Update docker/docker-compose.yaml

Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>

.github/workflows/web-e2e.yml

@@ -63,6 +63,7 @@ jobs:
           E2E_ADMIN_PASSWORD: E2eAdmin12345
           E2E_FORCE_WEB_BUILD: "1"
           E2E_INIT_PASSWORD: E2eInit12345
+          E2E_START_AGENT_BACKEND: "1"
         run: vp run e2e:full
 
       - name: Preserve Chromium E2E report and logs
@@ -138,21 +139,6 @@ jobs:
             exit 1
           fi
 
-          teardown_external_runtime() {
-            local run_status=$?
-            trap - EXIT
-            if ! vp run e2e:middleware:down; then
-              echo "::error title=E2E teardown failed::External runtime middleware did not shut down cleanly."
-              if [[ "$run_status" -eq 0 ]]; then
-                run_status=1
-              fi
-            fi
-            exit "$run_status"
-          }
-
-          trap teardown_external_runtime EXIT
-          vp run e2e:middleware:up
-          vp run e2e:post-merge:prepare
           vp run e2e:post-merge
 
       - name: Upload Cucumber report

api/Dockerfile

@@ -99,9 +99,9 @@ ENV VIRTUAL_ENV=/app/api/.venv
 COPY --from=packages --chown=dify:dify ${VIRTUAL_ENV} ${VIRTUAL_ENV}
 ENV PATH="${VIRTUAL_ENV}/bin:${PATH}"
 
-# Download nltk data
 RUN mkdir -p /usr/local/share/nltk_data \
-    && NLTK_DATA=/usr/local/share/nltk_data python -c "import nltk; nltk.download('punkt'); nltk.download('averaged_perceptron_tagger'); nltk.download('stopwords')" \
+    && NLTK_DATA=/usr/local/share/nltk_data python -m nltk.downloader punkt_tab averaged_perceptron_tagger_eng stopwords \
+    && NLTK_DATA=/usr/local/share/nltk_data python -c "import nltk; nltk.data.find('tokenizers/punkt_tab'); nltk.data.find('taggers/averaged_perceptron_tagger_eng'); nltk.data.find('corpora/stopwords')" \
     && chmod -R 755 /usr/local/share/nltk_data
 
 ENV TIKTOKEN_CACHE_DIR=/app/api/.tiktoken_cache

api/clients/agent_backend/__init__.py

@@ -5,8 +5,6 @@ API adapters: request building from Dify product concepts, a thin client wrapper
 event adaptation for future workflow integration, and deterministic fakes.
 """
 
-from dify_agent.protocol import RuntimeLayerSpec, extract_runtime_layer_specs
-
 from clients.agent_backend.client import AgentBackendRunClient, DifyAgentBackendRunClient
 from clients.agent_backend.errors import (
     AgentBackendError,
@@ -47,11 +45,6 @@ from clients.agent_backend.request_builder import (
     AgentBackendWorkflowNodeRunInput,
     redact_for_agent_backend_log,
 )
-from clients.agent_backend.session_cleanup import (
-    AgentBackendSessionCleanupPayload,
-    AgentBackendSessionCleanupResult,
-    cleanup_agent_backend_session,
-)
 
 __all__ = [
     "AGENT_SOUL_PROMPT_LAYER_ID",
@@ -80,8 +73,6 @@ __all__ = [
     "AgentBackendRunRequestBuilder",
     "AgentBackendRunStartedInternalEvent",
     "AgentBackendRunSucceededInternalEvent",
-    "AgentBackendSessionCleanupPayload",
-    "AgentBackendSessionCleanupResult",
     "AgentBackendStreamError",
     "AgentBackendStreamInternalEvent",
     "AgentBackendTransportError",
@@ -90,9 +81,6 @@ __all__ = [
     "DifyAgentBackendRunClient",
     "FakeAgentBackendRunClient",
     "FakeAgentBackendScenario",
-    "RuntimeLayerSpec",
-    "cleanup_agent_backend_session",
     "create_agent_backend_run_client",
-    "extract_runtime_layer_specs",
     "redact_for_agent_backend_log",
 ]

api/clients/agent_backend/request_builder.py

@@ -16,7 +16,6 @@ from collections.abc import Mapping
 from typing import ClassVar, Literal
 
 from agenton.compositor import CompositorSessionSnapshot
-from agenton.compositor.schemas import LayerSessionSnapshot
 from agenton.layers import ExitIntent
 from agenton_collections.layers.plain import PLAIN_PROMPT_LAYER_TYPE_ID, PromptLayerConfig
 from agenton_collections.layers.pydantic_ai import PYDANTIC_AI_HISTORY_LAYER_TYPE_ID
@@ -37,6 +36,7 @@ from dify_agent.layers.execution_context import (
 )
 from dify_agent.layers.knowledge import DIFY_KNOWLEDGE_BASE_LAYER_TYPE_ID, DifyKnowledgeBaseLayerConfig
 from dify_agent.layers.output import DIFY_OUTPUT_LAYER_TYPE_ID, DifyOutputLayerConfig
+from dify_agent.layers.runtime import DIFY_RUNTIME_LAYER_TYPE_ID, DifyRuntimeLayerConfig
 from dify_agent.layers.shell import DIFY_SHELL_LAYER_TYPE_ID, DifyShellLayerConfig
 from dify_agent.protocol import (
     DIFY_AGENT_HISTORY_LAYER_ID,
@@ -47,7 +47,6 @@ from dify_agent.protocol import (
     LayerExitSignals,
     RunComposition,
     RunLayerSpec,
-    RuntimeLayerSpec,
 )
 from pydantic import BaseModel, ConfigDict, Field, JsonValue, field_validator
 
@@ -56,6 +55,7 @@ WORKFLOW_NODE_JOB_PROMPT_LAYER_ID = "workflow_node_job_prompt"
 WORKFLOW_USER_PROMPT_LAYER_ID = "workflow_user_prompt"
 AGENT_APP_USER_PROMPT_LAYER_ID = "agent_app_user_prompt"
 DIFY_EXECUTION_CONTEXT_LAYER_ID = "execution_context"
+DIFY_RUNTIME_LAYER_ID = "runtime"
 DIFY_CONFIG_LAYER_ID = "config"
 DIFY_DRIVE_LAYER_ID = "drive"
 DIFY_PLUGIN_TOOLS_LAYER_ID = "tools"
@@ -66,25 +66,11 @@ DIFY_SHELL_LAYER_ID = "shell"
 type AgentConfigVersionKind = Literal["snapshot", "draft", "build_draft"]
 
 
-def _filter_snapshot_to_specs(
-    snapshot: CompositorSessionSnapshot,
-    specs: list[RuntimeLayerSpec],
-) -> CompositorSessionSnapshot:
-    """Keep only snapshot layers whose names appear in the cleanup spec list.
-
-    The agenton compositor rejects a snapshot whose layer-name sequence does
-    not match the active composition exactly. Cleanup-replay drops plugin
-    layers, so we must drop the matching snapshot entries here.
-    """
-    kept_names = {spec.name for spec in specs}
-    filtered_layers: list[LayerSessionSnapshot] = [layer for layer in snapshot.layers if layer.name in kept_names]
-    if len(filtered_layers) == len(snapshot.layers):
-        return snapshot
-    return CompositorSessionSnapshot(schema_version=snapshot.schema_version, layers=filtered_layers)
-
-
 def _shell_layer_deps() -> dict[str, str]:
-    return {"execution_context": DIFY_EXECUTION_CONTEXT_LAYER_ID}
+    return {
+        "execution_context": DIFY_EXECUTION_CONTEXT_LAYER_ID,
+        "runtime": DIFY_RUNTIME_LAYER_ID,
+    }
 
 
 def _drive_layer_deps() -> dict[str, str]:
@@ -214,6 +200,7 @@ class AgentBackendWorkflowNodeRunInput(BaseModel):
 
     model: AgentBackendModelConfig
     execution_context: DifyExecutionContextLayerConfig
+    backend_binding_ref: str = Field(min_length=1)
     workflow_node_job_prompt: str
     user_prompt: str
     agent_soul_prompt: str | None = None
@@ -231,8 +218,8 @@ class AgentBackendWorkflowNodeRunInput(BaseModel):
     # the Agent Soul configures human involvement; a deferred call ends the run and
     # the workflow pauses via the existing HITL form mechanism (ENG-635).
     ask_human_config: DifyAskHumanLayerConfig | None = None
-    # Inject the sandboxed shell layer (dify.shell). Requires the agent backend
-    # to be wired with a shellctl entrypoint; see configs AGENT_SHELL_ENABLED.
+    # Inject the sandboxed shell graph. Requires a deployment-selected runtime
+    # backend plus the product-resolved persistent Binding.
     include_shell: bool = False
     shell_config: DifyShellLayerConfig | None = None
     session_snapshot: CompositorSessionSnapshot | None = None
@@ -240,7 +227,6 @@ class AgentBackendWorkflowNodeRunInput(BaseModel):
     # (ENG-638). Keyed by the original deferred tool_call_id.
     deferred_tool_results: DeferredToolResultsPayload | None = None
     include_history: bool = True
-    suspend_on_exit: bool = True
     metadata: dict[str, JsonValue] = Field(default_factory=dict)
 
     model_config: ClassVar[ConfigDict] = ConfigDict(extra="forbid", arbitrary_types_allowed=True)
@@ -264,6 +250,7 @@ class AgentBackendAgentAppRunInput(BaseModel):
 
     model: AgentBackendModelConfig
     execution_context: DifyExecutionContextLayerConfig
+    backend_binding_ref: str = Field(min_length=1)
     user_prompt: str
     agent_soul_prompt: str | None = None
     agent_config_version_kind: AgentConfigVersionKind = "snapshot"
@@ -279,8 +266,8 @@ class AgentBackendAgentAppRunInput(BaseModel):
     # Human-in-the-loop ask_human deferred tool (dify.ask_human). Present only when
     # the Agent Soul configures human involvement (ENG-635).
     ask_human_config: DifyAskHumanLayerConfig | None = None
-    # Inject the sandboxed shell layer (dify.shell). Requires the agent backend
-    # to be wired with a shellctl entrypoint; see configs AGENT_SHELL_ENABLED.
+    # Inject the sandboxed shell graph. Requires a deployment-selected runtime
+    # backend plus the product-resolved persistent Binding.
     include_shell: bool = False
     shell_config: DifyShellLayerConfig | None = None
     session_snapshot: CompositorSessionSnapshot | None = None
@@ -288,7 +275,6 @@ class AgentBackendAgentAppRunInput(BaseModel):
     # (ENG-638). Keyed by the original deferred tool_call_id.
     deferred_tool_results: DeferredToolResultsPayload | None = None
     include_history: bool = True
-    suspend_on_exit: bool = True
     metadata: dict[str, JsonValue] = Field(default_factory=dict)
 
     model_config: ClassVar[ConfigDict] = ConfigDict(extra="forbid", arbitrary_types_allowed=True)
@@ -350,6 +336,14 @@ class AgentBackendRunRequestBuilder:
             run_input.include_shell or run_input.config_layer_config is not None or run_input.drive_config is not None
         )
         if include_shell:
+            layers.append(
+                RunLayerSpec(
+                    name=DIFY_RUNTIME_LAYER_ID,
+                    type=DIFY_RUNTIME_LAYER_TYPE_ID,
+                    metadata=run_input.metadata,
+                    config=DifyRuntimeLayerConfig(backend_binding_ref=run_input.backend_binding_ref),
+                )
+            )
             # Sandboxed bash workspace (dify.shell). It enters before config/drive
             # so eager pulls materialize content in the same filesystem used by
             # model commands.
@@ -481,53 +475,7 @@ class AgentBackendRunRequestBuilder:
             metadata=run_input.metadata,
             session_snapshot=run_input.session_snapshot,
             deferred_tool_results=run_input.deferred_tool_results,
-            on_exit=LayerExitSignals(
-                default=ExitIntent.SUSPEND if run_input.suspend_on_exit else ExitIntent.DELETE,
-            ),
-        )
-
-    def build_cleanup_request(
-        self,
-        *,
-        session_snapshot: CompositorSessionSnapshot,
-        runtime_layer_specs: list[RuntimeLayerSpec],
-        idempotency_key: str | None = None,
-        metadata: dict[str, JsonValue] | None = None,
-    ) -> CreateRunRequest:
-        """Build a lifecycle-only cleanup request that replays the prior layers.
-
-        The agenton compositor enforces that the session snapshot's layer names
-        match the active composition in order, so cleanup must replay the same
-        non-plugin layer graph that produced the snapshot. Plugin layers
-        (``dify.plugin.llm``, ``dify.plugin.tools``) are excluded from both the
-        composition and the snapshot before submission because their configs
-        may carry credentials or runtime-only declarations that are not
-        persisted between runs.
-        """
-        if not runtime_layer_specs:
-            raise ValueError(
-                "build_cleanup_request requires runtime_layer_specs; an empty "
-                "composition would fail the agent backend's snapshot validation."
-            )
-        request_metadata = dict(metadata or {})
-        request_metadata["agent_backend_lifecycle"] = "session_cleanup"
-        layers = [
-            RunLayerSpec(
-                name=spec.name,
-                type=spec.type,
-                deps=dict(spec.deps),
-                metadata=dict(spec.metadata),
-                config=spec.config,
-            )
-            for spec in runtime_layer_specs
-        ]
-        filtered_snapshot = _filter_snapshot_to_specs(session_snapshot, runtime_layer_specs)
-        return CreateRunRequest(
-            composition=RunComposition(layers=layers),
-            idempotency_key=idempotency_key,
-            metadata=request_metadata,
-            session_snapshot=filtered_snapshot,
-            on_exit=LayerExitSignals(default=ExitIntent.DELETE),
+            on_exit=LayerExitSignals(default=ExitIntent.SUSPEND),
         )
 
     def build_for_workflow_node(self, run_input: AgentBackendWorkflowNodeRunInput) -> CreateRunRequest:
@@ -580,6 +528,14 @@ class AgentBackendRunRequestBuilder:
             run_input.include_shell or run_input.config_layer_config is not None or run_input.drive_config is not None
         )
         if include_shell:
+            layers.append(
+                RunLayerSpec(
+                    name=DIFY_RUNTIME_LAYER_ID,
+                    type=DIFY_RUNTIME_LAYER_TYPE_ID,
+                    metadata=run_input.metadata,
+                    config=DifyRuntimeLayerConfig(backend_binding_ref=run_input.backend_binding_ref),
+                )
+            )
             # Sandboxed bash workspace (dify.shell). It enters before drive so
             # drive can materialize mentioned targets with `dify-agent drive pull`
             # in the same shell-visible filesystem used by model commands.
@@ -713,9 +669,7 @@ class AgentBackendRunRequestBuilder:
             metadata=run_input.metadata,
             session_snapshot=run_input.session_snapshot,
             deferred_tool_results=run_input.deferred_tool_results,
-            on_exit=LayerExitSignals(
-                default=ExitIntent.SUSPEND if run_input.suspend_on_exit else ExitIntent.DELETE,
-            ),
+            on_exit=LayerExitSignals(default=ExitIntent.SUSPEND),
         )
 
 

api/clients/agent_backend/session_cleanup.py

@@ -1,100 +0,0 @@
-"""Shared API-side helper for Agent backend lifecycle-only session cleanup.
-
-Product code owns local row retirement and background-task dispatch. This module
-only adapts persisted cleanup inputs into the public ``dify-agent`` run
-protocol, performs the synchronous ``create_run + wait_run`` loop used by Celery
-workers, and reports whether the backend cleanup succeeded, was skipped, or
-failed.
-"""
-
-from __future__ import annotations
-
-from dataclasses import dataclass
-from typing import ClassVar, Literal
-
-from agenton.compositor import CompositorSessionSnapshot
-from dify_agent.protocol import RuntimeLayerSpec
-from pydantic import BaseModel, ConfigDict, Field, JsonValue
-
-from clients.agent_backend.client import AgentBackendRunClient
-from clients.agent_backend.errors import AgentBackendError
-from clients.agent_backend.request_builder import AgentBackendRunRequestBuilder
-
-
-class AgentBackendSessionCleanupPayload(BaseModel):
-    """Serialized cleanup inputs preserved across API and Celery boundaries."""
-
-    session_snapshot: CompositorSessionSnapshot | None = None
-    runtime_layer_specs: list[RuntimeLayerSpec] = Field(default_factory=list)
-    idempotency_key: str | None = None
-    metadata: dict[str, JsonValue] = Field(default_factory=dict)
-    timeout_seconds: float = 30.0
-
-    model_config: ClassVar[ConfigDict] = ConfigDict(extra="forbid")
-
-
-@dataclass(frozen=True, slots=True)
-class AgentBackendSessionCleanupResult:
-    """Terminal outcome of one backend cleanup attempt."""
-
-    status: Literal["succeeded", "skipped", "failed"]
-    reason: str | None = None
-    cleanup_run_id: str | None = None
-
-    @classmethod
-    def succeeded(cls, cleanup_run_id: str) -> AgentBackendSessionCleanupResult:
-        return cls(status="succeeded", cleanup_run_id=cleanup_run_id)
-
-    @classmethod
-    def skipped(cls, reason: str) -> AgentBackendSessionCleanupResult:
-        return cls(status="skipped", reason=reason)
-
-    @classmethod
-    def failed(cls, reason: str, cleanup_run_id: str | None = None) -> AgentBackendSessionCleanupResult:
-        return cls(status="failed", reason=reason, cleanup_run_id=cleanup_run_id)
-
-
-def cleanup_agent_backend_session(
-    *,
-    payload: AgentBackendSessionCleanupPayload,
-    client: AgentBackendRunClient | None,
-    request_builder: AgentBackendRunRequestBuilder | None = None,
-) -> AgentBackendSessionCleanupResult:
-    """Run lifecycle-only cleanup against the Agent backend and report status."""
-    if client is None:
-        return AgentBackendSessionCleanupResult.skipped("no_agent_backend_client")
-    if payload.session_snapshot is None:
-        return AgentBackendSessionCleanupResult.skipped("missing_session_snapshot")
-    if not payload.runtime_layer_specs:
-        return AgentBackendSessionCleanupResult.skipped("missing_runtime_layer_specs")
-
-    builder = request_builder or AgentBackendRunRequestBuilder()
-    request = builder.build_cleanup_request(
-        session_snapshot=payload.session_snapshot,
-        runtime_layer_specs=payload.runtime_layer_specs,
-        idempotency_key=payload.idempotency_key,
-        metadata=payload.metadata,
-    )
-
-    try:
-        response = client.create_run(request)
-    except AgentBackendError as exc:
-        return AgentBackendSessionCleanupResult.failed(str(exc))
-
-    try:
-        status_response = client.wait_run(response.run_id, timeout_seconds=payload.timeout_seconds)
-    except AgentBackendError as exc:
-        return AgentBackendSessionCleanupResult.failed(str(exc), cleanup_run_id=response.run_id)
-
-    if status_response.status != "succeeded":
-        reason = status_response.error or f"cleanup run ended with status {status_response.status}"
-        return AgentBackendSessionCleanupResult.failed(reason, cleanup_run_id=response.run_id)
-
-    return AgentBackendSessionCleanupResult.succeeded(response.run_id)
-
-
-__all__ = [
-    "AgentBackendSessionCleanupPayload",
-    "AgentBackendSessionCleanupResult",
-    "cleanup_agent_backend_session",
-]

api/configs/extra/agent_backend_config.py

@@ -44,9 +44,8 @@ class AgentBackendConfig(BaseSettings):
 
     AGENT_SHELL_ENABLED: bool = Field(
         description=(
-            "Inject the dify.shell layer (sandboxed bash workspace) into Agent runs. "
-            "Requires the agent backend to be wired with a shellctl entrypoint before "
-            "shell-using Agent runs are executed."
+            "Inject the Home, Workspace, Sandbox, and Shell runtime layers into Agent runs. "
+            "Requires Dify Agent to have a deployment-selected runtime backend."
         ),
         default=True,
     )

api/controllers/common/session.py

@@ -52,7 +52,7 @@ def with_session[T, **P, R](
                         session.commit()
                         return result
                     except Exception:
-                        session.rollback()  # noqa: no-new-controller-sqlalchemy decorator owns transaction rollback
+                        session.rollback()  # guard-ignore: no-new-controller-sqlalchemy -- decorator owns rollback
                         raise
 
             with session_factory.create_session() as session:

api/controllers/console/agent/roster.py

@@ -62,7 +62,7 @@ from libs.datetime_utils import parse_time_range
 from libs.helper import dump_response
 from libs.login import login_required
 from models import Account
-from models.agent import Agent, AgentConfigDraftType, AgentStatus
+from models.agent import Agent, AgentStatus
 from models.agent_config_entities import AgentSoulConfig
 from models.enums import ApiTokenType
 from models.model import ApiToken, App, IconType
@@ -265,13 +265,6 @@ class AgentDebugConversationRefreshResponse(BaseModel):
     debug_conversation_message_count: int = 0
 
 
-class AgentDebugConversationRefreshPayload(BaseModel):
-    draft_type: AgentConfigDraftType = Field(
-        default=AgentConfigDraftType.DEBUG_BUILD,
-        description="Agent draft surface whose conversation should be refreshed",
-    )
-
-
 class AgentPublishPayload(BaseModel):
     version_note: str | None = Field(default=None, description="Optional note for this published Agent version")
 
@@ -315,7 +308,6 @@ register_schema_models(
     AgentAppCopyPayload,
     AgentPublishPayload,
     AgentBuildDraftCheckoutPayload,
-    AgentDebugConversationRefreshPayload,
     ComposerSavePayload,
     AgentApiStatusPayload,
     AgentInviteOptionsQuery,
@@ -395,11 +387,10 @@ def _serialize_agent_app_detail(
     payload["backing_app_id"] = roster_service.runtime_backing_app_id(agent)
     payload["hidden_app_backed"] = bool(agent.backing_app_id and agent.backing_app_id != agent.app_id)
     payload["id"] = agent.id
-    debug_conversation_id = roster_service.get_or_create_agent_app_debug_conversation_id(
+    debug_conversation_id = roster_service.get_or_create_build_conversation(
         tenant_id=app_model.tenant_id,
         agent_id=agent.id,
         account_id=current_user.id,
-        draft_type=AgentConfigDraftType.DEBUG_BUILD,
         commit=False,
     )
     message_count = roster_service.count_agent_app_debug_conversation_messages(
@@ -439,11 +430,10 @@ def _serialize_agent_app_pagination(session: Session, app_pagination, *, tenant_
         tenant_id=tenant_id,
         agent_ids=[agent.id for agent in agents_by_app_id.values()],
     )
-    debug_conversation_ids_by_agent_id = roster_service.load_or_create_agent_app_debug_conversation_ids_by_agent_id(
+    debug_conversation_ids_by_agent_id = roster_service.load_or_create_build_conversation_ids_by_agent_id(
         tenant_id=tenant_id,
         agents=list(agents_by_app_id.values()),
         account_id=current_user.id,
-        draft_type=AgentConfigDraftType.DEBUG_BUILD,
     )
     payload = AgentAppPagination.model_validate(
         app_pagination,
@@ -678,16 +668,6 @@ class AgentAppApi(Resource):
 
 @console_ns.route("/agent/<uuid:agent_id>/debug-conversation/refresh")
 class AgentDebugConversationRefreshApi(Resource):
-    @console_ns.expect(console_ns.models[AgentDebugConversationRefreshPayload.__name__])
-    @console_ns.doc(
-        params={
-            "payload": {
-                "in": "body",
-                "required": False,
-                "schema": {"$ref": f"#/components/schemas/{AgentDebugConversationRefreshPayload.__name__}"},
-            }
-        }
-    )
     @console_ns.response(
         200,
         "Agent debug conversation refreshed",
@@ -702,12 +682,10 @@ class AgentDebugConversationRefreshApi(Resource):
     @with_current_tenant_id
     @with_session
     def post(self, session: Session, tenant_id: str, current_user: Account, agent_id: UUID):
-        args = AgentDebugConversationRefreshPayload.model_validate(request.get_json(silent=True) or {})
-        debug_conversation_id = _agent_roster_service(session).refresh_agent_app_debug_conversation_id(
+        debug_conversation_id = _agent_roster_service(session).reset_build_conversation(
             tenant_id=tenant_id,
             agent_id=str(agent_id),
             account_id=current_user.id,
-            draft_type=args.draft_type,
         )
         return AgentDebugConversationRefreshResponse(
             debug_conversation_id=debug_conversation_id,
@@ -751,7 +729,7 @@ class AgentBuildDraftCheckoutApi(Resource):
     @rbac_permission_required(RBACResourceScope.WORKSPACE, RBACPermission.AGENT_MANAGE, resource_required=False)
     @with_current_user
     @with_current_tenant_id
-    @with_session
+    @with_session(write=False)
     def post(self, session: Session, tenant_id: str, current_user: Account, agent_id: UUID):
         args = AgentBuildDraftCheckoutPayload.model_validate(console_ns.payload or {})
         return AgentComposerService.checkout_agent_app_build_draft(
@@ -808,7 +786,7 @@ class AgentBuildDraftApi(Resource):
     @edit_permission_required
     @with_current_user
     @with_current_tenant_id
-    @with_session
+    @with_session(write=False)
     def delete(self, session: Session, tenant_id: str, current_user: Account, agent_id: UUID):
         return AgentComposerService.discard_agent_app_build_draft(
             session=session,
@@ -828,7 +806,7 @@ class AgentBuildDraftApplyApi(Resource):
     @rbac_permission_required(RBACResourceScope.WORKSPACE, RBACPermission.AGENT_MANAGE, resource_required=False)
     @with_current_user
     @with_current_tenant_id
-    @with_session
+    @with_session(write=False)
     def post(self, session: Session, tenant_id: str, current_user: Account, agent_id: UUID):
         return AgentComposerService.apply_agent_app_build_draft(
             session=session,

api/controllers/console/app/agent_app_sandbox.py

@@ -1,7 +1,7 @@
 """Console routes for Agent App and workflow Agent sandbox file access.
 
-The API keeps product-facing locators (conversation or workflow node identity)
-on this public boundary and proxies list/read/upload to the agent backend's new
+The API accepts product-facing Conversation, Build Draft, or Workflow Node
+Execution locators and proxies list/read/upload to the agent backend's
 ``/sandbox`` contract.
 """
 
@@ -26,10 +26,16 @@ from controllers.common.session import with_session
 from controllers.console import console_ns
 from controllers.console.agent.app_helpers import resolve_agent_runtime_app_model
 from controllers.console.app.wraps import get_app_model
-from controllers.console.wraps import account_initialization_required, setup_required, with_current_tenant_id
+from controllers.console.wraps import (
+    account_initialization_required,
+    setup_required,
+    with_current_tenant_id,
+    with_current_user,
+)
 from extensions.ext_database import db
 from fields.base import ResponseModel
 from libs.login import login_required
+from models import Account
 from models.model import App, AppMode
 from services.agent_app_sandbox_service import (
     AgentAppSandboxService,
@@ -37,52 +43,43 @@ from services.agent_app_sandbox_service import (
     WorkflowAgentSandboxService,
 )
 
-_NODE_EXECUTION_ID_DESCRIPTION = (
-    "Optional workflow node execution ID. When omitted, the latest active session for the node is used."
-)
-
 
 class AgentSandboxListQuery(BaseModel):
-    conversation_id: str = Field(min_length=1, description="Agent App conversation ID")
+    caller_type: Literal["conversation", "build_draft"]
+    caller_id: str = Field(min_length=1, description="Agent App caller ID")
     path: str = Field(default=".", description="Directory path relative to the sandbox workspace")
 
 
 class AgentSandboxInfoQuery(BaseModel):
-    conversation_id: str = Field(min_length=1, description="Agent App conversation ID")
+    caller_type: Literal["conversation", "build_draft"]
+    caller_id: str = Field(min_length=1, description="Agent App caller ID")
 
 
 class AgentSandboxFileQuery(BaseModel):
-    conversation_id: str = Field(min_length=1, description="Agent App conversation ID")
+    caller_type: Literal["conversation", "build_draft"]
+    caller_id: str = Field(min_length=1, description="Agent App caller ID")
     path: str = Field(min_length=1, description="File path relative to the sandbox workspace")
 
 
 class AgentSandboxUploadPayload(BaseModel):
-    conversation_id: str = Field(min_length=1, description="Agent App conversation ID")
+    caller_type: Literal["conversation", "build_draft"]
+    caller_id: str = Field(min_length=1, description="Agent App caller ID")
     path: str = Field(min_length=1, description="File path relative to the sandbox workspace")
 
 
 class WorkflowAgentSandboxListQuery(BaseModel):
+    node_execution_id: str = Field(min_length=1, description="Workflow node execution ID")
     path: str = Field(default=".", description="Directory path relative to the sandbox workspace")
-    node_execution_id: str | None = Field(
-        default=None,
-        description=_NODE_EXECUTION_ID_DESCRIPTION,
-    )
 
 
 class WorkflowAgentSandboxFileQuery(BaseModel):
+    node_execution_id: str = Field(min_length=1, description="Workflow node execution ID")
     path: str = Field(min_length=1, description="File path relative to the sandbox workspace")
-    node_execution_id: str | None = Field(
-        default=None,
-        description=_NODE_EXECUTION_ID_DESCRIPTION,
-    )
 
 
 class WorkflowAgentSandboxUploadPayload(BaseModel):
+    node_execution_id: str = Field(min_length=1, description="Workflow node execution ID")
     path: str = Field(min_length=1, description="File path relative to the sandbox workspace")
-    node_execution_id: str | None = Field(
-        default=None,
-        description=_NODE_EXECUTION_ID_DESCRIPTION,
-    )
 
 
 class SandboxFileEntryResponse(ResponseModel):
@@ -99,7 +96,6 @@ class SandboxListResponse(ResponseModel):
 
 
 class SandboxInfoResponse(ResponseModel):
-    session_id: str
     workspace_cwd: str
 
 
@@ -155,15 +151,19 @@ class AgentAppSandboxInfoResource(Resource):
     @login_required
     @account_initialization_required
     @with_current_tenant_id
+    @with_current_user
     @with_session(write=False)
-    def get(self, session: Session, tenant_id: str, agent_id: UUID):
+    def get(self, session: Session, current_user: Account, tenant_id: str, agent_id: UUID):
         app_model = resolve_agent_runtime_app_model(session=session, tenant_id=tenant_id, agent_id=agent_id)
         query = query_params_from_request(AgentSandboxInfoQuery)
         try:
             result = AgentAppSandboxService().get_info(
                 tenant_id=tenant_id,
                 app_id=app_model.id,
-                conversation_id=query.conversation_id,
+                agent_id=str(agent_id),
+                caller_type=query.caller_type,
+                caller_id=query.caller_id,
+                account_id=current_user.id,
             )
         except Exception as exc:
             return _handle(exc)
@@ -180,15 +180,19 @@ class AgentAppSandboxListResource(Resource):
     @login_required
     @account_initialization_required
     @with_current_tenant_id
+    @with_current_user
     @with_session(write=False)
-    def get(self, session: Session, tenant_id: str, agent_id: UUID):
+    def get(self, session: Session, current_user: Account, tenant_id: str, agent_id: UUID):
         app_model = resolve_agent_runtime_app_model(session=session, tenant_id=tenant_id, agent_id=agent_id)
         query = query_params_from_request(AgentSandboxListQuery)
         try:
             result = AgentAppSandboxService().list_files(
                 tenant_id=tenant_id,
                 app_id=app_model.id,
-                conversation_id=query.conversation_id,
+                agent_id=str(agent_id),
+                caller_type=query.caller_type,
+                caller_id=query.caller_id,
+                account_id=current_user.id,
                 path=query.path,
             )
         except Exception as exc:
@@ -206,15 +210,19 @@ class AgentAppSandboxReadResource(Resource):
     @login_required
     @account_initialization_required
     @with_current_tenant_id
+    @with_current_user
     @with_session(write=False)
-    def get(self, session: Session, tenant_id: str, agent_id: UUID):
+    def get(self, session: Session, current_user: Account, tenant_id: str, agent_id: UUID):
         app_model = resolve_agent_runtime_app_model(session=session, tenant_id=tenant_id, agent_id=agent_id)
         query = query_params_from_request(AgentSandboxFileQuery)
         try:
             result = AgentAppSandboxService().read_file(
                 tenant_id=tenant_id,
                 app_id=app_model.id,
-                conversation_id=query.conversation_id,
+                agent_id=str(agent_id),
+                caller_type=query.caller_type,
+                caller_id=query.caller_id,
+                account_id=current_user.id,
                 path=query.path,
             )
         except Exception as exc:
@@ -232,15 +240,19 @@ class AgentAppSandboxUploadResource(Resource):
     @login_required
     @account_initialization_required
     @with_current_tenant_id
+    @with_current_user
     @with_session(write=False)
-    def post(self, session: Session, tenant_id: str, agent_id: UUID):
+    def post(self, session: Session, current_user: Account, tenant_id: str, agent_id: UUID):
         app_model = resolve_agent_runtime_app_model(session=session, tenant_id=tenant_id, agent_id=agent_id)
         payload = AgentSandboxUploadPayload.model_validate(request.get_json(silent=True) or {})
         try:
             result = AgentAppSandboxService().upload_file(
                 tenant_id=tenant_id,
                 app_id=app_model.id,
-                conversation_id=payload.conversation_id,
+                agent_id=str(agent_id),
+                caller_type=payload.caller_type,
+                caller_id=payload.caller_id,
+                account_id=current_user.id,
                 path=payload.path,
             )
         except Exception as exc:

api/controllers/console/app/completion.py

@@ -36,7 +36,7 @@ from controllers.console.wraps import (
     with_current_user_id,
 )
 from controllers.web.error import InvokeRateLimitError as InvokeRateLimitHttpError
-from core.app.entities.app_invoke_entities import AGENT_RUNTIME_EXIT_INTENT_ARG, InvokeFrom
+from core.app.entities.app_invoke_entities import InvokeFrom
 from core.app.features.rate_limiting.rate_limit import RateLimitGenerator
 from core.errors.error import (
     ModelCurrentlyNotSupportError,
@@ -353,12 +353,7 @@ def _resolve_current_user_agent_debug_conversation_id(
     draft_type: AgentConfigDraftType,
     start_new: bool = False,
 ) -> str:
-    """Resolve or rotate the current editor's conversation within one draft surface.
-
-    ``start_new`` rotates the scoped mapping through ``AgentRosterService`` so
-    the old runtime session is retired before the new conversation is used.
-    Continuations and Build chat keep resolving the existing mapping.
-    """
+    """Resolve the current editor's Build or Preview conversation."""
 
     roster_service = AgentRosterService(session)
     resolved_agent_id = agent_id
@@ -368,17 +363,26 @@ def _resolve_current_user_agent_debug_conversation_id(
             raise AgentNotFoundError()
         resolved_agent_id = agent.id
 
-    resolve_conversation = (
-        roster_service.refresh_agent_app_debug_conversation_id
-        if start_new
-        else roster_service.get_or_create_agent_app_debug_conversation_id
-    )
-    return resolve_conversation(
+    if draft_type == AgentConfigDraftType.DEBUG_BUILD:
+        return roster_service.get_or_create_build_conversation(
+            tenant_id=current_tenant_id,
+            agent_id=resolved_agent_id,
+            account_id=current_user.id,
+        )
+    if start_new:
+        return roster_service.rotate_preview_conversation(
+            tenant_id=current_tenant_id,
+            agent_id=resolved_agent_id,
+            account_id=current_user.id,
+        )
+    conversation_id = roster_service.get_current_preview_conversation(
         tenant_id=current_tenant_id,
         agent_id=resolved_agent_id,
         account_id=current_user.id,
-        draft_type=draft_type,
     )
+    if conversation_id is None:
+        raise NotFound("Conversation Not Exists.")
+    return conversation_id
 
 
 def _create_chat_message(
@@ -450,7 +454,6 @@ def _create_build_chat_finalization_message(
         "draft_type": "debug_build",
         "conversation_id": debug_conversation_id,
         "auto_generate_name": False,
-        AGENT_RUNTIME_EXIT_INTENT_ARG: "delete",
     }
     external_trace_id = get_external_trace_id(request)
     if external_trace_id:

api/controllers/console/app/workflow.py

@@ -76,11 +76,13 @@ from models import Account, App
 from models.model import AppMode
 from models.workflow import Workflow
 from repositories.workflow_collaboration_repository import WORKFLOW_ONLINE_USERS_PREFIX
+from services.agent.retirement_service import WorkflowAgentRetirementService
 from services.app_generate_service import AppGenerateService
 from services.errors.app import IsDraftWorkflowError, WorkflowHashNotEqualError, WorkflowNotFoundError
 from services.errors.llm import InvokeRateLimitError
 from services.workflow_ref_service import WorkflowRefService
 from services.workflow_service import DraftWorkflowDeletionError, WorkflowInUseError, WorkflowService
+from tasks.collect_agent_resources_task import enqueue_agent_resource_collection
 
 logger = logging.getLogger(__name__)
 
@@ -317,7 +319,7 @@ class _WorkflowResponseSource:
         self._session = session
 
     def __getattr__(self, name: str) -> object:
-        return getattr(self._workflow, name)  # noqa: no-new-getattr response adapter delegates model fields
+        return getattr(self._workflow, name)  # guard-ignore: no-new-getattr -- delegates model fields
 
     @property
     def created_by_account(self) -> Account | None:
@@ -1245,7 +1247,7 @@ class PublishedWorkflowApi(Resource):
 
         workflow_service = WorkflowService()
         with sessionmaker(db.engine).begin() as session:
-            workflow = workflow_service.publish_workflow(
+            workflow, retirement_candidates = workflow_service.publish_workflow(
                 session=session,
                 app_model=app_model,
                 account=current_user,
@@ -1262,6 +1264,16 @@ class PublishedWorkflowApi(Resource):
 
             workflow_created_at = TimestampField().format(workflow.created_at)
 
+        binding_ids, home_snapshot_ids = WorkflowAgentRetirementService.retire_unowned(
+            tenant_id=app_model.tenant_id,
+            agent_ids=retirement_candidates,
+            account_id=current_user.id,
+        )
+        enqueue_agent_resource_collection(
+            tenant_id=app_model.tenant_id,
+            binding_ids=binding_ids,
+            home_snapshot_ids=home_snapshot_ids,
+        )
         return {
             "result": "success",
             "created_at": workflow_created_at,

api/controllers/console/datasets/datasets.py

@@ -218,7 +218,7 @@ class _DatasetQueryResponseSource:
         return self.query.get_queries(session=self.session)
 
     def __getattr__(self, name: str) -> Any:
-        return getattr(self.query, name)  # noqa: no-new-getattr response adapter delegates model fields
+        return getattr(self.query, name)  # guard-ignore: no-new-getattr -- delegates model fields
 
 
 class DatasetQueryListResponse(ResponseModel):
@@ -257,7 +257,7 @@ class _RelatedAppResponseSource:
         return self.app.mode_compatible_with_agent_with_session(session=self.session)
 
     def __getattr__(self, name: str) -> Any:
-        return getattr(self.app, name)  # noqa: no-new-getattr response adapter delegates model fields
+        return getattr(self.app, name)  # guard-ignore: no-new-getattr -- delegates model fields
 
 
 class RelatedAppListResponse(ResponseModel):

api/controllers/console/datasets/external.py

@@ -106,7 +106,7 @@ class ExternalKnowledgeApiResponseSource:
         return self.external_knowledge_api.get_dataset_bindings(session=self.session)
 
     def __getattr__(self, name: str) -> Any:
-        return getattr(self.external_knowledge_api, name)  # noqa: no-new-getattr response adapter delegates model fields
+        return getattr(self.external_knowledge_api, name)  # guard-ignore: no-new-getattr -- delegates model fields
 
 
 def external_knowledge_api_response(

api/controllers/console/explore/trial.py

@@ -404,7 +404,7 @@ class TrialWorkflowResponseSource:
         return self.workflow.get_tool_published(session=self.session)
 
     def __getattr__(self, name: str) -> Any:
-        return getattr(self.workflow, name)  # noqa: no-new-getattr response adapter delegates model fields
+        return getattr(self.workflow, name)  # guard-ignore: no-new-getattr -- delegates model fields
 
 
 register_schema_models(

api/controllers/console/snippets/snippet_workflow.py

@@ -56,10 +56,12 @@ from libs.helper import TimestampField
 from libs.login import current_account_with_tenant, login_required
 from models import Account
 from models.snippet import CustomizedSnippet
+from services.agent.retirement_service import WorkflowAgentRetirementService
 from services.agent.workflow_publish_service import WorkflowAgentPublishService
 from services.errors.app import IsDraftWorkflowError, WorkflowHashNotEqualError, WorkflowNotFoundError
 from services.snippet_generate_service import SnippetGenerateService
 from services.snippet_service import SnippetService
+from tasks.collect_agent_resources_task import enqueue_agent_resource_collection
 
 logger = logging.getLogger(__name__)
 
@@ -295,8 +297,9 @@ class SnippetPublishedWorkflowApi(Resource):
 
         with Session(db.engine) as session:
             snippet = session.merge(snippet)
+            tenant_id = snippet.tenant_id
             try:
-                workflow = snippet_service.publish_workflow(
+                workflow, retirement_candidates = snippet_service.publish_workflow(
                     session=session,
                     snippet=snippet,
                     account=current_user,
@@ -306,6 +309,16 @@ class SnippetPublishedWorkflowApi(Resource):
             except ValueError as e:
                 return {"message": str(e)}, 400
 
+        binding_ids, home_snapshot_ids = WorkflowAgentRetirementService.retire_unowned(
+            tenant_id=tenant_id,
+            agent_ids=retirement_candidates,
+            account_id=current_user.id,
+        )
+        enqueue_agent_resource_collection(
+            tenant_id=tenant_id,
+            binding_ids=binding_ids,
+            home_snapshot_ids=home_snapshot_ids,
+        )
         return {
             "result": "success",
             "created_at": workflow_created_at,

api/core/app/apps/advanced_chat/app_runner.py

@@ -16,6 +16,7 @@ from core.app.apps.workflow_app_runner import WorkflowBasedAppRunner
 from core.app.entities.app_invoke_entities import (
     AdvancedChatAppGenerateEntity,
     AppGenerateEntity,
+    DifyRunContext,
     InvokeFrom,
 )
 from core.app.entities.queue_entities import (
@@ -31,7 +32,7 @@ from core.moderation.base import ModerationError
 from core.moderation.input_moderation import InputModeration
 from core.repositories.factory import WorkflowExecutionRepository, WorkflowNodeExecutionRepository
 from core.workflow.node_factory import get_default_root_node_id
-from core.workflow.nodes.agent_v2.session_cleanup_layer import build_workflow_agent_session_cleanup_layer
+from core.workflow.nodes.agent_v2.workspace_retirement_layer import build_workflow_agent_workspace_retirement_layer
 from core.workflow.system_variables import (
     build_bootstrap_variables,
     build_system_variables,
@@ -267,7 +268,18 @@ class AdvancedChatAppRunner(WorkflowBasedAppRunner):
         )
 
         workflow_entry.graph_engine.layer(persistence_layer)
-        workflow_entry.graph_engine.layer(build_workflow_agent_session_cleanup_layer())
+        workflow_entry.graph_engine.layer(
+            build_workflow_agent_workspace_retirement_layer(
+                dify_run_context=DifyRunContext(
+                    tenant_id=self._workflow.tenant_id,
+                    app_id=self._workflow.app_id,
+                    user_id=self.application_generate_entity.user_id,
+                    user_from=user_from,
+                    invoke_from=invoke_from,
+                    trace_session_id=self.application_generate_entity.extras.get("trace_session_id"),
+                )
+            )
+        )
         conversation_variable_layer = ConversationVariablePersistenceLayer(
             ConversationVariableUpdater(session_factory.get_session_maker())
         )

api/core/app/apps/agent_app/app_generator.py

@@ -33,15 +33,13 @@ from core.app.apps.agent_app.app_runner import AgentAppRunner
 from core.app.apps.agent_app.errors import AgentAppGeneratorError, AgentAppNotPublishedError
 from core.app.apps.agent_app.generate_response_converter import AgentAppGenerateResponseConverter
 from core.app.apps.agent_app.runtime_request_builder import AgentAppRuntimeRequestBuilder
-from core.app.apps.agent_app.session_store import AgentAppRuntimeSessionStore
+from core.app.apps.agent_app.session_store import AgentAppWorkspaceStore
 from core.app.apps.base_app_queue_manager import AppQueueManager, PublishFrom
 from core.app.apps.exc import GenerateTaskStoppedError
 from core.app.apps.message_based_app_generator import MessageBasedAppGenerator
 from core.app.apps.message_based_app_queue_manager import MessageBasedAppQueueManager
 from core.app.entities.app_invoke_entities import (
-    AGENT_RUNTIME_EXIT_INTENT_ARG,
     AgentAppGenerateEntity,
-    AgentRuntimeExitIntent,
     DifyRunContext,
     InvokeFrom,
     UserFrom,
@@ -51,19 +49,24 @@ from core.db.session_factory import session_factory
 from core.ops.ops_trace_manager import TraceQueueManager
 from core.workflow.file_reference import build_file_reference, is_canonical_file_reference
 from extensions.ext_database import db
-from models import Account, App, AppModelConfig, EndUser, Message, MessageAnnotation
+from models import Account, App, AppModelConfig, Conversation, EndUser, Message, MessageAnnotation
 from models.agent import (
     APP_BACKED_AGENT_SOURCES,
     Agent,
     AgentConfigDraft,
     AgentConfigDraftType,
     AgentConfigSnapshot,
+    AgentConfigVersionKind,
     AgentScope,
     AgentSource,
     AgentStatus,
+    AgentWorkingResourceStatus,
+    AgentWorkspaceBinding,
+    AgentWorkspaceOwnerType,
 )
 from models.agent_config_entities import AgentSoulConfig
 from models.model import load_annotation_reply_config
+from services.agent.workspace_service import AgentWorkspaceService, WorkspaceOwnerScope
 from services.conversation_service import ConversationService
 
 logger = logging.getLogger(__name__)
@@ -150,19 +153,6 @@ class AgentAppGenerator(MessageBasedAppGenerator):
         inputs = args["inputs"]
         prompt_file_mappings = args.get("files") or []
 
-        # Resolve the bound roster Agent + its current Agent Soul snapshot.
-        agent, agent_config_id, agent_config_version_kind, agent_soul = self._resolve_agent(
-            app_model,
-            invoke_from=invoke_from,
-            draft_type=args.get("draft_type"),
-            user=user,
-            session=session,
-        )
-        runtime_session_snapshot_id = self._runtime_session_snapshot_id(
-            invoke_from=invoke_from,
-            snapshot_id=agent_config_id,
-        )
-
         conversation = None
         conversation_id = args.get("conversation_id")
         if conversation_id:
@@ -170,6 +160,21 @@ class AgentAppGenerator(MessageBasedAppGenerator):
                 app_model=app_model, conversation_id=conversation_id, user=user, session=session
             )
 
+        # New conversations use the current Agent generation. Existing
+        # conversations use the immutable generation named by their Binding.
+        agent, agent_config_id, agent_config_version_kind, agent_soul = self._resolve_agent(
+            app_model,
+            invoke_from=invoke_from,
+            draft_type=args.get("draft_type"),
+            user=user,
+            session=session,
+            conversation=conversation,
+        )
+        session_scope_config_version_id = self._session_scope_config_version_id(
+            invoke_from=invoke_from,
+            config_version_id=agent_config_id,
+        )
+
         # Build the EasyUI-shaped config from the Agent Soul so the chat pipeline
         # can persist usage; the answer itself comes from the agent backend.
         app_model_config = (
@@ -186,8 +191,6 @@ class AgentAppGenerator(MessageBasedAppGenerator):
         model_conf = ModelConfigConverter.convert(app_config)
 
         trace_manager = TraceQueueManager(app_model.id, user.id if isinstance(user, Account) else user.session_id)
-        agent_runtime_exit_intent = self._resolve_agent_runtime_exit_intent(args)
-
         application_generate_entity = AgentAppGenerateEntity(
             task_id=str(uuid.uuid4()),
             app_config=app_config,
@@ -215,8 +218,7 @@ class AgentAppGenerator(MessageBasedAppGenerator):
             agent_id=agent.id,
             agent_config_snapshot_id=agent_config_id,
             agent_config_version_kind=agent_config_version_kind,
-            agent_runtime_session_snapshot_id=runtime_session_snapshot_id,
-            agent_runtime_exit_intent=agent_runtime_exit_intent,
+            agent_session_scope_config_version_id=session_scope_config_version_id,
         )
 
         conversation, message = self._init_generate_records(
@@ -265,6 +267,7 @@ class AgentAppGenerator(MessageBasedAppGenerator):
         app_model: App,
         user: Account | EndUser,
         conversation_id: str,
+        form_id: str,
         invoke_from: InvokeFrom,
         session: Session,
     ) -> None:
@@ -279,14 +282,21 @@ class AgentAppGenerator(MessageBasedAppGenerator):
         conversation = ConversationService.get_conversation(
             app_model=app_model, conversation_id=conversation_id, user=user, session=session
         )
+        draft_type, draft_id = self._resolve_resume_draft(
+            app_model=app_model,
+            conversation=conversation,
+            user=user,
+            form_id=form_id,
+            session=session,
+        )
         agent, agent_config_id, agent_config_version_kind, agent_soul = self._resolve_agent(
             app_model,
             invoke_from=invoke_from,
-            draft_type=self._resume_draft_type(
-                app_model=app_model, conversation=conversation, user=user, session=session
-            ),
+            draft_type=draft_type,
+            draft_id=draft_id,
             user=user,
             session=session,
+            conversation=conversation,
         )
 
         app_model_config = (
@@ -384,30 +394,37 @@ class AgentAppGenerator(MessageBasedAppGenerator):
         )
 
     @staticmethod
-    def _resume_draft_type(
-        *, app_model: App, conversation: Any, user: Account | EndUser, session: Session
-    ) -> str | None:
+    def _resolve_resume_draft(
+        *,
+        app_model: App,
+        conversation: Any,
+        user: Account | EndUser,
+        form_id: str,
+        session: Session,
+    ) -> tuple[str | None, str | None]:
         if conversation.invoke_from != InvokeFrom.DEBUGGER:
-            return None
-        active_session = AgentAppRuntimeSessionStore().load_active_session_for_conversation(
-            tenant_id=app_model.tenant_id,
-            app_id=app_model.id,
-            conversation_id=conversation.id,
-        )
-        snapshot_id = active_session.scope.agent_config_snapshot_id if active_session is not None else None
-        if snapshot_id and isinstance(user, Account):
-            draft = session.scalar(
-                select(AgentConfigDraft).where(
-                    AgentConfigDraft.tenant_id == app_model.tenant_id,
-                    AgentConfigDraft.id == snapshot_id,
-                )
+            return None, None
+        if not isinstance(user, Account):
+            return AgentConfigDraftType.DRAFT.value, None
+
+        build_draft = session.scalar(
+            select(AgentConfigDraft)
+            .join(
+                AgentWorkspaceBinding,
+                AgentWorkspaceBinding.id == AgentConfigDraft.agent_workspace_binding_id,
             )
-            if draft is not None:
-                if draft.draft_type == AgentConfigDraftType.DEBUG_BUILD and draft.account_id == user.id:
-                    return AgentConfigDraftType.DEBUG_BUILD.value
-                if draft.draft_type == AgentConfigDraftType.DRAFT and draft.account_id is None:
-                    return AgentConfigDraftType.DRAFT.value
-        return AgentConfigDraftType.DRAFT.value
+            .where(
+                AgentConfigDraft.tenant_id == app_model.tenant_id,
+                AgentConfigDraft.draft_type == AgentConfigDraftType.DEBUG_BUILD,
+                AgentConfigDraft.account_id == user.id,
+                AgentWorkspaceBinding.tenant_id == app_model.tenant_id,
+                AgentWorkspaceBinding.status == AgentWorkingResourceStatus.ACTIVE,
+                AgentWorkspaceBinding.pending_form_id == form_id,
+            )
+        )
+        if build_draft is not None:
+            return AgentConfigDraftType.DEBUG_BUILD.value, build_draft.id
+        return AgentConfigDraftType.DRAFT.value, None
 
     def _generate_worker(
         self,
@@ -482,7 +499,7 @@ class AgentAppGenerator(MessageBasedAppGenerator):
                     invoke_from=application_generate_entity.invoke_from,
                 )
                 with session_factory.create_session() as session:
-                    _, _, agent_soul = self._resolve_agent_by_id(
+                    agent, config_version, agent_soul = self._resolve_agent_by_id(
                         tenant_id=app_config.tenant_id,
                         agent_id=application_generate_entity.agent_id,
                         snapshot_id=application_generate_entity.agent_config_snapshot_id,
@@ -496,13 +513,18 @@ class AgentAppGenerator(MessageBasedAppGenerator):
                     agent_config_snapshot_id=application_generate_entity.agent_config_snapshot_id,
                     agent_config_version_kind=application_generate_entity.agent_config_version_kind,
                     agent_soul=agent_soul,
+                    home_snapshot_id=config_version.home_snapshot_id,
                     conversation_id=conversation.id,
                     query=query,
                     message_id=message.id,
                     model_name=application_generate_entity.model_conf.model,
                     queue_manager=queue_manager,
-                    session_scope_snapshot_id=application_generate_entity.agent_runtime_session_snapshot_id,
-                    agent_runtime_exit_intent=application_generate_entity.agent_runtime_exit_intent,
+                    session_scope_snapshot_id=application_generate_entity.agent_session_scope_config_version_id,
+                    build_draft_id=(
+                        application_generate_entity.agent_config_snapshot_id
+                        if application_generate_entity.agent_config_version_kind == AgentConfigVersionKind.BUILD_DRAFT
+                        else None
+                    ),
                 )
             except GenerateTaskStoppedError:
                 pass
@@ -519,18 +541,6 @@ class AgentAppGenerator(MessageBasedAppGenerator):
             raise AgentAppGeneratorError("query is required")
         return query.replace("\x00", "")
 
-    @staticmethod
-    def _resolve_agent_runtime_exit_intent(args: Mapping[str, Any]) -> AgentRuntimeExitIntent:
-        """Resolve API-internal runtime exit policy from controller-owned args.
-
-        Only the private controller-injected "delete" value changes behavior.
-        Normal chat and resume flows default/fallback to "suspend" so public
-        payloads and invalid internal values preserve existing semantics.
-        """
-        if args.get(AGENT_RUNTIME_EXIT_INTENT_ARG) == "delete":
-            return "delete"
-        return "suspend"
-
     @staticmethod
     def _build_runner(dify_context: DifyRunContext) -> AgentAppRunner:
         credentials_provider, _ = build_dify_model_access(dify_context)
@@ -546,7 +556,7 @@ class AgentAppGenerator(MessageBasedAppGenerator):
                 stream_run_timeout_seconds=dify_config.AGENT_BACKEND_RUN_TIMEOUT_SECONDS,
             ),
             event_adapter=AgentBackendRunEventAdapter(),
-            session_store=AgentAppRuntimeSessionStore(),
+            session_store=AgentAppWorkspaceStore(),
             text_delta_debounce_seconds=dify_config.AGENT_APP_TEXT_DELTA_DEBOUNCE_SECONDS,
         )
 
@@ -610,8 +620,10 @@ class AgentAppGenerator(MessageBasedAppGenerator):
         *,
         invoke_from: InvokeFrom,
         draft_type: Any,
+        draft_id: str | None = None,
         user: Account | EndUser,
         session: Session,
+        conversation: Conversation | None = None,
     ) -> tuple[Agent, str, Literal["snapshot", "draft", "build_draft"], AgentSoulConfig]:
         agent = session.scalar(
             select(Agent)
@@ -643,6 +655,7 @@ class AgentAppGenerator(MessageBasedAppGenerator):
                 tenant_id=app_model.tenant_id,
                 agent=agent,
                 draft_type=draft_type,
+                draft_id=draft_id,
                 account_id=user.id if isinstance(user, Account) else None,
                 session=session,
             )
@@ -655,28 +668,82 @@ class AgentAppGenerator(MessageBasedAppGenerator):
         # Public runtime must keep serving the active snapshot even when unpublished draft edits exist.
         if not agent.active_config_snapshot_id:
             raise AgentAppNotPublishedError("Agent has not been published")
+        conversation_binding = self._resolve_conversation_binding(
+            session=session,
+            tenant_id=app_model.tenant_id,
+            app_id=app_model.id,
+            agent_id=agent.id,
+            conversation=conversation,
+        )
+        snapshot_id = (
+            conversation_binding.agent_config_version_id
+            if conversation_binding is not None
+            else agent.active_config_snapshot_id
+        )
         _, snapshot, agent_soul = self._resolve_agent_by_id(
             tenant_id=app_model.tenant_id,
             agent_id=agent.id,
-            snapshot_id=agent.active_config_snapshot_id,
+            snapshot_id=snapshot_id,
             session=session,
         )
+        if conversation_binding is not None:
+            AgentWorkspaceService.validate_binding_generation(
+                conversation_binding,
+                base_home_snapshot_id=snapshot.home_snapshot_id,
+                agent_config_version_id=snapshot.id,
+                agent_config_version_kind=AgentConfigVersionKind.SNAPSHOT,
+            )
         return agent, snapshot.id, "snapshot", agent_soul
 
     @staticmethod
-    def _runtime_session_snapshot_id(*, invoke_from: InvokeFrom, snapshot_id: str) -> str | None:
-        """Return the session scope snapshot id for Agent App runtime state.
+    def _resolve_conversation_binding(
+        *,
+        session: Session,
+        tenant_id: str,
+        app_id: str,
+        agent_id: str,
+        conversation: Conversation | None,
+    ) -> AgentWorkspaceBinding | None:
+        """Resolve the exact participant generation owned by an existing conversation."""
+
+        if conversation is None or conversation.agent_workspace_binding_id is None:
+            return None
+        binding = AgentWorkspaceService.get_active_binding(
+            session=session,
+            tenant_id=tenant_id,
+            binding_id=conversation.agent_workspace_binding_id,
+            expected_owner_scope=WorkspaceOwnerScope(
+                tenant_id=tenant_id,
+                app_id=app_id,
+                owner_type=AgentWorkspaceOwnerType.CONVERSATION,
+                owner_id=conversation.id,
+            ),
+        )
+        if binding is None or binding.agent_id != agent_id:
+            raise AgentAppGeneratorError("Conversation participant Binding is unavailable")
+        return binding
+
+    @staticmethod
+    def _session_scope_config_version_id(*, invoke_from: InvokeFrom, config_version_id: str) -> str | None:
+        """Return the config version id that scopes Agent App session reuse.
 
         Console preview/debug chat uses a stable Agent draft row id; build mode
         uses the current user's build-draft row id. Published/web/API runs use
-        immutable published snapshot ids. This keeps runtime session continuity
+        immutable published snapshot ids. This keeps Workspace Binding continuity
         inside one editable surface without mixing draft/build/published state.
         """
-        return snapshot_id
+        del invoke_from
+        return config_version_id
 
     @staticmethod
     def _resolve_debug_draft(
-        *, tenant_id: str, agent: Agent, draft_type: Any, account_id: str | None, session: Session
+        *,
+        tenant_id: str,
+        agent: Agent,
+        draft_type: Any,
+        account_id: str | None,
+        session: Session,
+        draft_id: str | None = None,
     ) -> AgentConfigDraft:
         effective_draft_type = (
             AgentConfigDraftType.DEBUG_BUILD
@@ -700,6 +767,8 @@ class AgentAppGenerator(MessageBasedAppGenerator):
             AgentConfigDraft.draft_type == AgentConfigDraftType.DEBUG_BUILD,
             AgentConfigDraft.account_id == account_id,
         )
+        if draft_id is not None:
+            stmt = stmt.where(AgentConfigDraft.id == draft_id)
         draft = session.scalar(stmt.order_by(AgentConfigDraft.updated_at.desc()).limit(1))
         if draft is not None:
             return draft

api/core/app/apps/agent_app/app_runner.py

@@ -3,8 +3,7 @@
 Unlike the legacy ``AgentChatAppRunner`` (which runs an in-process ReAct loop),
 this runner delegates to the Agent backend, consumes the streamed event flow,
 republishes the assistant answer through the existing EasyUI chat task
-pipeline, and then either saves or retires the conversation-owned runtime
-session depending on the turn's exit policy.
+pipeline, and saves the latest Agenton snapshot on the persistent Binding.
 """
 
 from __future__ import annotations
@@ -31,22 +30,20 @@ from clients.agent_backend import (
     AgentBackendRunFailedInternalEvent,
     AgentBackendRunSucceededInternalEvent,
     AgentBackendStreamInternalEvent,
-    extract_runtime_layer_specs,
 )
-from clients.agent_backend.session_cleanup import AgentBackendSessionCleanupPayload
 from core.app.apps.agent_app.runtime_request_builder import (
     AgentAppRuntimeBuildContext,
     AgentAppRuntimeRequest,
     AgentAppRuntimeRequestBuilder,
 )
 from core.app.apps.agent_app.session_store import (
-    AgentAppRuntimeSessionStore,
     AgentAppSessionScope,
+    AgentAppWorkspaceStore,
     StoredAgentAppSession,
 )
 from core.app.apps.base_app_queue_manager import AppQueueManager, PublishFrom
 from core.app.apps.exc import GenerateTaskStoppedError
-from core.app.entities.app_invoke_entities import AgentRuntimeExitIntent, DifyRunContext
+from core.app.entities.app_invoke_entities import DifyRunContext
 from core.app.entities.queue_entities import (
     QueueAgentMessageEvent,
     QueueAgentThoughtEvent,
@@ -67,10 +64,10 @@ from graphon.model_runtime.errors.invoke import (
     InvokeRateLimitError,
     InvokeServerUnavailableError,
 )
+from models.agent import AgentConfigVersionKind
 from models.agent_config_entities import AgentSoulConfig
 from models.enums import CreatorUserRole
 from models.model import MessageAgentThought
-from tasks.agent_backend_session_cleanup_task import cleanup_conversation_agent_runtime_session
 
 logger = logging.getLogger(__name__)
 
@@ -620,7 +617,7 @@ class AgentAppRunner:
         request_builder: AgentAppRuntimeRequestBuilder,
         agent_backend_client: AgentBackendRunClient,
         event_adapter: AgentBackendRunEventAdapter,
-        session_store: AgentAppRuntimeSessionStore,
+        session_store: AgentAppWorkspaceStore,
         text_delta_debounce_seconds: float,
     ) -> None:
         self._request_builder = request_builder
@@ -637,37 +634,41 @@ class AgentAppRunner:
         agent_config_snapshot_id: str,
         agent_config_version_kind: Literal["snapshot", "draft", "build_draft"] = "snapshot",
         agent_soul: AgentSoulConfig,
+        home_snapshot_id: str,
         conversation_id: str,
         query: str,
         message_id: str,
         model_name: str,
         queue_manager: AppQueueManager,
         session_scope_snapshot_id: str | None | _DefaultSessionScopeSnapshotId = _DEFAULT_SESSION_SCOPE_SNAPSHOT_ID,
-        agent_runtime_exit_intent: AgentRuntimeExitIntent = "suspend",
+        build_draft_id: str | None = None,
     ) -> None:
-        preserve_session = agent_runtime_exit_intent == "suspend"
         scope = self._build_session_scope(
             dify_context=dify_context,
             agent_id=agent_id,
             agent_config_snapshot_id=agent_config_snapshot_id,
+            home_snapshot_id=home_snapshot_id,
             conversation_id=conversation_id,
             session_scope_snapshot_id=session_scope_snapshot_id,
+            agent_config_version_kind=AgentConfigVersionKind(agent_config_version_kind),
+            build_draft_id=build_draft_id,
         )
         # ENG-638: if a prior turn paused on ask_human and the form is now answered,
         # resume by threading the human's reply into this run as deferred_tool_results.
-        stored = self._session_store.load_active_session(scope)
+        stored = self._session_store.load_or_create(scope)
         runtime = self._build_runtime(
             dify_context=dify_context,
             agent_id=agent_id,
             agent_config_snapshot_id=agent_config_snapshot_id,
             agent_config_version_kind=agent_config_version_kind,
             agent_soul=agent_soul,
+            binding_id=stored.binding_id,
+            backend_binding_ref=stored.backend_binding_ref,
             conversation_id=conversation_id,
             query=query,
             idempotency_key=message_id,
             stored=stored,
             message_id=message_id,
-            suspend_on_exit=preserve_session,
         )
 
         create_response = self._agent_backend_client.create_run(runtime.request)
@@ -681,9 +682,6 @@ class AgentAppRunner:
         )
 
         if isinstance(terminal, AgentBackendDeferredToolCallInternalEvent):
-            if not preserve_session:
-                self._mark_session_cleaned(scope=scope, backend_run_id=terminal.run_id)
-                raise AgentBackendError("Agent App finalization cannot pause for human input.")
             # ENG-635: the agent asked a human. End this turn with the question and
             # a conversation-owned HITL form; a form submission resumes the run.
             self._pause_for_ask_human(
@@ -703,8 +701,8 @@ class AgentAppRunner:
         if not isinstance(terminal, AgentBackendRunSucceededInternalEvent):
             if isinstance(terminal, AgentBackendRunFailedInternalEvent):
                 reason = terminal.reason
-                if reason == "sandbox_expired":
-                    raise AgentBackendError("The agent session sandbox has expired. Please start a new conversation.")
+                if reason == "binding_lost":
+                    raise AgentBackendError("The retained agent working environment is no longer available.")
                 raise _agent_backend_failure_to_exception(terminal)
             raise AgentBackendError("Agent backend run did not complete successfully.")
 
@@ -719,38 +717,18 @@ class AgentAppRunner:
                 message_id,
                 exc_info=True,
             )
-        if preserve_session:
-            superseded_sessions = self._load_superseded_sessions(scope=scope)
-            self._publish_terminal_answer(
-                queue_manager=queue_manager,
-                model_name=model_name,
-                answer=answer,
-                query=query,
-                usage=_llm_usage_from_agent_backend(terminal.usage),
-            )
-            session_saved = self._save_session(
-                scope=scope,
-                backend_run_id=terminal.run_id,
-                snapshot=terminal.session_snapshot,
-                runtime_layer_specs=extract_runtime_layer_specs(runtime.request.composition),
-            )
-            if session_saved:
-                self._cleanup_superseded_sessions(superseded_sessions)
-        else:
-            # The backend has already accepted a terminal success with
-            # delete-on-exit semantics. Local publish/persistence errors must
-            # not keep the API-side session row active, and cleanup failures
-            # must not replace the original publish/error outcome.
-            try:
-                self._publish_terminal_answer(
-                    queue_manager=queue_manager,
-                    model_name=model_name,
-                    answer=answer,
-                    query=query,
-                    usage=_llm_usage_from_agent_backend(terminal.usage),
-                )
-            finally:
-                self._mark_session_cleaned(scope=scope, backend_run_id=terminal.run_id)
+        self._publish_terminal_answer(
+            queue_manager=queue_manager,
+            model_name=model_name,
+            answer=answer,
+            query=query,
+            usage=_llm_usage_from_agent_backend(terminal.usage),
+        )
+        self._save_session(
+            scope=scope,
+            binding_id=runtime.binding_id,
+            snapshot=terminal.session_snapshot,
+        )
 
     def _build_session_scope(
         self,
@@ -758,8 +736,11 @@ class AgentAppRunner:
         dify_context: DifyRunContext,
         agent_id: str,
         agent_config_snapshot_id: str,
+        home_snapshot_id: str,
         conversation_id: str,
         session_scope_snapshot_id: str | None | _DefaultSessionScopeSnapshotId,
+        agent_config_version_kind: AgentConfigVersionKind,
+        build_draft_id: str | None = None,
     ) -> AgentAppSessionScope:
         if isinstance(session_scope_snapshot_id, _DefaultSessionScopeSnapshotId):
             effective_session_scope_snapshot_id: str | None = agent_config_snapshot_id
@@ -770,7 +751,10 @@ class AgentAppRunner:
             app_id=dify_context.app_id,
             conversation_id=conversation_id,
             agent_id=agent_id,
-            agent_config_snapshot_id=effective_session_scope_snapshot_id,
+            agent_config_snapshot_id=effective_session_scope_snapshot_id or agent_config_snapshot_id,
+            home_snapshot_id=home_snapshot_id,
+            agent_config_version_kind=agent_config_version_kind,
+            build_draft_id=build_draft_id,
         )
 
     def _build_runtime(
@@ -781,14 +765,15 @@ class AgentAppRunner:
         agent_config_snapshot_id: str,
         agent_config_version_kind: Literal["snapshot", "draft", "build_draft"],
         agent_soul: AgentSoulConfig,
+        binding_id: str,
+        backend_binding_ref: str,
         conversation_id: str,
         query: str,
         idempotency_key: str,
-        stored: StoredAgentAppSession | None,
+        stored: StoredAgentAppSession,
         message_id: str | None,
-        suspend_on_exit: bool,
     ) -> AgentAppRuntimeRequest:
-        session_snapshot = stored.session_snapshot if stored is not None else None
+        session_snapshot = stored.session_snapshot
         deferred_tool_results = (
             self._resolve_pending_ask_human(stored=stored, dify_context=dify_context, message_id=message_id)
             if message_id is not None
@@ -804,9 +789,10 @@ class AgentAppRunner:
                 conversation_id=conversation_id,
                 user_query=query,
                 idempotency_key=idempotency_key,
+                binding_id=binding_id,
+                backend_binding_ref=backend_binding_ref,
                 session_snapshot=session_snapshot,
                 deferred_tool_results=deferred_tool_results,
-                suspend_on_exit=suspend_on_exit,
             )
         )
 
@@ -843,9 +829,8 @@ class AgentAppRunner:
         # second run with the human's answer (ENG-637/638 columns, conversation owner).
         self._save_session(
             scope=scope,
-            backend_run_id=terminal.run_id,
+            binding_id=runtime.binding_id,
             snapshot=terminal.session_snapshot,
-            runtime_layer_specs=extract_runtime_layer_specs(runtime.request.composition),
             pending_form_id=created.form_id,
             pending_tool_call_id=terminal.deferred_tool_call.tool_call_id,
         )
@@ -862,12 +847,12 @@ class AgentAppRunner:
     def _resolve_pending_ask_human(
         self,
         *,
-        stored: StoredAgentAppSession | None,
+        stored: StoredAgentAppSession,
         dify_context: DifyRunContext,
         message_id: str,
     ) -> DeferredToolResultsPayload | None:
         """Build deferred_tool_results when a pending ask_human form is answered."""
-        if stored is None or stored.pending_form_id is None or stored.pending_tool_call_id is None:
+        if stored.pending_form_id is None or stored.pending_tool_call_id is None:
             return None
         outcome = resolve_ask_human_form(
             form_id=stored.pending_form_id,
@@ -1036,18 +1021,16 @@ class AgentAppRunner:
         self,
         *,
         scope: AgentAppSessionScope,
-        backend_run_id: str,
+        binding_id: str,
         snapshot: Any,
-        runtime_layer_specs: Any,
         pending_form_id: str | None = None,
         pending_tool_call_id: str | None = None,
     ) -> bool:
         try:
             self._session_store.save_active_snapshot(
                 scope=scope,
-                backend_run_id=backend_run_id,
+                binding_id=binding_id,
                 snapshot=snapshot,
-                runtime_layer_specs=runtime_layer_specs,
                 pending_form_id=pending_form_id,
                 pending_tool_call_id=pending_tool_call_id,
             )
@@ -1064,87 +1047,6 @@ class AgentAppRunner:
             )
             return False
 
-    def _load_superseded_sessions(self, *, scope: AgentAppSessionScope) -> list[StoredAgentAppSession]:
-        try:
-            stored_sessions = self._session_store.list_active_sessions_for_conversation(
-                tenant_id=scope.tenant_id,
-                app_id=scope.app_id,
-                conversation_id=scope.conversation_id,
-            )
-        except Exception:
-            logger.warning(
-                "Failed to load existing Agent App conversation sessions before snapshot save: "
-                "tenant_id=%s app_id=%s conversation_id=%s agent_id=%s",
-                scope.tenant_id,
-                scope.app_id,
-                scope.conversation_id,
-                scope.agent_id,
-                exc_info=True,
-            )
-            return []
-
-        return [stored for stored in stored_sessions if stored.scope != scope]
-
-    def _cleanup_superseded_sessions(self, stored_sessions: list[StoredAgentAppSession]) -> None:
-        for stored_session in stored_sessions:
-            try:
-                if stored_session.runtime_layer_specs:
-                    payload = AgentBackendSessionCleanupPayload(
-                        session_snapshot=stored_session.session_snapshot,
-                        runtime_layer_specs=stored_session.runtime_layer_specs,
-                        idempotency_key=(
-                            f"{stored_session.scope.tenant_id}:{stored_session.scope.app_id}:"
-                            f"{stored_session.scope.conversation_id}:{stored_session.scope.agent_id}:"
-                            f"{stored_session.scope.agent_config_snapshot_id or 'no-config'}:"
-                            f"superseded-session-cleanup:{stored_session.backend_run_id or 'no-run'}"
-                        ),
-                        metadata={
-                            "tenant_id": stored_session.scope.tenant_id,
-                            "app_id": stored_session.scope.app_id,
-                            "conversation_id": stored_session.scope.conversation_id,
-                            "agent_id": stored_session.scope.agent_id,
-                            "agent_config_snapshot_id": stored_session.scope.agent_config_snapshot_id,
-                            "previous_agent_backend_run_id": stored_session.backend_run_id,
-                        },
-                    )
-                    cleanup_conversation_agent_runtime_session.delay(payload.model_dump(mode="json"))
-            except Exception:
-                logger.warning(
-                    "Failed to enqueue Agent backend cleanup for superseded Agent App session: "
-                    "tenant_id=%s app_id=%s conversation_id=%s agent_id=%s backend_run_id=%s",
-                    stored_session.scope.tenant_id,
-                    stored_session.scope.app_id,
-                    stored_session.scope.conversation_id,
-                    stored_session.scope.agent_id,
-                    stored_session.backend_run_id,
-                    exc_info=True,
-                )
-
-    def _mark_session_cleaned(
-        self,
-        *,
-        scope: AgentAppSessionScope,
-        backend_run_id: str,
-    ) -> None:
-        """Best-effort delete-on-exit cleanup for the API-side session row.
-
-        Once the Agent backend reaches a terminal event, cleanup persistence
-        must not replace the original publish/error outcome for that turn.
-        """
-        try:
-            self._session_store.mark_cleaned(scope=scope, backend_run_id=backend_run_id)
-        except Exception:
-            logger.warning(
-                "Failed to retire Agent App conversation session after delete-on-exit: "
-                "tenant_id=%s app_id=%s conversation_id=%s agent_id=%s backend_run_id=%s",
-                scope.tenant_id,
-                scope.app_id,
-                scope.conversation_id,
-                scope.agent_id,
-                backend_run_id,
-                exc_info=True,
-            )
-
     @staticmethod
     def _terminal_output_to_answer(output: JsonValue) -> str:
         """Normalize the backend's terminal output to assistant text.

api/core/app/apps/agent_app/runtime_request_builder.py

@@ -70,11 +70,12 @@ class AgentAppRuntimeBuildContext:
     conversation_id: str
     user_query: str
     idempotency_key: str
+    binding_id: str
+    backend_binding_ref: str
     agent_config_version_kind: Literal["snapshot", "draft", "build_draft"] = "snapshot"
     session_snapshot: CompositorSessionSnapshot | None = None
     # ENG-638: set when resuming a chat turn after a submitted ask_human form.
     deferred_tool_results: DeferredToolResultsPayload | None = None
-    suspend_on_exit: bool = True
 
 
 @dataclass(frozen=True, slots=True)
@@ -82,6 +83,7 @@ class AgentAppRuntimeRequest:
     request: CreateRunRequest
     redacted_request: dict[str, Any]
     metadata: dict[str, Any]
+    binding_id: str
 
 
 class AgentAppRuntimeRequestBuilder:
@@ -160,6 +162,7 @@ class AgentAppRuntimeRequestBuilder:
                     invoke_from=cast(DifyExecutionContextInvokeFrom, context.dify_context.invoke_from.value),
                     agent_mode="agent_app",
                 ),
+                backend_binding_ref=context.backend_binding_ref,
                 # ENG-616: expand slash-menu mention tokens to canonical names so
                 # no frontend-internal {{#…#}} marker ever reaches the model.
                 agent_soul_prompt=expand_prompt_mentions(agent_soul.prompt.system_prompt, soul_prompt_resolver).strip()
@@ -175,13 +178,17 @@ class AgentAppRuntimeRequestBuilder:
                 shell_config=build_shell_layer_config(agent_soul),
                 session_snapshot=context.session_snapshot,
                 deferred_tool_results=context.deferred_tool_results,
-                suspend_on_exit=context.suspend_on_exit,
                 idempotency_key=context.idempotency_key,
                 metadata=metadata,
             )
         )
         redacted = cast(dict[str, Any], redact_for_agent_backend_log(request))
-        return AgentAppRuntimeRequest(request=request, redacted_request=redacted, metadata=metadata)
+        return AgentAppRuntimeRequest(
+            request=request,
+            redacted_request=redacted,
+            metadata=metadata,
+            binding_id=context.binding_id,
+        )
 
     def _build_tool_layers(
         self,

api/core/app/apps/agent_app/session_store.py

@@ -1,255 +1,176 @@
-"""Conversation-keyed Agent backend session store for the Agent App type.
-
-Shares the unified ``agent_runtime_sessions`` table with the workflow Agent
-Node store, but owns rows with ``owner_type = conversation``: one Agent App
-conversation maps to one Agent session, so multi-turn chat re-enters the same
-``session_snapshot``. Cross-conversation memory (PRD Global / Per app) is a
-phase-2 concern and not modeled here.
-"""
+"""Persist and resolve the exact participant owned by an Agent App caller."""
 
 from __future__ import annotations
 
-from dataclasses import dataclass, field
+from dataclasses import dataclass
 
 from agenton.compositor import CompositorSessionSnapshot
-from dify_agent.protocol import RuntimeLayerSpec
-from pydantic import TypeAdapter
 from sqlalchemy import select
+from sqlalchemy.orm import Session
 
 from core.db.session_factory import session_factory
-from libs.datetime_utils import naive_utc_now
 from models.agent import (
-    AgentRuntimeSession,
-    AgentRuntimeSessionOwnerType,
-    AgentRuntimeSessionStatus,
+    AgentConfigDraft,
+    AgentConfigDraftType,
+    AgentConfigVersionKind,
+    AgentWorkspaceBinding,
+    AgentWorkspaceOwnerType,
+)
+from models.model import App, Conversation
+from services.agent.workspace_service import (
+    AgentWorkspaceNotFoundError,
+    AgentWorkspaceService,
+    WorkspaceOwnerScope,
 )
-
-_RUNTIME_LAYER_SPECS_ADAPTER: TypeAdapter[list[RuntimeLayerSpec]] = TypeAdapter(list[RuntimeLayerSpec])
-
-
-def _serialize_runtime_layer_specs(specs: list[RuntimeLayerSpec]) -> str:
-    return _RUNTIME_LAYER_SPECS_ADAPTER.dump_json(specs).decode()
-
-
-def _deserialize_runtime_layer_specs(value: str | None) -> list[RuntimeLayerSpec]:
-    if not value:
-        return []
-    return _RUNTIME_LAYER_SPECS_ADAPTER.validate_json(value)
 
 
 @dataclass(frozen=True, slots=True)
 class AgentAppSessionScope:
-    """Identity of one Agent App conversation session."""
-
     tenant_id: str
     app_id: str
     conversation_id: str
     agent_id: str
-    agent_config_snapshot_id: str | None
+    agent_config_snapshot_id: str
+    home_snapshot_id: str
+    agent_config_version_kind: AgentConfigVersionKind = AgentConfigVersionKind.SNAPSHOT
+    build_draft_id: str | None = None
+
+    @property
+    def workspace_owner(self) -> WorkspaceOwnerScope:
+        owner_type = (
+            AgentWorkspaceOwnerType.BUILD_DRAFT if self.build_draft_id else AgentWorkspaceOwnerType.CONVERSATION
+        )
+        return WorkspaceOwnerScope(
+            tenant_id=self.tenant_id,
+            app_id=self.app_id,
+            owner_type=owner_type,
+            owner_id=self.build_draft_id or self.conversation_id,
+        )
 
 
 @dataclass(frozen=True, slots=True)
 class StoredAgentAppSession:
-    """Persisted Agent App conversation session with reusable runtime specs."""
-
     scope: AgentAppSessionScope
-    session_snapshot: CompositorSessionSnapshot
-    backend_run_id: str | None
-    runtime_layer_specs: list[RuntimeLayerSpec] = field(default_factory=list)
-    # ENG-635: set while the conversation turn is paused on a dify.ask_human
-    # deferred call, awaiting a HITL form submission.
+    binding_id: str
+    workspace_id: str
+    backend_binding_ref: str
+    session_snapshot: CompositorSessionSnapshot | None
     pending_form_id: str | None = None
     pending_tool_call_id: str | None = None
 
 
-class AgentAppRuntimeSessionStore:
-    """Persists Agent backend session snapshots for Agent App conversations."""
+class AgentAppWorkspaceStore:
+    """Resolve Agent App sessions through a caller-owned Binding pointer."""
 
-    def load_active_snapshot(self, scope: AgentAppSessionScope) -> CompositorSessionSnapshot | None:
-        stored = self.load_active_session(scope)
-        return stored.session_snapshot if stored is not None else None
-
-    def load_active_session(self, scope: AgentAppSessionScope) -> StoredAgentAppSession | None:
+    def load_or_create(self, scope: AgentAppSessionScope) -> StoredAgentAppSession:
         with session_factory.create_session() as session:
-            row = session.scalar(self._active_stmt(scope))
-            if row is None:
-                return None
-            return StoredAgentAppSession(
-                scope=scope,
-                session_snapshot=CompositorSessionSnapshot.model_validate_json(row.session_snapshot),
-                backend_run_id=row.backend_run_id,
-                runtime_layer_specs=_deserialize_runtime_layer_specs(row.composition_layer_specs),
-                pending_form_id=row.pending_form_id,
-                pending_tool_call_id=row.pending_tool_call_id,
-            )
-
-    def load_active_session_for_conversation(
-        self, *, tenant_id: str, app_id: str, conversation_id: str
-    ) -> StoredAgentAppSession | None:
-        """Load the latest ACTIVE session for one conversation-level sandbox lookup.
-
-        Sandbox inspection only knows the product locator
-        ``tenant_id + app_id + conversation_id``; it does not know which
-        ``agent_id`` or Agent Soul snapshot produced the active shell session.
-        This method therefore resolves the newest ACTIVE conversation-owned row
-        for that conversation and returns both the resumable snapshot and the
-        persisted non-sensitive runtime layer specs needed to build a
-        ``SandboxLocator``.
-        """
-        stmt = (
-            select(AgentRuntimeSession)
-            .where(
-                AgentRuntimeSession.owner_type == AgentRuntimeSessionOwnerType.CONVERSATION,
-                AgentRuntimeSession.tenant_id == tenant_id,
-                AgentRuntimeSession.app_id == app_id,
-                AgentRuntimeSession.conversation_id == conversation_id,
-                AgentRuntimeSession.status == AgentRuntimeSessionStatus.ACTIVE,
-            )
-            .order_by(AgentRuntimeSession.updated_at.desc())
-        )
-        with session_factory.create_session() as session:
-            row = session.scalar(stmt)
-            if row is None:
-                return None
-            return StoredAgentAppSession(
-                scope=AgentAppSessionScope(
-                    tenant_id=row.tenant_id,
-                    app_id=row.app_id,
-                    conversation_id=row.conversation_id or "",
-                    agent_id=row.agent_id,
-                    agent_config_snapshot_id=row.agent_config_snapshot_id or "",
-                ),
-                session_snapshot=CompositorSessionSnapshot.model_validate_json(row.session_snapshot),
-                backend_run_id=row.backend_run_id,
-                runtime_layer_specs=_deserialize_runtime_layer_specs(row.composition_layer_specs),
-            )
-
-    def list_active_sessions_for_conversation(
-        self, *, tenant_id: str, app_id: str, conversation_id: str
-    ) -> list[StoredAgentAppSession]:
-        """List all ACTIVE conversation-owned sessions for lifecycle cleanup."""
-        stmt = (
-            select(AgentRuntimeSession)
-            .where(
-                AgentRuntimeSession.owner_type == AgentRuntimeSessionOwnerType.CONVERSATION,
-                AgentRuntimeSession.tenant_id == tenant_id,
-                AgentRuntimeSession.app_id == app_id,
-                AgentRuntimeSession.conversation_id == conversation_id,
-                AgentRuntimeSession.status == AgentRuntimeSessionStatus.ACTIVE,
-            )
-            .order_by(AgentRuntimeSession.updated_at.desc())
-        )
-        with session_factory.create_session() as session:
-            rows = session.scalars(stmt).all()
-            return [
-                StoredAgentAppSession(
-                    scope=AgentAppSessionScope(
-                        tenant_id=row.tenant_id,
-                        app_id=row.app_id,
-                        conversation_id=row.conversation_id or "",
-                        agent_id=row.agent_id,
-                        agent_config_snapshot_id=row.agent_config_snapshot_id,
-                    ),
-                    session_snapshot=CompositorSessionSnapshot.model_validate_json(row.session_snapshot),
-                    backend_run_id=row.backend_run_id,
-                    runtime_layer_specs=_deserialize_runtime_layer_specs(row.composition_layer_specs),
-                    pending_form_id=row.pending_form_id,
-                    pending_tool_call_id=row.pending_tool_call_id,
+            caller = self._load_caller(session=session, scope=scope)
+            binding_id = caller.agent_workspace_binding_id
+            if binding_id is None:
+                binding = AgentWorkspaceService.create_binding(
+                    session=session,
+                    scope=scope.workspace_owner,
+                    agent_id=scope.agent_id,
+                    base_home_snapshot_id=scope.home_snapshot_id,
+                    agent_config_version_id=scope.agent_config_snapshot_id,
+                    agent_config_version_kind=scope.agent_config_version_kind,
                 )
-                for row in rows
-            ]
+                caller.agent_workspace_binding_id = binding.id
+                session.commit()
+            else:
+                binding = self._get_binding(session=session, scope=scope, binding_id=binding_id)
+            return self._stored(scope, binding)
+
+    @staticmethod
+    def _load_caller(*, session: Session, scope: AgentAppSessionScope) -> Conversation | AgentConfigDraft:
+        if scope.build_draft_id is not None:
+            if scope.agent_config_version_kind != AgentConfigVersionKind.BUILD_DRAFT:
+                raise AgentWorkspaceNotFoundError("Build Draft caller requires build_draft generation")
+            draft = session.scalar(
+                select(AgentConfigDraft).where(
+                    AgentConfigDraft.id == scope.build_draft_id,
+                    AgentConfigDraft.tenant_id == scope.tenant_id,
+                    AgentConfigDraft.agent_id == scope.agent_id,
+                    AgentConfigDraft.draft_type == AgentConfigDraftType.DEBUG_BUILD,
+                )
+            )
+            if draft is None:
+                raise AgentWorkspaceNotFoundError("Build Draft caller is unavailable")
+            return draft
+        if scope.agent_config_version_kind == AgentConfigVersionKind.BUILD_DRAFT:
+            raise AgentWorkspaceNotFoundError("Build Draft caller ID is required")
+        conversation = session.scalar(
+            select(Conversation)
+            .join(App, App.id == Conversation.app_id)
+            .where(
+                App.tenant_id == scope.tenant_id,
+                Conversation.id == scope.conversation_id,
+                Conversation.app_id == scope.app_id,
+                Conversation.is_deleted.is_(False),
+            )
+        )
+        if conversation is None:
+            raise AgentWorkspaceNotFoundError("Conversation caller is unavailable")
+        return conversation
+
+    @staticmethod
+    def _get_binding(
+        *,
+        session: Session,
+        scope: AgentAppSessionScope,
+        binding_id: str,
+    ) -> AgentWorkspaceBinding:
+        binding = AgentWorkspaceService.get_active_binding(
+            session=session,
+            tenant_id=scope.tenant_id,
+            binding_id=binding_id,
+            expected_owner_scope=scope.workspace_owner,
+        )
+        if binding is None or binding.agent_id != scope.agent_id:
+            raise AgentWorkspaceNotFoundError("Caller participant Binding is unavailable")
+        AgentWorkspaceService.validate_binding_generation(
+            binding,
+            base_home_snapshot_id=scope.home_snapshot_id,
+            agent_config_version_id=scope.agent_config_snapshot_id,
+            agent_config_version_kind=scope.agent_config_version_kind,
+        )
+        return binding
 
     def save_active_snapshot(
         self,
         *,
         scope: AgentAppSessionScope,
-        backend_run_id: str,
+        binding_id: str,
         snapshot: CompositorSessionSnapshot | None,
-        runtime_layer_specs: list[RuntimeLayerSpec],
         pending_form_id: str | None = None,
         pending_tool_call_id: str | None = None,
     ) -> None:
-        """Persist the current conversation snapshot and enforce one ACTIVE row.
-
-        Agent App chat treats one conversation as one resumable runtime shell.
-        Saving the latest snapshot therefore upserts the scoped row back to
-        ACTIVE and retires any other ACTIVE conversation-owned rows for the
-        same ``tenant_id + app_id + conversation_id`` so later lookups see a
-        single active session.
-        """
         if snapshot is None:
             return
-        snapshot_json = snapshot.model_dump_json()
-        runtime_layer_specs_json = _serialize_runtime_layer_specs(runtime_layer_specs)
-        with session_factory.create_session() as session:
-            row = session.scalar(self._scope_stmt(scope))
-            if row is None:
-                row = AgentRuntimeSession(
-                    tenant_id=scope.tenant_id,
-                    app_id=scope.app_id,
-                    owner_type=AgentRuntimeSessionOwnerType.CONVERSATION,
-                    agent_id=scope.agent_id,
-                    agent_config_snapshot_id=scope.agent_config_snapshot_id,
-                    conversation_id=scope.conversation_id,
-                    backend_run_id=backend_run_id,
-                    session_snapshot=snapshot_json,
-                    composition_layer_specs=runtime_layer_specs_json,
-                    status=AgentRuntimeSessionStatus.ACTIVE,
-                    pending_form_id=pending_form_id,
-                    pending_tool_call_id=pending_tool_call_id,
-                )
-                session.add(row)
-            else:
-                row.backend_run_id = backend_run_id
-                row.session_snapshot = snapshot_json
-                row.composition_layer_specs = runtime_layer_specs_json
-                row.status = AgentRuntimeSessionStatus.ACTIVE
-                row.cleaned_at = None
-                # Set (or clear, when omitted) the ask_human pause correlation.
-                row.pending_form_id = pending_form_id
-                row.pending_tool_call_id = pending_tool_call_id
-            session.flush()
-            other_rows = session.scalars(
-                select(AgentRuntimeSession).where(
-                    AgentRuntimeSession.owner_type == AgentRuntimeSessionOwnerType.CONVERSATION,
-                    AgentRuntimeSession.tenant_id == scope.tenant_id,
-                    AgentRuntimeSession.app_id == scope.app_id,
-                    AgentRuntimeSession.conversation_id == scope.conversation_id,
-                    AgentRuntimeSession.status == AgentRuntimeSessionStatus.ACTIVE,
-                    AgentRuntimeSession.id != row.id,
-                )
-            ).all()
-            for other_row in other_rows:
-                other_row.status = AgentRuntimeSessionStatus.CLEANED
-                other_row.cleaned_at = naive_utc_now()
-            session.commit()
-
-    def mark_cleaned(self, *, scope: AgentAppSessionScope, backend_run_id: str | None = None) -> None:
-        with session_factory.create_session() as session:
-            row = session.scalar(self._active_stmt(scope))
-            if row is None:
-                return
-            if backend_run_id is not None:
-                row.backend_run_id = backend_run_id
-            row.status = AgentRuntimeSessionStatus.CLEANED
-            row.cleaned_at = naive_utc_now()
-            session.commit()
+        AgentWorkspaceService.save_binding_session_snapshot(
+            tenant_id=scope.tenant_id,
+            binding_id=binding_id,
+            session_snapshot=snapshot.model_dump_json(),
+            pending_form_id=pending_form_id,
+            pending_tool_call_id=pending_tool_call_id,
+        )
 
     @staticmethod
-    def _scope_stmt(scope: AgentAppSessionScope):
-        stmt = select(AgentRuntimeSession).where(
-            AgentRuntimeSession.owner_type == AgentRuntimeSessionOwnerType.CONVERSATION,
-            AgentRuntimeSession.tenant_id == scope.tenant_id,
-            AgentRuntimeSession.conversation_id == scope.conversation_id,
-            AgentRuntimeSession.agent_id == scope.agent_id,
+    def _stored(scope: AgentAppSessionScope, binding: AgentWorkspaceBinding) -> StoredAgentAppSession:
+        snapshot = (
+            CompositorSessionSnapshot.model_validate_json(binding.session_snapshot)
+            if binding.session_snapshot
+            else None
+        )
+        return StoredAgentAppSession(
+            scope=scope,
+            binding_id=binding.id,
+            workspace_id=binding.workspace_id,
+            backend_binding_ref=binding.backend_binding_ref,
+            session_snapshot=snapshot,
+            pending_form_id=binding.pending_form_id,
+            pending_tool_call_id=binding.pending_tool_call_id,
         )
-        if scope.agent_config_snapshot_id is None:
-            return stmt.where(AgentRuntimeSession.agent_config_snapshot_id.is_(None))
-        return stmt.where(AgentRuntimeSession.agent_config_snapshot_id == scope.agent_config_snapshot_id)
-
-    @classmethod
-    def _active_stmt(cls, scope: AgentAppSessionScope):
-        return cls._scope_stmt(scope).where(AgentRuntimeSession.status == AgentRuntimeSessionStatus.ACTIVE)
 
 
-__all__ = ["AgentAppRuntimeSessionStore", "AgentAppSessionScope", "StoredAgentAppSession"]
+__all__ = ["AgentAppSessionScope", "AgentAppWorkspaceStore", "StoredAgentAppSession"]

api/core/app/apps/workflow/app_runner.py

@@ -10,11 +10,11 @@ from core.app.apps.workflow.command_channels import (
     CombinedCommandChannel,
 )
 from core.app.apps.workflow_app_runner import WorkflowBasedAppRunner
-from core.app.entities.app_invoke_entities import InvokeFrom, WorkflowAppGenerateEntity
+from core.app.entities.app_invoke_entities import DifyRunContext, InvokeFrom, WorkflowAppGenerateEntity
 from core.app.workflow.layers.persistence import PersistenceWorkflowInfo, WorkflowPersistenceLayer
 from core.repositories.factory import WorkflowExecutionRepository, WorkflowNodeExecutionRepository
 from core.workflow.node_factory import get_default_root_node_id
-from core.workflow.nodes.agent_v2.session_cleanup_layer import build_workflow_agent_session_cleanup_layer
+from core.workflow.nodes.agent_v2.workspace_retirement_layer import build_workflow_agent_workspace_retirement_layer
 from core.workflow.snippet_start import get_compatible_start_aliases
 from core.workflow.system_variables import build_bootstrap_variables, build_system_variables
 from core.workflow.variable_pool_initializer import add_node_inputs_to_pool, add_variables_to_pool
@@ -197,7 +197,18 @@ class WorkflowAppRunner(WorkflowBasedAppRunner):
         )
 
         workflow_entry.graph_engine.layer(persistence_layer)
-        workflow_entry.graph_engine.layer(build_workflow_agent_session_cleanup_layer())
+        workflow_entry.graph_engine.layer(
+            build_workflow_agent_workspace_retirement_layer(
+                dify_run_context=DifyRunContext(
+                    tenant_id=self._workflow.tenant_id,
+                    app_id=self._workflow.app_id,
+                    user_id=self.application_generate_entity.user_id,
+                    user_from=user_from,
+                    invoke_from=invoke_from,
+                    trace_session_id=self.application_generate_entity.extras.get("trace_session_id"),
+                )
+            )
+        )
         for layer in self._graph_engine_layers:
             workflow_entry.graph_engine.layer(layer)
 

api/core/app/entities/app_invoke_entities.py

@@ -15,8 +15,6 @@ if TYPE_CHECKING:
 
 
 DIFY_RUN_CONTEXT_KEY = "_dify"
-AGENT_RUNTIME_EXIT_INTENT_ARG = "_agent_runtime_exit_intent"
-type AgentRuntimeExitIntent = Literal["suspend", "delete"]
 
 
 class UserFrom(StrEnum):
@@ -227,12 +225,8 @@ class AgentAppGenerateEntity(ChatAppGenerateEntity):
     backend should read from: immutable snapshot, shared draft, or per-user
     build draft.
 
-    ``agent_runtime_session_snapshot_id`` carries the runtime session scope
-    used to resume or suspend within the same editable config surface.
-
-    ``agent_runtime_exit_intent`` is API-internal lifecycle policy for the
-    Agent backend session after this turn finishes. Normal chat/resume turns
-    suspend on exit; build-chat finalization deletes the backend runtime.
+    ``agent_session_scope_config_version_id`` identifies the draft or immutable
+    config version whose Workspace Binding should be reused for this session.
 
     ``prompt_file_mappings`` preserves the raw request ``files`` array for the
     Agent backend prompt. These references are appended to the backend prompt
@@ -242,8 +236,7 @@ class AgentAppGenerateEntity(ChatAppGenerateEntity):
     agent_id: str
     agent_config_snapshot_id: str
     agent_config_version_kind: Literal["snapshot", "draft", "build_draft"] = "snapshot"
-    agent_runtime_session_snapshot_id: str | None = None
-    agent_runtime_exit_intent: AgentRuntimeExitIntent = "suspend"
+    agent_session_scope_config_version_id: str | None = None
     prompt_file_mappings: Sequence[JsonValue] = Field(default_factory=list)
 
 

api/core/app/workflow/layers/persistence.py

@@ -20,11 +20,13 @@ from core.helper.trace_id_helper import ParentTraceContext
 from core.ops.entities.trace_entity import TraceTaskName
 from core.ops.ops_trace_manager import TraceQueueManager, TraceTask
 from core.repositories.factory import WorkflowExecutionRepository, WorkflowNodeExecutionRepository
+from core.workflow.node_execution_process_data import preserve_workflow_agent_binding_id
 from core.workflow.system_variables import SystemVariableKey
 from core.workflow.variable_prefixes import SYSTEM_VARIABLE_NODE_ID
 from core.workflow.workflow_run_outputs import project_node_outputs_for_workflow_run
 from graphon.entities import WorkflowExecution, WorkflowNodeExecution
 from graphon.enums import (
+    BuiltinNodeTypes,
     WorkflowExecutionStatus,
     WorkflowNodeExecutionMetadataKey,
     WorkflowNodeExecutionStatus,
@@ -241,7 +243,10 @@ class WorkflowPersistenceLayer(GraphEngineLayer):
         )
 
         self._node_execution_cache[event.id] = domain_execution
-        self._workflow_node_execution_repository.save(domain_execution)
+        if event.node_type == BuiltinNodeTypes.AGENT and event.node_version == "2":
+            self._workflow_node_execution_repository.save_synchronously(domain_execution)
+        else:
+            self._workflow_node_execution_repository.save(domain_execution)
 
         snapshot = _NodeRuntimeSnapshot(
             node_id=event.node_id,
@@ -361,7 +366,11 @@ class WorkflowPersistenceLayer(GraphEngineLayer):
     def _append_retry_history(self, execution: WorkflowNodeExecution, event: NodeRunRetryEvent) -> None:
         """Append a validated full attempt before repository truncation or offload."""
         finished_at = naive_utc_now()
-        process_data = dict(execution.process_data or {})
+        process_data = preserve_workflow_agent_binding_id(
+            event.node_run_result.process_data,
+            execution.process_data,
+        )
+        process_data = dict(process_data or {})
         raw_history = process_data.get(RETRY_HISTORY_PROCESS_DATA_KEY)
         history = list(raw_history) if isinstance(raw_history, list) else []
         projected_outputs = project_node_outputs_for_workflow_run(
@@ -390,11 +399,12 @@ class WorkflowPersistenceLayer(GraphEngineLayer):
         next_process_data: Mapping[str, Any] | None,
     ) -> Mapping[str, Any] | None:
         """Keep internal retry history while replacing node-specific Process Data."""
+        merged_process_data = preserve_workflow_agent_binding_id(existing_process_data, next_process_data)
         raw_history = (existing_process_data or {}).get(RETRY_HISTORY_PROCESS_DATA_KEY)
         if not isinstance(raw_history, list) or not raw_history:
-            return next_process_data
+            return merged_process_data
 
-        merged_process_data = dict(next_process_data or {})
+        merged_process_data = dict(merged_process_data or {})
         merged_process_data[RETRY_HISTORY_PROCESS_DATA_KEY] = raw_history
         return merged_process_data
 
@@ -440,6 +450,11 @@ class WorkflowPersistenceLayer(GraphEngineLayer):
                 outputs=projected_outputs,
                 metadata=node_result.metadata,
             )
+        else:
+            domain_execution.process_data = preserve_workflow_agent_binding_id(
+                node_result.process_data,
+                domain_execution.process_data,
+            )
 
         self._workflow_node_execution_repository.save(domain_execution)
         self._workflow_node_execution_repository.save_execution_data(domain_execution)

api/core/repositories/celery_workflow_node_execution_repository.py

@@ -16,6 +16,9 @@ from core.repositories.factory import (
     OrderConfig,
     WorkflowNodeExecutionRepository,
 )
+from core.repositories.sqlalchemy_workflow_node_execution_repository import (
+    SQLAlchemyWorkflowNodeExecutionRepository,
+)
 from graphon.entities import WorkflowNodeExecution
 from models import Account, CreatorUserRole, EndUser
 from models.workflow import WorkflowNodeExecutionTriggeredFrom
@@ -49,6 +52,7 @@ class CeleryWorkflowNodeExecutionRepository(WorkflowNodeExecutionRepository):
     _creator_user_role: CreatorUserRole
     _execution_cache: dict[str, WorkflowNodeExecution]
     _workflow_execution_mapping: dict[str, list[str]]
+    _sql_repository: SQLAlchemyWorkflowNodeExecutionRepository
 
     def __init__(
         self,
@@ -98,6 +102,13 @@ class CeleryWorkflowNodeExecutionRepository(WorkflowNodeExecutionRepository):
 
         # Cache for mapping workflow_execution_ids to execution IDs for efficient retrieval
         self._workflow_execution_mapping = {}
+        self._sql_repository = SQLAlchemyWorkflowNodeExecutionRepository(
+            session_factory=session_factory,
+            tenant_id=tenant_id,
+            user=user,
+            app_id=app_id,
+            triggered_from=triggered_from,
+        )
 
         logger.info(
             "Initialized CeleryWorkflowNodeExecutionRepository for tenant %s, app %s, triggered_from %s",
@@ -149,6 +160,17 @@ class CeleryWorkflowNodeExecutionRepository(WorkflowNodeExecutionRepository):
             # For now, we'll re-raise the exception
             raise
 
+    @override
+    def save_synchronously(self, execution: WorkflowNodeExecution) -> None:
+        """Create the Agent v2 caller row before runtime participant allocation."""
+
+        self._sql_repository.save_synchronously(execution)
+        self._execution_cache[execution.id] = execution
+        if execution.workflow_execution_id:
+            execution_ids = self._workflow_execution_mapping.setdefault(execution.workflow_execution_id, [])
+            if execution.id not in execution_ids:
+                execution_ids.append(execution.id)
+
     @override
     def get_by_workflow_execution(
         self,

api/core/repositories/factory.py

@@ -35,6 +35,8 @@ class WorkflowExecutionRepository(Protocol):
 class WorkflowNodeExecutionRepository(Protocol):
     def save(self, execution: WorkflowNodeExecution): ...
 
+    def save_synchronously(self, execution: WorkflowNodeExecution) -> None: ...
+
     def save_execution_data(self, execution: WorkflowNodeExecution): ...
 
     def get_by_workflow_execution(

api/core/repositories/sqlalchemy_workflow_node_execution_repository.py

@@ -18,6 +18,7 @@ from tenacity import before_sleep_log, retry, retry_if_exception, stop_after_att
 
 from configs import dify_config
 from core.repositories.factory import OrderConfig, WorkflowNodeExecutionRepository
+from core.workflow.node_execution_process_data import preserve_workflow_agent_binding_id
 from extensions.ext_storage import storage
 from graphon.entities import WorkflowNodeExecution
 from graphon.enums import WorkflowNodeExecutionMetadataKey, WorkflowNodeExecutionStatus
@@ -372,6 +373,12 @@ class SQLAlchemyWorkflowNodeExecutionRepository(WorkflowNodeExecutionRepository)
             logger.exception("Failed to save workflow node execution after all retries")
             raise
 
+    @override
+    def save_synchronously(self, execution: WorkflowNodeExecution) -> None:
+        """Persist a caller row before an Agent v2 participant is materialized."""
+
+        self.save(execution)
+
     def _persist_to_database(self, db_model: WorkflowNodeExecutionModel):
         """
         Persist the database model to the database.
@@ -386,6 +393,13 @@ class SQLAlchemyWorkflowNodeExecutionRepository(WorkflowNodeExecutionRepository)
             existing = session.get(WorkflowNodeExecutionModel, db_model.id)
 
             if existing:
+                merged_process_data = preserve_workflow_agent_binding_id(
+                    existing.process_data_dict,
+                    db_model.process_data_dict,
+                )
+                db_model.process_data = (
+                    _deterministic_json_dump(merged_process_data) if merged_process_data is not None else None
+                )
                 # Update existing record by copying all non-private attributes
                 for key, value in db_model.__dict__.items():
                     if not key.startswith("_"):
@@ -442,18 +456,25 @@ class SQLAlchemyWorkflowNodeExecutionRepository(WorkflowNodeExecutionRepository)
             else:
                 db_model.outputs = self._json_encode(domain_model.outputs)
 
-        if domain_model.process_data is not None:
+        process_data = preserve_workflow_agent_binding_id(db_model.process_data_dict, domain_model.process_data)
+        if process_data is not None:
             result = self._truncate_and_upload(
-                domain_model.process_data,
+                process_data,
                 domain_model.id,
                 ExecutionOffLoadType.PROCESS_DATA,
             )
             if result is not None:
-                db_model.process_data = self._json_encode(result.truncated_value)
-                domain_model.set_truncated_process_data(result.truncated_value)
+                truncated_process_data = preserve_workflow_agent_binding_id(
+                    process_data,
+                    result.truncated_value,
+                )
+                if truncated_process_data is None:
+                    raise ValueError("truncated process data is unavailable")
+                db_model.process_data = self._json_encode(truncated_process_data)
+                domain_model.set_truncated_process_data(truncated_process_data)
                 offload_data = _replace_or_append_offload(offload_data, result.offload)
             else:
-                db_model.process_data = self._json_encode(domain_model.process_data)
+                db_model.process_data = self._json_encode(process_data)
 
         db_model.offload_data = offload_data
         with self._session_factory() as session, session.begin():

api/core/workflow/node_execution_process_data.py

@@ -0,0 +1,27 @@
+from collections.abc import Mapping
+from typing import Any
+
+WORKFLOW_AGENT_BINDING_ID_KEY = "workflow_agent_binding_id"
+
+
+def preserve_workflow_agent_binding_id(
+    identity_source: Mapping[str, Any] | None,
+    process_data: Mapping[str, Any] | None,
+) -> dict[str, Any] | None:
+    source_id = (identity_source or {}).get(WORKFLOW_AGENT_BINDING_ID_KEY)
+    target_id = (process_data or {}).get(WORKFLOW_AGENT_BINDING_ID_KEY)
+    for value in (source_id, target_id):
+        if value is not None and not isinstance(value, str):
+            raise ValueError("workflow_agent_binding_id must be a string")
+    if source_id is not None and target_id is not None and source_id != target_id:
+        raise ValueError("workflow_agent_binding_id does not match")
+
+    if process_data is None and source_id is None:
+        return None
+    merged = dict(process_data or {})
+    if source_id is not None:
+        merged[WORKFLOW_AGENT_BINDING_ID_KEY] = source_id
+    return merged
+
+
+__all__ = ["WORKFLOW_AGENT_BINDING_ID_KEY", "preserve_workflow_agent_binding_id"]

api/core/workflow/node_factory.py

@@ -487,7 +487,7 @@ class DifyNodeFactory(NodeFactory):
             from core.workflow.nodes.agent_v2.output_failure_orchestrator import OutputFailureOrchestrator
             from core.workflow.nodes.agent_v2.output_file_rebacker import reback_tool_file_output
             from core.workflow.nodes.agent_v2.output_type_checker import PerOutputTypeChecker
-            from core.workflow.nodes.agent_v2.session_store import WorkflowAgentRuntimeSessionStore
+            from core.workflow.nodes.agent_v2.session_store import WorkflowAgentWorkspaceStore
 
             return {
                 "binding_resolver": WorkflowAgentBindingResolver(),
@@ -512,7 +512,7 @@ class DifyNodeFactory(NodeFactory):
                 # tenant validator resolves ToolFile (canonical) + UploadFile refs.
                 "type_checker": PerOutputTypeChecker(file_validator=AgentOutputFileTenantValidator()),
                 "failure_orchestrator": OutputFailureOrchestrator(),
-                "session_store": WorkflowAgentRuntimeSessionStore(),
+                "session_store": WorkflowAgentWorkspaceStore(),
             }
         return {
             "strategy_resolver": self._agent_strategy_resolver,

api/core/workflow/nodes/agent_v2/agent_node.py

@@ -18,13 +18,10 @@ from clients.agent_backend import (
     AgentBackendRunEventAdapter,
     AgentBackendRunFailedInternalEvent,
     AgentBackendRunSucceededInternalEvent,
-    AgentBackendSessionCleanupPayload,
     AgentBackendStreamError,
     AgentBackendStreamInternalEvent,
     AgentBackendTransportError,
     AgentBackendValidationError,
-    RuntimeLayerSpec,
-    extract_runtime_layer_specs,
 )
 from core.app.entities.app_invoke_entities import DIFY_RUN_CONTEXT_KEY, DifyRunContext
 from core.repositories.human_input_repository import HumanInputFormRepository, HumanInputFormRepositoryImpl
@@ -33,11 +30,12 @@ from core.workflow.nodes.human_input.session_binding import default_session_bind
 from core.workflow.system_variables import SystemVariableKey, get_system_text
 from graphon.entities.pause_reason import HitlRequired, SchedulingPause
 from graphon.enums import BuiltinNodeTypes, WorkflowNodeExecutionMetadataKey, WorkflowNodeExecutionStatus
-from graphon.node_events import NodeEventBase, NodeRunResult, PauseRequestedEvent, StreamCompletedEvent
+from graphon.graph_events import NodeRunPauseRequestedEvent
+from graphon.node_events import NodeEventBase, NodeRunResult, StreamCompletedEvent
 from graphon.nodes.base.node import Node
 from models.agent_config_entities import AgentSoulConfig, WorkflowNodeJobConfig
 from services.agent.prompt_mentions import extract_workflow_node_output_selectors
-from tasks.agent_backend_session_cleanup_task import cleanup_workflow_agent_runtime_session
+from services.agent.workspace_service import AgentWorkspaceNotFoundError
 
 from .ask_human_hitl import AskHumanFormBuildError, build_ask_human_pause_reason
 from .ask_human_resume import build_deferred_tool_results, resolve_ask_human_form
@@ -56,7 +54,7 @@ from .runtime_request_builder import (
     WorkflowAgentRuntimeRequestBuilder,
     WorkflowAgentRuntimeRequestBuildError,
 )
-from .session_store import WorkflowAgentRuntimeSessionStore, WorkflowAgentSessionScope
+from .session_store import WorkflowAgentSessionScope, WorkflowAgentWorkspaceStore
 
 if TYPE_CHECKING:
     from graphon.entities import GraphInitParams
@@ -68,7 +66,7 @@ logger = logging.getLogger(__name__)
 # Stage 4 §5+§7: the terminal events that `_consume_event_stream` may return.
 # Stream + started events are filtered out before we yield; transport errors
 # are surfaced as a separate StreamCompletedEvent in the second tuple slot.
-_TerminalAgentBackendEvent = (
+type _TerminalAgentBackendEvent = (
     AgentBackendRunSucceededInternalEvent
     | AgentBackendRunFailedInternalEvent
     | AgentBackendRunCancelledInternalEvent
@@ -93,7 +91,7 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
         output_adapter: WorkflowAgentOutputAdapter,
         type_checker: PerOutputTypeChecker,
         failure_orchestrator: OutputFailureOrchestrator,
-        session_store: WorkflowAgentRuntimeSessionStore | None = None,
+        session_store: WorkflowAgentWorkspaceStore,
     ) -> None:
         super().__init__(
             node_id=node_id,
@@ -130,7 +128,34 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
         return reason
 
     @override
-    def _run(self) -> Generator[NodeEventBase, None, None]:
+    def _run(self) -> Generator[NodeEventBase | NodeRunPauseRequestedEvent, None, None]:
+        inputs: dict[str, Any] = {}
+        process_data: dict[str, Any] = {}
+        metadata: dict[str, Any] = {
+            "agent_backend": {
+                "status": "not_started",
+            }
+        }
+        try:
+            yield from self._run_inner(inputs=inputs, process_data=process_data, metadata=metadata)
+        except Exception as error:
+            if not process_data:
+                raise
+            yield self._failure_event(
+                inputs=inputs,
+                process_data=process_data,
+                metadata=metadata,
+                error=str(error),
+                error_type="agent_workflow_node_runtime_error",
+            )
+
+    def _run_inner(
+        self,
+        *,
+        inputs: dict[str, Any],
+        process_data: dict[str, Any],
+        metadata: dict[str, Any],
+    ) -> Generator[NodeEventBase | NodeRunPauseRequestedEvent, None, None]:
         dify_ctx = DifyRunContext.model_validate(self.require_run_context_value(DIFY_RUN_CONTEXT_KEY))
         workflow_id = self.graph_init_params.workflow_id
         workflow_run_id = get_system_text(
@@ -143,21 +168,24 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
             self.graph_runtime_state.variable_pool,
             SystemVariableKey.CONVERSATION_ID,
         )
-        inputs: dict[str, Any] = {}
-        process_data: dict[str, Any] = {}
-        metadata: dict[str, Any] = {
-            "agent_backend": {
-                "status": "not_started",
-            }
-        }
 
         # ──── Setup: resolve binding once + extract declared outputs for stage 4 checks ────
         try:
+            existing_scope = self._session_store.load_existing_node_execution_scope(
+                tenant_id=dify_ctx.tenant_id,
+                app_id=dify_ctx.app_id,
+                workflow_id=workflow_id,
+                workflow_run_id=workflow_run_id,
+                node_id=self._node_id,
+                node_execution_id=self.execution_id,
+            )
             bundle = self._binding_resolver.resolve(
                 tenant_id=dify_ctx.tenant_id,
                 app_id=dify_ctx.app_id,
                 workflow_id=workflow_id,
                 node_id=self._node_id,
+                binding_id=existing_scope.workflow_agent_binding_id if existing_scope is not None else None,
+                snapshot_id=existing_scope.agent_config_snapshot_id if existing_scope is not None else None,
             )
         except WorkflowAgentBindingError as error:
             yield self._failure_event(
@@ -168,20 +196,31 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
                 error_type=error.error_code,
             )
             return
+        except AgentWorkspaceNotFoundError as error:
+            yield self._failure_event(
+                inputs=inputs,
+                process_data=process_data,
+                metadata=metadata,
+                error=str(error),
+                error_type="agent_workflow_node_runtime_error",
+            )
+            return
 
-        process_data = {
-            "agent_id": bundle.agent.id,
-            "agent_config_snapshot_id": bundle.snapshot.id,
-            "binding_id": bundle.binding.id,
-        }
-        session_scope = WorkflowAgentSessionScope(
+        process_data.update(
+            {
+                "agent_id": bundle.agent.id,
+                "agent_config_snapshot_id": bundle.snapshot.id,
+                "workflow_agent_binding_id": bundle.binding.id,
+            }
+        )
+        session_scope = existing_scope or WorkflowAgentSessionScope(
             tenant_id=dify_ctx.tenant_id,
             app_id=dify_ctx.app_id,
             workflow_id=workflow_id,
             workflow_run_id=workflow_run_id,
             node_id=self._node_id,
-            node_execution_id=self.id,
-            binding_id=bundle.binding.id,
+            node_execution_id=self.execution_id,
+            workflow_agent_binding_id=bundle.binding.id,
             agent_id=bundle.agent.id,
             agent_config_snapshot_id=bundle.snapshot.id,
         )
@@ -200,47 +239,53 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
         # the second Agent run as deferred_tool_results; if it is somehow still
         # waiting, re-emit the same pause defensively.
         deferred_tool_results = None
-        if self._session_store is not None:
-            stored_session = self._session_store.load_active_session(session_scope)
-            if stored_session is not None and stored_session.pending_form_id is not None:
-                resume_outcome = resolve_ask_human_form(
-                    form_id=stored_session.pending_form_id,
-                    tenant_id=dify_ctx.tenant_id,
-                    node_id=self._node_id,
+        stored_session = self._session_store.load_or_create_node_execution_session(
+            session_scope,
+            home_snapshot_id=bundle.snapshot.home_snapshot_id,
+        )
+        if stored_session.pending_form_id is not None:
+            resume_outcome = resolve_ask_human_form(
+                form_id=stored_session.pending_form_id,
+                tenant_id=dify_ctx.tenant_id,
+                node_id=self._node_id,
+            )
+            if resume_outcome is not None and resume_outcome.repause is not None:
+                yield self._pause_event(
+                    reason=resume_outcome.repause,
+                    inputs=inputs,
+                    process_data=process_data,
+                    metadata=metadata,
+                )
+                return
+            if (
+                resume_outcome is not None
+                and resume_outcome.deferred_result is not None
+                and stored_session.pending_tool_call_id is not None
+            ):
+                deferred_tool_results = build_deferred_tool_results(
+                    tool_call_id=stored_session.pending_tool_call_id,
+                    result=resume_outcome.deferred_result,
                 )
-                if resume_outcome is not None and resume_outcome.repause is not None:
-                    yield PauseRequestedEvent(reason=self._to_graph_pause_reason(resume_outcome.repause))
-                    return
-                if (
-                    resume_outcome is not None
-                    and resume_outcome.deferred_result is not None
-                    and stored_session.pending_tool_call_id is not None
-                ):
-                    deferred_tool_results = build_deferred_tool_results(
-                        tool_call_id=stored_session.pending_tool_call_id,
-                        result=resume_outcome.deferred_result,
-                    )
 
         # ──── Retry loop (Stage 4 §7) ────
         attempt = 0
         while True:
             try:
-                session_snapshot = None
-                if self._session_store is not None:
-                    session_snapshot = self._session_store.load_active_snapshot(session_scope)
                 runtime_request = self._runtime_request_builder.build(
                     WorkflowAgentRuntimeBuildContext(
                         dify_context=dify_ctx,
                         workflow_id=workflow_id,
                         workflow_run_id=workflow_run_id,
                         node_id=self._node_id,
-                        node_execution_id=self.id,
+                        node_execution_id=self.execution_id,
                         variable_pool=self.graph_runtime_state.variable_pool,
                         binding=bundle.binding,
                         agent=bundle.agent,
                         snapshot=bundle.snapshot,
+                        binding_id=stored_session.binding_id,
+                        backend_binding_ref=stored_session.backend_binding_ref,
                         attempt=attempt,
-                        session_snapshot=session_snapshot,
+                        session_snapshot=stored_session.session_snapshot,
                         deferred_tool_results=deferred_tool_results,
                     )
                 )
@@ -266,8 +311,9 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
             # Capture inputs only from the first attempt so retry doesn't churn the
             # node's "inputs" payload that ends up in the workflow detail view.
             if attempt == 0:
-                inputs = {"agent_backend_request": runtime_request.redacted_request}
-            metadata = dict(runtime_request.metadata)
+                inputs["agent_backend_request"] = runtime_request.redacted_request
+            metadata.clear()
+            metadata.update(runtime_request.metadata)
             metadata["attempt"] = attempt
 
             try:
@@ -288,7 +334,12 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
                 "status": create_response.status,
             }
 
-            terminal_event, exhausted = self._consume_event_stream(create_response.run_id, metadata)
+            terminal_event, exhausted = self._consume_event_stream(
+                create_response.run_id,
+                inputs=inputs,
+                process_data=process_data,
+                metadata=metadata,
+            )
             if exhausted is not None:
                 # Streaming error / unexpected end — surface immediately without
                 # retrying because the failure is transport-level.
@@ -349,29 +400,23 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
                     )
                 self._save_session_snapshot(
                     session_scope=session_scope,
-                    backend_run_id=terminal_event.run_id,
+                    binding_id=stored_session.binding_id,
                     snapshot=terminal_event.session_snapshot,
-                    runtime_layer_specs=extract_runtime_layer_specs(runtime_request.request.composition),
                     metadata=metadata,
                     pending_form_id=pending_form_id,
                     pending_tool_call_id=pending_tool_call_id,
                 )
-                yield PauseRequestedEvent(reason=self._to_graph_pause_reason(pause_reason))
-                return
-
-            # Non-success terminal (failed / cancelled) skips per-output
-            # post-processing — the backend itself already failed. We also retire
-            # the local ACTIVE session row so a workflow loop back into the same
-            # Agent node cannot resume from a stale snapshot. The failed agent
-            # backend layers (suspended per ``on_exit``) are left for agent
-            # backend's own GC; this row will no longer be picked up by the
-            # workflow-terminal cleanup layer.
-            if not isinstance(terminal_event, AgentBackendRunSucceededInternalEvent):
-                self._mark_session_cleaned_on_failure(
-                    session_scope=session_scope,
-                    backend_run_id=terminal_event.run_id,
+                yield self._pause_event(
+                    reason=pause_reason,
+                    inputs=inputs,
+                    process_data=process_data,
                     metadata=metadata,
                 )
+                return
+
+            # A failed attempt does not retire the product-owned Binding. The
+            # Workflow Run terminal lifecycle event owns that transition.
+            if not isinstance(terminal_event, AgentBackendRunSucceededInternalEvent):
                 yield StreamCompletedEvent(
                     node_run_result=self._output_adapter.build_failure_result(
                         event=terminal_event,
@@ -384,9 +429,8 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
 
             self._save_session_snapshot(
                 session_scope=session_scope,
-                backend_run_id=terminal_event.run_id,
+                binding_id=stored_session.binding_id,
                 snapshot=terminal_event.session_snapshot,
-                runtime_layer_specs=extract_runtime_layer_specs(runtime_request.request.composition),
                 metadata=metadata,
             )
 
@@ -458,6 +502,9 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
     def _consume_event_stream(
         self,
         run_id: str,
+        *,
+        inputs: dict[str, Any],
+        process_data: dict[str, Any],
         metadata: dict[str, Any],
     ) -> tuple[
         _TerminalAgentBackendEvent | None,
@@ -507,8 +554,8 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
                         return internal_event, None
                     self._cancel_backend_run(run_id, reason="unexpected_event")
                     return None, self._failure_event(
-                        inputs={},
-                        process_data={},
+                        inputs=inputs,
+                        process_data=process_data,
                         metadata=metadata,
                         error=f"Unexpected internal event type {internal_event.type!r}",
                         error_type="agent_backend_stream_error",
@@ -516,8 +563,8 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
         except AgentBackendError as error:
             self._cancel_backend_run(run_id, reason=self._stream_stop_reason())
             return None, self._failure_event(
-                inputs={},
-                process_data={},
+                inputs=inputs,
+                process_data=process_data,
                 metadata=metadata,
                 error=str(error),
                 error_type=self._agent_backend_error_type(error),
@@ -525,8 +572,8 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
         except Exception as error:
             self._cancel_backend_run(run_id, reason=self._stream_stop_reason())
             return None, self._failure_event(
-                inputs={},
-                process_data={},
+                inputs=inputs,
+                process_data=process_data,
                 metadata=metadata,
                 error=str(error),
                 error_type="agent_backend_stream_error",
@@ -596,21 +643,17 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
         self,
         *,
         session_scope: WorkflowAgentSessionScope,
-        backend_run_id: str,
+        binding_id: str,
         snapshot: CompositorSessionSnapshot | None,
-        runtime_layer_specs: list[RuntimeLayerSpec],
         metadata: dict[str, Any],
         pending_form_id: str | None = None,
         pending_tool_call_id: str | None = None,
     ) -> None:
-        if self._session_store is None:
-            return
         try:
             self._session_store.save_active_snapshot(
                 scope=session_scope,
-                backend_run_id=backend_run_id,
+                binding_id=binding_id,
                 snapshot=snapshot,
-                runtime_layer_specs=runtime_layer_specs,
                 pending_form_id=pending_form_id,
                 pending_tool_call_id=pending_tool_call_id,
             )
@@ -619,88 +662,18 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
             metadata["agent_backend"] = agent_backend
         except Exception:
             logger.warning(
-                "Failed to persist workflow Agent runtime session snapshot: "
-                "tenant_id=%s workflow_run_id=%s node_id=%s binding_id=%s agent_id=%s backend_run_id=%s",
+                "Failed to persist workflow Agent Binding session snapshot: "
+                "tenant_id=%s workflow_run_id=%s node_id=%s binding_id=%s agent_id=%s",
                 session_scope.tenant_id,
                 session_scope.workflow_run_id,
                 session_scope.node_id,
-                session_scope.binding_id,
+                session_scope.workflow_agent_binding_id,
                 session_scope.agent_id,
-                backend_run_id,
                 exc_info=True,
             )
             agent_backend = dict(metadata.get("agent_backend") or {})
             agent_backend["session_snapshot_persisted"] = False
-            agent_backend["session_snapshot_persist_error"] = "workflow_agent_runtime_session_store_error"
-            metadata["agent_backend"] = agent_backend
-
-    def _mark_session_cleaned_on_failure(
-        self,
-        *,
-        session_scope: WorkflowAgentSessionScope,
-        backend_run_id: str,
-        metadata: dict[str, Any],
-    ) -> None:
-        if self._session_store is None:
-            return
-        stored_session = self._session_store.load_active_session(session_scope)
-        try:
-            if stored_session is not None and stored_session.runtime_layer_specs:
-                payload = AgentBackendSessionCleanupPayload(
-                    session_snapshot=stored_session.session_snapshot,
-                    runtime_layer_specs=stored_session.runtime_layer_specs,
-                    idempotency_key=(
-                        f"{session_scope.tenant_id}:{session_scope.workflow_run_id}:{session_scope.node_id}:"
-                        f"{session_scope.binding_id}:workflow-agent-failure-cleanup:"
-                        f"{stored_session.backend_run_id or 'no-stored-run'}:{backend_run_id}"
-                    ),
-                    metadata={
-                        "tenant_id": session_scope.tenant_id,
-                        "app_id": session_scope.app_id,
-                        "workflow_id": session_scope.workflow_id,
-                        "workflow_run_id": session_scope.workflow_run_id,
-                        "node_id": session_scope.node_id,
-                        "node_execution_id": session_scope.node_execution_id,
-                        "binding_id": session_scope.binding_id,
-                        "agent_id": session_scope.agent_id,
-                        "agent_config_snapshot_id": session_scope.agent_config_snapshot_id,
-                        "previous_agent_backend_run_id": stored_session.backend_run_id,
-                        "failed_agent_backend_run_id": backend_run_id,
-                    },
-                )
-                cleanup_workflow_agent_runtime_session.delay(payload.model_dump(mode="json"))
-        except Exception:
-            logger.warning(
-                "Failed to enqueue workflow Agent backend cleanup on agent run failure: "
-                "tenant_id=%s workflow_run_id=%s node_id=%s binding_id=%s agent_id=%s backend_run_id=%s",
-                session_scope.tenant_id,
-                session_scope.workflow_run_id,
-                session_scope.node_id,
-                session_scope.binding_id,
-                session_scope.agent_id,
-                backend_run_id,
-                exc_info=True,
-            )
-        try:
-            self._session_store.mark_cleaned(scope=session_scope, backend_run_id=backend_run_id)
-            agent_backend = dict(metadata.get("agent_backend") or {})
-            agent_backend["session_snapshot_cleaned_on_failure"] = True
-            metadata["agent_backend"] = agent_backend
-        except Exception:
-            logger.warning(
-                "Failed to mark workflow Agent runtime session cleaned on agent run failure: "
-                "tenant_id=%s workflow_run_id=%s node_id=%s binding_id=%s agent_id=%s backend_run_id=%s",
-                session_scope.tenant_id,
-                session_scope.workflow_run_id,
-                session_scope.node_id,
-                session_scope.binding_id,
-                session_scope.agent_id,
-                backend_run_id,
-                exc_info=True,
-            )
-            agent_backend = dict(metadata.get("agent_backend") or {})
-            agent_backend["session_snapshot_cleaned_on_failure"] = False
-            agent_backend["session_snapshot_cleanup_error"] = "workflow_agent_runtime_session_store_error"
+            agent_backend["session_snapshot_persist_error"] = "workflow_agent_workspace_store_error"
             metadata["agent_backend"] = agent_backend
 
     @staticmethod
@@ -742,6 +715,27 @@ class DifyAgentNode(Node[DifyAgentNodeData]):
             )
         )
 
+    def _pause_event(
+        self,
+        *,
+        reason: HumanInputRequired | SchedulingPause,
+        inputs: dict[str, Any],
+        process_data: dict[str, Any],
+        metadata: dict[str, Any],
+    ) -> NodeRunPauseRequestedEvent:
+        return NodeRunPauseRequestedEvent(
+            id=self.execution_id,
+            node_id=self._node_id,
+            node_type=self.node_type,
+            node_run_result=NodeRunResult(
+                status=WorkflowNodeExecutionStatus.PAUSED,
+                inputs=inputs,
+                process_data=process_data,
+                metadata={WorkflowNodeExecutionMetadataKey.AGENT_LOG: metadata},
+            ),
+            reason=self._to_graph_pause_reason(reason),
+        )
+
     @staticmethod
     def _agent_backend_error_type(error: AgentBackendError) -> str:
         if isinstance(error, AgentBackendValidationError):

api/core/workflow/nodes/agent_v2/binding_resolver.py

@@ -41,18 +41,27 @@ class WorkflowAgentBindingResolver:
         app_id: str,
         workflow_id: str,
         node_id: str,
+        binding_id: str | None = None,
+        snapshot_id: str | None = None,
     ) -> WorkflowAgentBindingBundle:
-        with session_factory.create_session() as session:
-            binding = session.scalar(
-                select(WorkflowAgentNodeBinding)
-                .where(
-                    WorkflowAgentNodeBinding.tenant_id == tenant_id,
-                    WorkflowAgentNodeBinding.app_id == app_id,
-                    WorkflowAgentNodeBinding.workflow_id == workflow_id,
-                    WorkflowAgentNodeBinding.node_id == node_id,
-                )
-                .limit(1)
+        """Resolve the current binding, optionally at a generation pinned by an existing execution."""
+
+        if (binding_id is None) != (snapshot_id is None):
+            raise WorkflowAgentBindingError(
+                "agent_binding_generation_invalid",
+                "Workflow Agent binding and config snapshot must be pinned together.",
             )
+
+        with session_factory.create_session() as session:
+            binding_stmt = select(WorkflowAgentNodeBinding).where(
+                WorkflowAgentNodeBinding.tenant_id == tenant_id,
+                WorkflowAgentNodeBinding.app_id == app_id,
+                WorkflowAgentNodeBinding.workflow_id == workflow_id,
+                WorkflowAgentNodeBinding.node_id == node_id,
+            )
+            if binding_id is not None:
+                binding_stmt = binding_stmt.where(WorkflowAgentNodeBinding.id == binding_id)
+            binding = session.scalar(binding_stmt.limit(1))
             if binding is None:
                 raise WorkflowAgentBindingError(
                     "agent_binding_not_found",
@@ -77,12 +86,16 @@ class WorkflowAgentBindingResolver:
                     f"Agent {binding.agent_id} is not available or has not been published.",
                 )
 
-            snapshot_id = (
-                agent.active_config_snapshot_id
-                if binding.binding_type == WorkflowAgentBindingType.ROSTER_AGENT
-                else binding.current_snapshot_id
+            effective_snapshot_id = (
+                (
+                    agent.active_config_snapshot_id
+                    if binding.binding_type == WorkflowAgentBindingType.ROSTER_AGENT
+                    else binding.current_snapshot_id
+                )
+                if snapshot_id is None
+                else snapshot_id
             )
-            if snapshot_id is None:
+            if effective_snapshot_id is None:
                 raise WorkflowAgentBindingError(
                     "agent_config_snapshot_not_found",
                     "Workflow Agent binding has no current config snapshot.",
@@ -93,14 +106,14 @@ class WorkflowAgentBindingResolver:
                 .where(
                     AgentConfigSnapshot.tenant_id == tenant_id,
                     AgentConfigSnapshot.agent_id == agent.id,
-                    AgentConfigSnapshot.id == snapshot_id,
+                    AgentConfigSnapshot.id == effective_snapshot_id,
                 )
                 .limit(1)
             )
             if snapshot is None:
                 raise WorkflowAgentBindingError(
                     "agent_config_snapshot_not_found",
-                    f"Agent config snapshot {snapshot_id} not found.",
+                    f"Agent config snapshot {effective_snapshot_id} not found.",
                 )
 
             session.expunge(binding)

api/core/workflow/nodes/agent_v2/runtime_request_builder.py

@@ -33,7 +33,6 @@ from dify_agent.layers.shell import (
     DifyShellCliToolConfig,
     DifyShellEnvVarConfig,
     DifyShellLayerConfig,
-    DifyShellSandboxConfig,
     DifyShellSecretRefConfig,
 )
 from dify_agent.protocol import CreateRunRequest, DeferredToolResultsPayload
@@ -136,6 +135,8 @@ class WorkflowAgentRuntimeBuildContext:
     binding: WorkflowAgentNodeBinding
     agent: Agent
     snapshot: AgentConfigSnapshot
+    binding_id: str
+    backend_binding_ref: str
     # Stage 4 §7 / D-4: 0 for the first run, then incremented per retry. Drives the
     # idempotency key so the backend treats each retry as a fresh request.
     attempt: int = 0
@@ -251,6 +252,7 @@ class WorkflowAgentRuntimeRequestBuilder:
                     agent_mode=self._agent_backend_agent_mode(context.dify_context.invoke_from),
                     invoke_from=cast(DifyExecutionContextInvokeFrom, context.dify_context.invoke_from.value),
                 ),
+                backend_binding_ref=context.backend_binding_ref,
                 agent_soul_prompt=soul_prompt or None,
                 workflow_node_job_prompt=workflow_job_prompt,
                 user_prompt=user_prompt,
@@ -739,7 +741,6 @@ class WorkflowAgentRuntimeRequestBuilder:
 
 def build_shell_layer_config(agent_soul: AgentSoulConfig) -> DifyShellLayerConfig:
     """Map Agent Soul shell-adjacent fields into the Agent backend shell config."""
-    sandbox_config = _plain_mapping(agent_soul.sandbox.config)
     return DifyShellLayerConfig(
         cli_tools=[
             tool
@@ -750,12 +751,6 @@ def build_shell_layer_config(agent_soul: AgentSoulConfig) -> DifyShellLayerConfi
         secret_refs=[
             secret for secret in (_shell_secret_ref(item) for item in agent_soul.env.secret_refs) if secret is not None
         ],
-        sandbox=DifyShellSandboxConfig(
-            provider=agent_soul.sandbox.provider,
-            config=sandbox_config,
-        )
-        if agent_soul.sandbox.provider or sandbox_config
-        else None,
     )
 
 

api/core/workflow/nodes/agent_v2/session_cleanup_layer.py

@@ -1,126 +0,0 @@
-"""Workflow terminal layer that retires Agent backend sessions asynchronously."""
-
-from __future__ import annotations
-
-import logging
-from typing import override
-
-from clients.agent_backend import AgentBackendSessionCleanupPayload
-from core.workflow.system_variables import SystemVariableKey, get_system_text
-from graphon.graph_engine.layers import GraphEngineLayer
-from graphon.graph_events import (
-    GraphEngineEvent,
-    GraphRunAbortedEvent,
-    GraphRunFailedEvent,
-    GraphRunPartialSucceededEvent,
-    GraphRunSucceededEvent,
-)
-from tasks.agent_backend_session_cleanup_task import cleanup_workflow_agent_runtime_session
-
-from .session_store import StoredWorkflowAgentSession, WorkflowAgentRuntimeSessionStore
-
-logger = logging.getLogger(__name__)
-
-
-class WorkflowAgentSessionCleanupLayer(GraphEngineLayer):
-    """Retire workflow-owned Agent runtime sessions when the workflow ends.
-
-    Workflow termination is a product-lifecycle boundary: once the run reaches a
-    terminal graph event, the local session row must no longer be resumable. The
-    actual Agent backend cleanup is therefore dispatched asynchronously with the
-    persisted snapshot/specs payload, while the local row is marked CLEANED
-    immediately afterwards regardless of enqueue outcome.
-    """
-
-    _TERMINAL_EVENTS = (
-        GraphRunSucceededEvent,
-        GraphRunPartialSucceededEvent,
-        GraphRunFailedEvent,
-        GraphRunAbortedEvent,
-    )
-
-    def __init__(self, *, session_store: WorkflowAgentRuntimeSessionStore) -> None:
-        super().__init__()
-        self._session_store = session_store
-
-    @override
-    def on_graph_start(self) -> None:
-        return
-
-    @override
-    def on_event(self, event: GraphEngineEvent) -> None:
-        if not isinstance(event, self._TERMINAL_EVENTS):
-            return
-        workflow_run_id = get_system_text(
-            self.graph_runtime_state.variable_pool,
-            SystemVariableKey.WORKFLOW_EXECUTION_ID,
-        )
-        if not workflow_run_id:
-            logger.warning("Skipping workflow Agent session cleanup: workflow_run_id is missing.")
-            return
-
-        for stored_session in self._session_store.list_active_sessions(workflow_run_id=workflow_run_id):
-            self._cleanup_session(stored_session)
-
-    @override
-    def on_graph_end(self, error: Exception | None) -> None:
-        return
-
-    def _cleanup_session(self, stored_session: StoredWorkflowAgentSession) -> None:
-        scope = stored_session.scope
-        try:
-            if stored_session.runtime_layer_specs:
-                payload = AgentBackendSessionCleanupPayload(
-                    session_snapshot=stored_session.session_snapshot,
-                    runtime_layer_specs=stored_session.runtime_layer_specs,
-                    idempotency_key=f"{scope.workflow_run_id}:{scope.node_id}:{scope.binding_id}:agent-session-cleanup",
-                    metadata={
-                        "tenant_id": scope.tenant_id,
-                        "app_id": scope.app_id,
-                        "workflow_id": scope.workflow_id,
-                        "workflow_run_id": scope.workflow_run_id,
-                        "node_id": scope.node_id,
-                        "node_execution_id": scope.node_execution_id,
-                        "binding_id": scope.binding_id,
-                        "agent_id": scope.agent_id,
-                        "agent_config_snapshot_id": scope.agent_config_snapshot_id,
-                        "previous_agent_backend_run_id": stored_session.backend_run_id,
-                    },
-                )
-                cleanup_workflow_agent_runtime_session.delay(payload.model_dump(mode="json"))
-            else:
-                logger.warning(
-                    "Skipping workflow Agent backend cleanup enqueue: no runtime_layer_specs persisted. "
-                    "workflow_run_id=%s node_id=%s agent_id=%s",
-                    scope.workflow_run_id,
-                    scope.node_id,
-                    scope.agent_id,
-                )
-        except Exception:
-            logger.warning(
-                "Failed to enqueue workflow Agent backend cleanup: "
-                "workflow_run_id=%s node_id=%s agent_id=%s previous_run_id=%s",
-                scope.workflow_run_id,
-                scope.node_id,
-                scope.agent_id,
-                stored_session.backend_run_id,
-                exc_info=True,
-            )
-        finally:
-            try:
-                self._session_store.mark_cleaned(scope=scope, backend_run_id=stored_session.backend_run_id)
-            except Exception:
-                logger.warning(
-                    "Failed to retire workflow Agent runtime session after cleanup enqueue: "
-                    "workflow_run_id=%s node_id=%s agent_id=%s previous_run_id=%s",
-                    scope.workflow_run_id,
-                    scope.node_id,
-                    scope.agent_id,
-                    stored_session.backend_run_id,
-                    exc_info=True,
-                )
-
-
-def build_workflow_agent_session_cleanup_layer() -> WorkflowAgentSessionCleanupLayer:
-    """Wire the cleanup layer with the standard workflow-owned session store."""
-    return WorkflowAgentSessionCleanupLayer(session_store=WorkflowAgentRuntimeSessionStore())

api/core/workflow/nodes/agent_v2/session_store.py

@@ -1,31 +1,32 @@
+"""Workflow Agent participant persistence keyed by node execution."""
+
 from __future__ import annotations
 
-from dataclasses import dataclass, field
+import json
+import time
+from dataclasses import dataclass
 
 from agenton.compositor import CompositorSessionSnapshot
-from dify_agent.protocol import RuntimeLayerSpec
-from pydantic import TypeAdapter
 from sqlalchemy import select
+from sqlalchemy.orm import Session
 
 from core.db.session_factory import session_factory
-from libs.datetime_utils import naive_utc_now
 from models.agent import (
-    AgentRuntimeSessionOwnerType,
-    WorkflowAgentRuntimeSession,
-    WorkflowAgentRuntimeSessionStatus,
+    AgentConfigVersionKind,
+    AgentWorkingResourceStatus,
+    AgentWorkspace,
+    AgentWorkspaceBinding,
+    AgentWorkspaceOwnerType,
+)
+from models.workflow import WorkflowNodeExecutionModel
+from services.agent.workspace_service import (
+    AgentWorkspaceNotFoundError,
+    AgentWorkspaceService,
+    WorkspaceOwnerScope,
 )
 
-_SPECS_ADAPTER: TypeAdapter[list[RuntimeLayerSpec]] = TypeAdapter(list[RuntimeLayerSpec])
-
-
-def _serialize_specs(specs: list[RuntimeLayerSpec]) -> str:
-    return _SPECS_ADAPTER.dump_json(specs).decode()
-
-
-def _deserialize_specs(value: str | None) -> list[RuntimeLayerSpec]:
-    if not value:
-        return []
-    return _SPECS_ADAPTER.validate_json(value)
+_CALLER_VISIBILITY_ATTEMPTS = 60
+_CALLER_VISIBILITY_INTERVAL_SECONDS = 0.05
 
 
 @dataclass(frozen=True, slots=True)
@@ -36,171 +37,251 @@ class WorkflowAgentSessionScope:
     workflow_run_id: str | None
     node_id: str
     node_execution_id: str
-    binding_id: str
+    workflow_agent_binding_id: str
     agent_id: str
     agent_config_snapshot_id: str
 
+    @property
+    def workspace_owner(self) -> WorkspaceOwnerScope:
+        return WorkspaceOwnerScope(
+            tenant_id=self.tenant_id,
+            app_id=self.app_id,
+            owner_type=AgentWorkspaceOwnerType.WORKFLOW_RUN,
+            owner_id=self.workflow_run_id or self.node_execution_id,
+            owner_scope_key=f"{self.node_id}:{self.workflow_agent_binding_id}",
+        )
+
 
 @dataclass(frozen=True, slots=True)
 class StoredWorkflowAgentSession:
     scope: WorkflowAgentSessionScope
-    session_snapshot: CompositorSessionSnapshot
-    backend_run_id: str | None
-    runtime_layer_specs: list[RuntimeLayerSpec] = field(default_factory=list)
-    # ENG-637: set while the session is paused on a dify.ask_human deferred call.
+    binding_id: str
+    workspace_id: str
+    backend_binding_ref: str
+    session_snapshot: CompositorSessionSnapshot | None
     pending_form_id: str | None = None
     pending_tool_call_id: str | None = None
 
 
-class WorkflowAgentRuntimeSessionStore:
-    """Stores Agent backend session snapshots for workflow Agent node re-entry."""
+class WorkflowAgentWorkspaceStore:
+    """Load or create the participant named by a node execution caller row."""
 
-    def load_active_snapshot(self, scope: WorkflowAgentSessionScope) -> CompositorSessionSnapshot | None:
-        stored = self.load_active_session(scope)
-        return stored.session_snapshot if stored is not None else None
-
-    def load_active_session(self, scope: WorkflowAgentSessionScope) -> StoredWorkflowAgentSession | None:
-        """Load the active session row including any pending ask_human correlation."""
-        if scope.workflow_run_id is None:
-            return None
+    def load_existing_node_execution_scope(
+        self,
+        *,
+        tenant_id: str,
+        app_id: str,
+        workflow_id: str,
+        workflow_run_id: str | None,
+        node_id: str,
+        node_execution_id: str,
+    ) -> WorkflowAgentSessionScope | None:
+        """Return the generation pinned by an existing node execution participant."""
 
         with session_factory.create_session() as session:
-            row = session.scalar(
-                select(WorkflowAgentRuntimeSession).where(
-                    WorkflowAgentRuntimeSession.tenant_id == scope.tenant_id,
-                    WorkflowAgentRuntimeSession.workflow_run_id == scope.workflow_run_id,
-                    WorkflowAgentRuntimeSession.node_id == scope.node_id,
-                    WorkflowAgentRuntimeSession.binding_id == scope.binding_id,
-                    WorkflowAgentRuntimeSession.agent_id == scope.agent_id,
-                    WorkflowAgentRuntimeSession.status == WorkflowAgentRuntimeSessionStatus.ACTIVE,
-                )
+            execution = self._load_execution_by_identity(
+                session=session,
+                tenant_id=tenant_id,
+                app_id=app_id,
+                workflow_id=workflow_id,
+                workflow_run_id=workflow_run_id,
+                node_id=node_id,
+                node_execution_id=node_execution_id,
             )
-            if row is None:
+            binding_id = execution.agent_workspace_binding_id
+            if binding_id is None:
                 return None
-            return StoredWorkflowAgentSession(
-                scope=scope,
-                session_snapshot=CompositorSessionSnapshot.model_validate_json(row.session_snapshot),
-                backend_run_id=row.backend_run_id,
-                runtime_layer_specs=_deserialize_specs(row.composition_layer_specs),
-                pending_form_id=row.pending_form_id,
-                pending_tool_call_id=row.pending_tool_call_id,
+            process_data = execution.process_data_dict
+            if not isinstance(process_data, dict):
+                raise AgentWorkspaceNotFoundError("Workflow node execution caller identity is invalid")
+            workflow_agent_binding_id = process_data.get("workflow_agent_binding_id")
+            if not isinstance(workflow_agent_binding_id, str):
+                raise AgentWorkspaceNotFoundError("Workflow node execution caller identity is missing")
+            owner_scope = WorkspaceOwnerScope(
+                tenant_id=tenant_id,
+                app_id=app_id,
+                owner_type=AgentWorkspaceOwnerType.WORKFLOW_RUN,
+                owner_id=workflow_run_id or node_execution_id,
+                owner_scope_key=f"{node_id}:{workflow_agent_binding_id}",
+            )
+            binding = AgentWorkspaceService.get_active_binding(
+                session=session,
+                tenant_id=tenant_id,
+                binding_id=binding_id,
+                expected_owner_scope=owner_scope,
+            )
+            if binding is None or binding.agent_config_version_kind != AgentConfigVersionKind.SNAPSHOT:
+                raise AgentWorkspaceNotFoundError("Workflow node participant Binding is unavailable")
+            return WorkflowAgentSessionScope(
+                tenant_id=tenant_id,
+                app_id=app_id,
+                workflow_id=workflow_id,
+                workflow_run_id=workflow_run_id,
+                node_id=node_id,
+                node_execution_id=node_execution_id,
+                workflow_agent_binding_id=workflow_agent_binding_id,
+                agent_id=binding.agent_id,
+                agent_config_snapshot_id=binding.agent_config_version_id,
             )
 
-    def list_active_sessions(self, *, workflow_run_id: str) -> list[StoredWorkflowAgentSession]:
+    def load_or_create_node_execution_session(
+        self, scope: WorkflowAgentSessionScope, *, home_snapshot_id: str
+    ) -> StoredWorkflowAgentSession:
         with session_factory.create_session() as session:
-            rows = session.scalars(
-                select(WorkflowAgentRuntimeSession).where(
-                    WorkflowAgentRuntimeSession.workflow_run_id == workflow_run_id,
-                    WorkflowAgentRuntimeSession.status == WorkflowAgentRuntimeSessionStatus.ACTIVE,
+            execution = self._load_execution(session=session, scope=scope)
+            process_data = execution.process_data_dict
+            if process_data is None:
+                process_data = {}
+            if not isinstance(process_data, dict):
+                raise AgentWorkspaceNotFoundError("Workflow node execution caller identity is invalid")
+            stored_workflow_binding_id = process_data.get("workflow_agent_binding_id")
+            if stored_workflow_binding_id is not None and stored_workflow_binding_id != scope.workflow_agent_binding_id:
+                raise AgentWorkspaceNotFoundError("Workflow node execution caller identity does not match")
+
+            binding_id = execution.agent_workspace_binding_id
+            if binding_id is None:
+                binding = AgentWorkspaceService.create_binding(
+                    session=session,
+                    scope=scope.workspace_owner,
+                    agent_id=scope.agent_id,
+                    base_home_snapshot_id=home_snapshot_id,
+                    agent_config_version_id=scope.agent_config_snapshot_id,
+                    agent_config_version_kind=AgentConfigVersionKind.SNAPSHOT,
                 )
-            ).all()
-            return [
-                StoredWorkflowAgentSession(
-                    scope=WorkflowAgentSessionScope(
-                        tenant_id=row.tenant_id,
-                        app_id=row.app_id,
-                        # These columns are nullable on the unified runtime-session
-                        # table (workflow_run ⊕ conversation owner), but are always
-                        # populated for a workflow-owned row; coerce for the typed scope.
-                        workflow_id=row.workflow_id or "",
-                        workflow_run_id=row.workflow_run_id,
-                        node_id=row.node_id or "",
-                        node_execution_id=row.node_execution_id or "",
-                        binding_id=row.binding_id or "",
-                        agent_id=row.agent_id,
-                        agent_config_snapshot_id=row.agent_config_snapshot_id or "",
-                    ),
-                    session_snapshot=CompositorSessionSnapshot.model_validate_json(row.session_snapshot),
-                    backend_run_id=row.backend_run_id,
-                    runtime_layer_specs=_deserialize_specs(row.composition_layer_specs),
+                execution.agent_workspace_binding_id = binding.id
+                execution.process_data = json.dumps(
+                    {
+                        **process_data,
+                        "workflow_agent_binding_id": scope.workflow_agent_binding_id,
+                    },
+                    ensure_ascii=False,
                 )
-                for row in rows
-            ]
+                session.commit()
+            else:
+                if stored_workflow_binding_id is None:
+                    raise AgentWorkspaceNotFoundError("Workflow node execution caller identity is missing")
+                resolved_binding = AgentWorkspaceService.get_active_binding(
+                    session=session,
+                    tenant_id=scope.tenant_id,
+                    binding_id=binding_id,
+                    expected_owner_scope=scope.workspace_owner,
+                )
+                if resolved_binding is None or resolved_binding.agent_id != scope.agent_id:
+                    raise AgentWorkspaceNotFoundError("Workflow node participant Binding is unavailable")
+                binding = resolved_binding
+                AgentWorkspaceService.validate_binding_generation(
+                    binding,
+                    base_home_snapshot_id=home_snapshot_id,
+                    agent_config_version_id=scope.agent_config_snapshot_id,
+                    agent_config_version_kind=AgentConfigVersionKind.SNAPSHOT,
+                )
+            return self._stored(scope, binding)
 
     def save_active_snapshot(
         self,
         *,
         scope: WorkflowAgentSessionScope,
-        backend_run_id: str,
+        binding_id: str,
         snapshot: CompositorSessionSnapshot | None,
-        runtime_layer_specs: list[RuntimeLayerSpec],
         pending_form_id: str | None = None,
         pending_tool_call_id: str | None = None,
     ) -> None:
-        if scope.workflow_run_id is None or snapshot is None:
+        if snapshot is None:
             return
+        AgentWorkspaceService.save_binding_session_snapshot(
+            tenant_id=scope.tenant_id,
+            binding_id=binding_id,
+            session_snapshot=snapshot.model_dump_json(),
+            pending_form_id=pending_form_id,
+            pending_tool_call_id=pending_tool_call_id,
+        )
 
-        snapshot_json = snapshot.model_dump_json()
-        specs_json = _serialize_specs(runtime_layer_specs)
+    def retire_workflow_run(self, *, tenant_id: str, app_id: str, workflow_run_id: str) -> list[str]:
+        """Retire active Workspaces, commit, and return active or already-retired IDs for collection."""
+
+        retired: list[str] = []
         with session_factory.create_session() as session:
-            row = session.scalar(
-                select(WorkflowAgentRuntimeSession).where(
-                    WorkflowAgentRuntimeSession.tenant_id == scope.tenant_id,
-                    WorkflowAgentRuntimeSession.workflow_run_id == scope.workflow_run_id,
-                    WorkflowAgentRuntimeSession.node_id == scope.node_id,
-                    WorkflowAgentRuntimeSession.binding_id == scope.binding_id,
-                    WorkflowAgentRuntimeSession.agent_id == scope.agent_id,
+            workspaces = session.scalars(
+                select(AgentWorkspace).where(
+                    AgentWorkspace.tenant_id == tenant_id,
+                    AgentWorkspace.app_id == app_id,
+                    AgentWorkspace.owner_type == AgentWorkspaceOwnerType.WORKFLOW_RUN,
+                    AgentWorkspace.owner_id == workflow_run_id,
+                    AgentWorkspace.status.in_((AgentWorkingResourceStatus.ACTIVE, AgentWorkingResourceStatus.RETIRED)),
                 )
-            )
-            if row is None:
-                row = WorkflowAgentRuntimeSession(
-                    tenant_id=scope.tenant_id,
-                    app_id=scope.app_id,
-                    owner_type=AgentRuntimeSessionOwnerType.WORKFLOW_RUN,
-                    workflow_id=scope.workflow_id,
-                    workflow_run_id=scope.workflow_run_id,
-                    node_id=scope.node_id,
-                    node_execution_id=scope.node_execution_id,
-                    binding_id=scope.binding_id,
-                    agent_id=scope.agent_id,
-                    agent_config_snapshot_id=scope.agent_config_snapshot_id,
-                    backend_run_id=backend_run_id,
-                    session_snapshot=snapshot_json,
-                    composition_layer_specs=specs_json,
-                    status=WorkflowAgentRuntimeSessionStatus.ACTIVE,
-                    pending_form_id=pending_form_id,
-                    pending_tool_call_id=pending_tool_call_id,
+            ).all()
+            for workspace in workspaces:
+                if workspace.status == AgentWorkingResourceStatus.RETIRED:
+                    retired.append(workspace.id)
+                    continue
+                workspace_id = AgentWorkspaceService.retire_workspace(
+                    session=session,
+                    tenant_id=tenant_id,
+                    workspace_id=workspace.id,
                 )
-                session.add(row)
-            else:
-                row.node_execution_id = scope.node_execution_id
-                row.agent_config_snapshot_id = scope.agent_config_snapshot_id
-                row.backend_run_id = backend_run_id
-                row.session_snapshot = snapshot_json
-                row.composition_layer_specs = specs_json
-                row.status = WorkflowAgentRuntimeSessionStatus.ACTIVE
-                row.cleaned_at = None
-                # Set (or clear, when omitted) the ask_human pause correlation.
-                row.pending_form_id = pending_form_id
-                row.pending_tool_call_id = pending_tool_call_id
+                if workspace_id is not None:
+                    retired.append(workspace_id)
             session.commit()
+        return retired
 
-    def mark_cleaned(self, *, scope: WorkflowAgentSessionScope, backend_run_id: str | None = None) -> None:
-        if scope.workflow_run_id is None:
-            return
+    @staticmethod
+    def _load_execution(*, session: Session, scope: WorkflowAgentSessionScope) -> WorkflowNodeExecutionModel:
+        return WorkflowAgentWorkspaceStore._load_execution_by_identity(
+            session=session,
+            tenant_id=scope.tenant_id,
+            app_id=scope.app_id,
+            workflow_id=scope.workflow_id,
+            workflow_run_id=scope.workflow_run_id,
+            node_id=scope.node_id,
+            node_execution_id=scope.node_execution_id,
+        )
 
-        with session_factory.create_session() as session:
-            row = session.scalar(
-                select(WorkflowAgentRuntimeSession).where(
-                    WorkflowAgentRuntimeSession.tenant_id == scope.tenant_id,
-                    WorkflowAgentRuntimeSession.workflow_run_id == scope.workflow_run_id,
-                    WorkflowAgentRuntimeSession.node_id == scope.node_id,
-                    WorkflowAgentRuntimeSession.binding_id == scope.binding_id,
-                    WorkflowAgentRuntimeSession.agent_id == scope.agent_id,
-                    WorkflowAgentRuntimeSession.status == WorkflowAgentRuntimeSessionStatus.ACTIVE,
-                )
-            )
-            if row is None:
-                return
-            if backend_run_id is not None:
-                row.backend_run_id = backend_run_id
-            row.status = WorkflowAgentRuntimeSessionStatus.CLEANED
-            row.cleaned_at = naive_utc_now()
-            session.commit()
+    @staticmethod
+    def _load_execution_by_identity(
+        *,
+        session: Session,
+        tenant_id: str,
+        app_id: str,
+        workflow_id: str,
+        workflow_run_id: str | None,
+        node_id: str,
+        node_execution_id: str,
+    ) -> WorkflowNodeExecutionModel:
+        """Wait briefly for the already-emitted node-start event to persist its caller row."""
+
+        stmt = select(WorkflowNodeExecutionModel).where(
+            WorkflowNodeExecutionModel.id == node_execution_id,
+            WorkflowNodeExecutionModel.tenant_id == tenant_id,
+            WorkflowNodeExecutionModel.app_id == app_id,
+            WorkflowNodeExecutionModel.workflow_id == workflow_id,
+            WorkflowNodeExecutionModel.node_id == node_id,
+            WorkflowNodeExecutionModel.workflow_run_id == workflow_run_id,
+        )
+        for attempt in range(_CALLER_VISIBILITY_ATTEMPTS):
+            execution = session.scalar(stmt)
+            if execution is not None:
+                return execution
+            if attempt < _CALLER_VISIBILITY_ATTEMPTS - 1:
+                time.sleep(_CALLER_VISIBILITY_INTERVAL_SECONDS)
+
+        raise AgentWorkspaceNotFoundError("Workflow node execution caller is unavailable")
+
+    @staticmethod
+    def _stored(scope: WorkflowAgentSessionScope, binding: AgentWorkspaceBinding) -> StoredWorkflowAgentSession:
+        snapshot = (
+            CompositorSessionSnapshot.model_validate_json(binding.session_snapshot)
+            if binding.session_snapshot
+            else None
+        )
+        return StoredWorkflowAgentSession(
+            scope=scope,
+            binding_id=binding.id,
+            workspace_id=binding.workspace_id,
+            backend_binding_ref=binding.backend_binding_ref,
+            session_snapshot=snapshot,
+            pending_form_id=binding.pending_form_id,
+            pending_tool_call_id=binding.pending_tool_call_id,
+        )
 
 
-__all__ = [
-    "StoredWorkflowAgentSession",
-    "WorkflowAgentRuntimeSessionStore",
-    "WorkflowAgentSessionScope",
-]
+__all__ = ["StoredWorkflowAgentSession", "WorkflowAgentSessionScope", "WorkflowAgentWorkspaceStore"]

api/core/workflow/nodes/agent_v2/workspace_retirement_layer.py

@@ -0,0 +1,89 @@
+"""Retire Workflow Agent Workspaces when the Workflow Run terminates."""
+
+from __future__ import annotations
+
+import logging
+from typing import override
+
+from core.app.entities.app_invoke_entities import DifyRunContext
+from core.workflow.nodes.agent_v2.session_store import WorkflowAgentWorkspaceStore
+from core.workflow.system_variables import SystemVariableKey, get_system_text
+from graphon.graph_engine.layers import GraphEngineLayer
+from graphon.graph_events import (
+    GraphEngineEvent,
+    GraphRunAbortedEvent,
+    GraphRunFailedEvent,
+    GraphRunPartialSucceededEvent,
+    GraphRunSucceededEvent,
+)
+from tasks.collect_agent_resources_task import enqueue_agent_resource_collection
+
+logger = logging.getLogger(__name__)
+
+
+class WorkflowAgentWorkspaceRetirementLayer(GraphEngineLayer):
+    """Synchronously retire run Workspaces, then enqueue physical collection."""
+
+    _TERMINAL_EVENTS = (
+        GraphRunSucceededEvent,
+        GraphRunPartialSucceededEvent,
+        GraphRunFailedEvent,
+        GraphRunAbortedEvent,
+    )
+
+    def __init__(
+        self,
+        *,
+        dify_run_context: DifyRunContext,
+    ) -> None:
+        super().__init__()
+        self._dify_run_context = dify_run_context
+
+    @override
+    def on_graph_start(self) -> None:
+        return
+
+    @override
+    def on_event(self, event: GraphEngineEvent) -> None:
+        if not isinstance(event, self._TERMINAL_EVENTS):
+            return
+        workflow_run_id = get_system_text(
+            self.graph_runtime_state.variable_pool,
+            SystemVariableKey.WORKFLOW_EXECUTION_ID,
+        )
+        if not workflow_run_id:
+            logger.warning("Skipping Workflow Agent Workspace retirement: workflow_run_id is missing")
+            return
+        try:
+            workspace_ids = WorkflowAgentWorkspaceStore().retire_workflow_run(
+                tenant_id=self._dify_run_context.tenant_id,
+                app_id=self._dify_run_context.app_id,
+                workflow_run_id=workflow_run_id,
+            )
+        except Exception:
+            logger.exception(
+                "Failed to retire Workflow Agent Workspaces",
+                extra={
+                    "tenant_id": self._dify_run_context.tenant_id,
+                    "app_id": self._dify_run_context.app_id,
+                    "workflow_run_id": workflow_run_id,
+                },
+            )
+            return
+        enqueue_agent_resource_collection(
+            tenant_id=self._dify_run_context.tenant_id,
+            workspace_ids=workspace_ids,
+        )
+
+    @override
+    def on_graph_end(self, error: Exception | None) -> None:
+        return
+
+
+def build_workflow_agent_workspace_retirement_layer(
+    *, dify_run_context: DifyRunContext
+) -> WorkflowAgentWorkspaceRetirementLayer:
+    return WorkflowAgentWorkspaceRetirementLayer(dify_run_context=dify_run_context)
+
+
+__all__ = ["WorkflowAgentWorkspaceRetirementLayer", "build_workflow_agent_workspace_retirement_layer"]

api/extensions/ext_celery.py

@@ -166,6 +166,7 @@ def init_app(app: DifyApp) -> Celery:
 
     imports = [
         "tasks.async_workflow_tasks",  # trigger workers
+        "tasks.collect_agent_resources_task",  # retired Agent resource collection
         "tasks.trigger_processing_tasks",  # async trigger processing
         "tasks.generate_summary_index_task",  # summary index generation
         "tasks.regenerate_summary_index_task",  # summary index regeneration

api/extensions/logstore/repositories/logstore_workflow_node_execution_repository.py

@@ -277,6 +277,12 @@ class LogstoreWorkflowNodeExecutionRepository(WorkflowNodeExecutionRepository):
                 logger.exception("Failed to dual-write node execution to SQL database: id=%s", execution.id)
                 # Don't raise - LogStore write succeeded, SQL is just a backup
 
+    @override
+    def save_synchronously(self, execution: WorkflowNodeExecution) -> None:
+        """Create the SQL caller row required by Agent v2 participant ownership."""
+
+        self.sql_repository.save_synchronously(execution)
+
     @override
     def save_execution_data(self, execution: WorkflowNodeExecution) -> None:
         """

api/fields/conversation_fields.py

@@ -34,7 +34,7 @@ class _SessionResponseSource[SourceT]:
         self._session = session
 
     def __getattr__(self, name: str) -> object:
-        return getattr(self._source, name)  # noqa: no-new-getattr response adapter delegates model fields
+        return getattr(self._source, name)  # guard-ignore: no-new-getattr -- delegates model fields
 
 
 class _FeedbackResponseSource(_SessionResponseSource[MessageFeedback]):

api/fields/dataset_fields.py

@@ -227,7 +227,7 @@ class DatasetDetailResponseSource:
         return self.dataset.get_total_available_documents(session=self.session)
 
     def __getattr__(self, name: str) -> Any:
-        return getattr(self.dataset, name)  # noqa: no-new-getattr response adapter delegates model fields
+        return getattr(self.dataset, name)  # guard-ignore: no-new-getattr -- delegates model fields
 
 
 def dataset_detail_response_source(dataset: Any, *, session: Session) -> DatasetDetailResponseSource:

api/fields/document_fields.py

@@ -90,7 +90,7 @@ class DocumentWithSession:
         return self.document.get_doc_metadata_details(session=self.session)
 
     def __getattr__(self, name: str) -> Any:
-        return getattr(self.document, name)  # noqa: no-new-getattr response adapter delegates model fields
+        return getattr(self.document, name)  # guard-ignore: no-new-getattr -- delegates model fields
 
 
 def document_response(document: Document, *, session: Session) -> DocumentResponse:

api/migrations/versions/2026_07_21_2251-2f39536b3feb_add_agent_home_snapshot_ledger.py

@@ -0,0 +1,64 @@
+"""add agent home snapshot ledger
+
+Revision ID: 2f39536b3feb
+Revises: 6f5a9c2d8e1b
+Create Date: 2026-07-21 22:51:07.268658
+
+"""
+from alembic import op
+import models as models
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision = '2f39536b3feb'
+down_revision = '6f5a9c2d8e1b'
+branch_labels = None
+depends_on = None
+
+
+def upgrade():
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.create_table('agent_home_snapshots',
+    sa.Column('id', models.types.StringUUID(), nullable=False),
+    sa.Column('tenant_id', models.types.StringUUID(), nullable=False),
+    sa.Column('agent_id', models.types.StringUUID(), nullable=False),
+    sa.Column('snapshot_ref', sa.String(length=255), nullable=False),
+    sa.Column('created_at', sa.DateTime(), server_default=sa.text('CURRENT_TIMESTAMP'), nullable=False),
+    sa.PrimaryKeyConstraint('id', name='agent_home_snapshot_pkey')
+    )
+    with op.batch_alter_table('agent_home_snapshots', schema=None) as batch_op:
+        batch_op.create_index('agent_home_snapshot_tenant_agent_idx', ['tenant_id', 'agent_id'], unique=False)
+
+    with op.batch_alter_table('agent_config_drafts', schema=None) as batch_op:
+        batch_op.add_column(sa.Column('home_snapshot_id', models.types.StringUUID(), nullable=False))
+
+    with op.batch_alter_table('agent_config_snapshots', schema=None) as batch_op:
+        batch_op.add_column(sa.Column('home_snapshot_id', models.types.StringUUID(), nullable=False))
+
+    with op.batch_alter_table('agent_runtime_sessions', schema=None) as batch_op:
+        batch_op.add_column(sa.Column('home_snapshot_id', models.types.StringUUID(), nullable=False))
+        batch_op.drop_index(batch_op.f('agent_runtime_session_conversation_scope_unique'), postgresql_where='(conversation_id IS NOT NULL)')
+        batch_op.create_index('agent_runtime_session_conversation_scope_unique', ['tenant_id', 'conversation_id', 'agent_id', 'agent_config_snapshot_id', 'home_snapshot_id'], unique=True, postgresql_where=sa.text('conversation_id IS NOT NULL'))
+
+    # ### end Alembic commands ###
+
+
+def downgrade():
+    # ### commands auto generated by Alembic - please adjust! ###
+    with op.batch_alter_table('agent_runtime_sessions', schema=None) as batch_op:
+        batch_op.drop_index('agent_runtime_session_conversation_scope_unique', postgresql_where=sa.text('conversation_id IS NOT NULL'))
+        batch_op.create_index(batch_op.f('agent_runtime_session_conversation_scope_unique'), ['tenant_id', 'conversation_id', 'agent_id', 'agent_config_snapshot_id'], unique=True, postgresql_where='(conversation_id IS NOT NULL)')
+        batch_op.drop_column('home_snapshot_id')
+
+    with op.batch_alter_table('agent_config_snapshots', schema=None) as batch_op:
+        batch_op.drop_column('home_snapshot_id')
+
+    with op.batch_alter_table('agent_config_drafts', schema=None) as batch_op:
+        batch_op.drop_column('home_snapshot_id')
+
+    with op.batch_alter_table('agent_home_snapshots', schema=None) as batch_op:
+        batch_op.drop_index('agent_home_snapshot_tenant_agent_idx')
+
+    op.drop_table('agent_home_snapshots')
+    # ### end Alembic commands ###

api/migrations/versions/2026_07_23_0203-f6e4c5686857_replace_agent_runtime_sessions_with_.py

@@ -0,0 +1,153 @@
+"""replace agent runtime sessions with workspaces and bindings
+
+Revision ID: f6e4c5686857
+Revises: 2f39536b3feb
+Create Date: 2026-07-23 02:03:05.641638
+
+"""
+from alembic import op
+import models as models
+import sqlalchemy as sa
+from sqlalchemy.dialects import postgresql
+
+# revision identifiers, used by Alembic.
+revision = 'f6e4c5686857'
+down_revision = '2f39536b3feb'
+branch_labels = None
+depends_on = None
+
+
+def upgrade():
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.create_table('agent_workspace_bindings',
+    sa.Column('tenant_id', models.types.StringUUID(), nullable=False),
+    sa.Column('app_id', models.types.StringUUID(), nullable=False),
+    sa.Column('workspace_id', models.types.StringUUID(), nullable=False),
+    sa.Column('agent_id', models.types.StringUUID(), nullable=False),
+    sa.Column('base_home_snapshot_id', models.types.StringUUID(), nullable=False),
+    sa.Column('agent_config_version_id', models.types.StringUUID(), nullable=False),
+    sa.Column('agent_config_version_kind', sa.String(length=32), nullable=False),
+    sa.Column('backend_binding_ref', sa.String(length=255), nullable=False),
+    sa.Column('session_snapshot', models.types.LongText(), nullable=True),
+    sa.Column('status', sa.String(length=32), server_default='active', nullable=False),
+    sa.Column('retired_at', sa.DateTime(), nullable=True),
+    sa.Column('pending_form_id', models.types.StringUUID(), nullable=True),
+    sa.Column('pending_tool_call_id', sa.String(length=255), nullable=True),
+    sa.Column('id', models.types.StringUUID(), nullable=False),
+    sa.Column('created_at', sa.DateTime(), server_default=sa.text('CURRENT_TIMESTAMP'), nullable=False),
+    sa.Column('updated_at', sa.DateTime(), server_default=sa.text('CURRENT_TIMESTAMP'), nullable=False),
+    sa.PrimaryKeyConstraint('id', name='agent_workspace_binding_pkey')
+    )
+    with op.batch_alter_table('agent_workspace_bindings', schema=None) as batch_op:
+        batch_op.create_index('agent_workspace_binding_agent_status_idx', ['tenant_id', 'agent_id', 'status'], unique=False)
+        batch_op.create_index('agent_workspace_binding_status_retired_idx', ['status', 'retired_at'], unique=False)
+        batch_op.create_index('agent_workspace_binding_workspace_status_idx', ['tenant_id', 'workspace_id', 'status'], unique=False)
+
+    op.create_table('agent_workspaces',
+    sa.Column('tenant_id', models.types.StringUUID(), nullable=False),
+    sa.Column('app_id', models.types.StringUUID(), nullable=False),
+    sa.Column('owner_type', sa.String(length=32), nullable=False),
+    sa.Column('owner_id', models.types.StringUUID(), nullable=False),
+    sa.Column('owner_scope_key', sa.String(length=255), nullable=False),
+    sa.Column('backend_workspace_ref', sa.String(length=255), nullable=False),
+    sa.Column('status', sa.String(length=32), server_default='active', nullable=False),
+    sa.Column('active_guard', sa.SmallInteger(), server_default='1', nullable=True),
+    sa.Column('retired_at', sa.DateTime(), nullable=True),
+    sa.Column('id', models.types.StringUUID(), nullable=False),
+    sa.Column('created_at', sa.DateTime(), server_default=sa.text('CURRENT_TIMESTAMP'), nullable=False),
+    sa.Column('updated_at', sa.DateTime(), server_default=sa.text('CURRENT_TIMESTAMP'), nullable=False),
+    sa.PrimaryKeyConstraint('id', name='agent_workspace_pkey')
+    )
+    with op.batch_alter_table('agent_workspaces', schema=None) as batch_op:
+        batch_op.create_index('agent_workspace_owner_active_unique', ['tenant_id', 'owner_type', 'owner_id', 'owner_scope_key', 'active_guard'], unique=True)
+        batch_op.create_index('agent_workspace_status_retired_idx', ['status', 'retired_at'], unique=False)
+        batch_op.create_index('agent_workspace_tenant_app_status_idx', ['tenant_id', 'app_id', 'status'], unique=False)
+        batch_op.create_index('agent_workspace_tenant_status_idx', ['tenant_id', 'status'], unique=False)
+
+    with op.batch_alter_table('agent_runtime_sessions', schema=None) as batch_op:
+        batch_op.drop_index(batch_op.f('agent_runtime_session_backend_run_idx'))
+        batch_op.drop_index(batch_op.f('agent_runtime_session_conversation_lookup_idx'))
+        batch_op.drop_index(batch_op.f('agent_runtime_session_conversation_scope_unique'), postgresql_where='(conversation_id IS NOT NULL)')
+        batch_op.drop_index(batch_op.f('agent_runtime_session_workflow_lookup_idx'))
+        batch_op.drop_index(batch_op.f('agent_runtime_session_workflow_scope_unique'), postgresql_where='(workflow_run_id IS NOT NULL)')
+
+    op.drop_table('agent_runtime_sessions')
+    with op.batch_alter_table('agent_home_snapshots', schema=None) as batch_op:
+        batch_op.add_column(sa.Column('status', sa.String(length=32), server_default='active', nullable=False))
+        batch_op.add_column(sa.Column('retired_at', sa.DateTime(), nullable=True))
+        batch_op.create_index('agent_home_snapshot_status_retired_idx', ['status', 'retired_at'], unique=False)
+
+    with op.batch_alter_table('conversations', schema=None) as batch_op:
+        batch_op.add_column(sa.Column('agent_workspace_binding_id', models.types.StringUUID(), nullable=True))
+
+    with op.batch_alter_table('agent_config_drafts', schema=None) as batch_op:
+        batch_op.add_column(sa.Column('agent_workspace_binding_id', models.types.StringUUID(), nullable=True))
+
+    with op.batch_alter_table('workflow_node_executions', schema=None) as batch_op:
+        batch_op.add_column(sa.Column('agent_workspace_binding_id', models.types.StringUUID(), nullable=True))
+
+    # ### end Alembic commands ###
+
+
+def downgrade():
+    # ### commands auto generated by Alembic - please adjust! ###
+    with op.batch_alter_table('workflow_node_executions', schema=None) as batch_op:
+        batch_op.drop_column('agent_workspace_binding_id')
+
+    with op.batch_alter_table('agent_config_drafts', schema=None) as batch_op:
+        batch_op.drop_column('agent_workspace_binding_id')
+
+    with op.batch_alter_table('conversations', schema=None) as batch_op:
+        batch_op.drop_column('agent_workspace_binding_id')
+
+    with op.batch_alter_table('agent_home_snapshots', schema=None) as batch_op:
+        batch_op.drop_index('agent_home_snapshot_status_retired_idx')
+        batch_op.drop_column('retired_at')
+        batch_op.drop_column('status')
+
+    op.create_table('agent_runtime_sessions',
+    sa.Column('id', sa.UUID(), server_default=sa.text('uuidv7()'), autoincrement=False, nullable=False),
+    sa.Column('tenant_id', sa.UUID(), autoincrement=False, nullable=False),
+    sa.Column('app_id', sa.UUID(), autoincrement=False, nullable=False),
+    sa.Column('owner_type', sa.VARCHAR(length=32), autoincrement=False, nullable=False),
+    sa.Column('agent_id', sa.UUID(), autoincrement=False, nullable=False),
+    sa.Column('backend_run_id', sa.VARCHAR(length=255), autoincrement=False, nullable=True),
+    sa.Column('session_snapshot', sa.TEXT(), autoincrement=False, nullable=False),
+    sa.Column('workflow_id', sa.UUID(), autoincrement=False, nullable=True),
+    sa.Column('workflow_run_id', sa.UUID(), autoincrement=False, nullable=True),
+    sa.Column('node_id', sa.VARCHAR(length=255), autoincrement=False, nullable=True),
+    sa.Column('node_execution_id', sa.VARCHAR(length=255), autoincrement=False, nullable=True),
+    sa.Column('binding_id', sa.UUID(), autoincrement=False, nullable=True),
+    sa.Column('agent_config_snapshot_id', sa.UUID(), autoincrement=False, nullable=True),
+    sa.Column('composition_layer_specs', sa.TEXT(), autoincrement=False, nullable=False),
+    sa.Column('conversation_id', sa.UUID(), autoincrement=False, nullable=True),
+    sa.Column('status', sa.VARCHAR(length=32), server_default=sa.text("'active'::character varying"), autoincrement=False, nullable=False),
+    sa.Column('cleaned_at', postgresql.TIMESTAMP(), autoincrement=False, nullable=True),
+    sa.Column('created_at', postgresql.TIMESTAMP(), server_default=sa.text('CURRENT_TIMESTAMP'), autoincrement=False, nullable=False),
+    sa.Column('updated_at', postgresql.TIMESTAMP(), server_default=sa.text('CURRENT_TIMESTAMP'), autoincrement=False, nullable=False),
+    sa.Column('pending_form_id', sa.UUID(), autoincrement=False, nullable=True),
+    sa.Column('pending_tool_call_id', sa.VARCHAR(length=255), autoincrement=False, nullable=True),
+    sa.Column('home_snapshot_id', sa.UUID(), autoincrement=False, nullable=False),
+    sa.PrimaryKeyConstraint('id', name=op.f('agent_runtime_session_pkey'))
+    )
+    with op.batch_alter_table('agent_runtime_sessions', schema=None) as batch_op:
+        batch_op.create_index(batch_op.f('agent_runtime_session_workflow_scope_unique'), ['tenant_id', 'workflow_run_id', 'node_id', 'binding_id', 'agent_id'], unique=True, postgresql_where='(workflow_run_id IS NOT NULL)')
+        batch_op.create_index(batch_op.f('agent_runtime_session_workflow_lookup_idx'), ['tenant_id', 'workflow_run_id', 'node_id', 'status'], unique=False)
+        batch_op.create_index(batch_op.f('agent_runtime_session_conversation_scope_unique'), ['tenant_id', 'conversation_id', 'agent_id', 'agent_config_snapshot_id', 'home_snapshot_id'], unique=True, postgresql_where='(conversation_id IS NOT NULL)')
+        batch_op.create_index(batch_op.f('agent_runtime_session_conversation_lookup_idx'), ['tenant_id', 'conversation_id', 'status'], unique=False)
+        batch_op.create_index(batch_op.f('agent_runtime_session_backend_run_idx'), ['backend_run_id'], unique=False)
+
+    with op.batch_alter_table('agent_workspaces', schema=None) as batch_op:
+        batch_op.drop_index('agent_workspace_tenant_status_idx')
+        batch_op.drop_index('agent_workspace_tenant_app_status_idx')
+        batch_op.drop_index('agent_workspace_status_retired_idx')
+        batch_op.drop_index('agent_workspace_owner_active_unique')
+
+    op.drop_table('agent_workspaces')
+    with op.batch_alter_table('agent_workspace_bindings', schema=None) as batch_op:
+        batch_op.drop_index('agent_workspace_binding_workspace_status_idx')
+        batch_op.drop_index('agent_workspace_binding_status_retired_idx')
+        batch_op.drop_index('agent_workspace_binding_agent_status_idx')
+
+    op.drop_table('agent_workspace_bindings')
+    # ### end Alembic commands ###

api/models/__init__.py

@@ -15,21 +15,22 @@ from .agent import (
     AgentConfigRevision,
     AgentConfigRevisionOperation,
     AgentConfigSnapshot,
+    AgentConfigVersionKind,
     AgentDebugConversation,
     AgentDriveFile,
     AgentDriveFileKind,
+    AgentHomeSnapshot,
     AgentIconType,
     AgentKind,
-    AgentRuntimeSession,
-    AgentRuntimeSessionOwnerType,
-    AgentRuntimeSessionStatus,
     AgentScope,
     AgentSource,
     AgentStatus,
+    AgentWorkingResourceStatus,
+    AgentWorkspace,
+    AgentWorkspaceBinding,
+    AgentWorkspaceOwnerType,
     WorkflowAgentBindingType,
     WorkflowAgentNodeBinding,
-    WorkflowAgentRuntimeSession,
-    WorkflowAgentRuntimeSessionStatus,
 )
 from .api_based_extension import APIBasedExtension, APIBasedExtensionPoint
 from .comment import (
@@ -164,17 +165,20 @@ __all__ = [
     "AgentConfigRevision",
     "AgentConfigRevisionOperation",
     "AgentConfigSnapshot",
+    "AgentConfigVersionKind",
     "AgentDebugConversation",
     "AgentDriveFile",
     "AgentDriveFileKind",
+    "AgentHomeSnapshot",
     "AgentIconType",
     "AgentKind",
-    "AgentRuntimeSession",
-    "AgentRuntimeSessionOwnerType",
-    "AgentRuntimeSessionStatus",
     "AgentScope",
     "AgentSource",
     "AgentStatus",
+    "AgentWorkingResourceStatus",
+    "AgentWorkspace",
+    "AgentWorkspaceBinding",
+    "AgentWorkspaceOwnerType",
     "ApiRequest",
     "ApiToken",
     "ApiToolProvider",
@@ -271,8 +275,6 @@ __all__ = [
     "Workflow",
     "WorkflowAgentBindingType",
     "WorkflowAgentNodeBinding",
-    "WorkflowAgentRuntimeSession",
-    "WorkflowAgentRuntimeSessionStatus",
     "WorkflowAppLog",
     "WorkflowAppLogCreatedFrom",
     "WorkflowArchiveLog",

api/models/agent.py

@@ -116,35 +116,29 @@ class WorkflowAgentBindingType(StrEnum):
     INLINE_AGENT = "inline_agent"
 
 
-class AgentRuntimeSessionStatus(StrEnum):
-    """Lifecycle state of an Agent backend session snapshot.
+class AgentWorkingResourceStatus(StrEnum):
+    """Product lifecycle state for a persistent working-environment resource."""
 
-    Owner-agnostic: applies both to workflow Agent Node runs (owner =
-    workflow_run) and to Agent App conversations (owner = conversation).
-    """
-
-    # Snapshot can be reused by a later Agent run in the same session.
     ACTIVE = "active"
-    # Snapshot has been retired and must not be submitted to Agent backend again.
-    CLEANED = "cleaned"
+    RETIRED = "retired"
 
 
-class AgentRuntimeSessionOwnerType(StrEnum):
-    """Which product surface owns an Agent runtime session row."""
+class AgentWorkspaceOwnerType(StrEnum):
+    """Product scope that owns a Workspace."""
 
-    # Owned by one workflow Agent Node execution scope.
     WORKFLOW_RUN = "workflow_run"
-    # Owned by one Agent App conversation (multi-turn chat).
     CONVERSATION = "conversation"
+    BUILD_DRAFT = "build_draft"
 
 
-# Back-compat alias: the workflow lifecycle code (shipped in PR #36724) imports
-# the old name. Kept so unifying the table does not churn that path.
-WorkflowAgentRuntimeSessionStatus = AgentRuntimeSessionStatus
+class AgentConfigVersionKind(StrEnum):
+    SNAPSHOT = "snapshot"
+    DRAFT = "draft"
+    BUILD_DRAFT = "build_draft"
 
 
 class Agent(DefaultFieldsMixin, Base):
-    """Workspace-scoped Agent identity used by Agent Roster and workflow-only agents."""
+    """Agent Soul and source lineage; ``AgentWorkspaceBinding.id`` identifies each materialized participant."""
 
     __tablename__ = "agents"
     __table_args__ = (
@@ -221,14 +215,42 @@ class Agent(DefaultFieldsMixin, Base):
     archived_at: Mapped[datetime | None] = mapped_column(DateTime, nullable=True)
 
 
-class AgentDebugConversation(DefaultFieldsMixin, Base):
-    """Per-account, per-draft console debug conversation for an Agent App.
+class AgentHomeSnapshot(Base):
+    """Append-only mapping from one Agent-owned Home identity to its backend ref.
 
-    Agent App preview state must be isolated by editor account. The Agent row is
-    shared by everyone in the workspace, so this table owns the user-specific
-    conversation pointers used by console debug chat. ``draft`` is the Preview
-    conversation and ``debug_build`` is the Build conversation; they must never
-    share persisted messages or runtime sessions.
+    Product tables reference ``id``. ``snapshot_ref`` remains an opaque
+    deployment-specific handle and is only consumed at Dify Agent boundaries.
+    Snapshot bytes and ``snapshot_ref`` are immutable. Lifecycle metadata can
+    transition ACTIVE -> RETIRED; successful physical collection deletes row.
+    """
+
+    __tablename__ = "agent_home_snapshots"
+    __table_args__ = (
+        sa.PrimaryKeyConstraint("id", name="agent_home_snapshot_pkey"),
+        Index("agent_home_snapshot_tenant_agent_idx", "tenant_id", "agent_id"),
+        Index("agent_home_snapshot_status_retired_idx", "status", "retired_at"),
+    )
+
+    id: Mapped[str] = mapped_column(StringUUID, default=lambda: str(uuidv7()))
+    tenant_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    agent_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    snapshot_ref: Mapped[str] = mapped_column(String(255), nullable=False)
+    status: Mapped[AgentWorkingResourceStatus] = mapped_column(
+        EnumText(AgentWorkingResourceStatus, length=32),
+        nullable=False,
+        default=AgentWorkingResourceStatus.ACTIVE,
+        server_default=AgentWorkingResourceStatus.ACTIVE.value,
+    )
+    retired_at: Mapped[datetime | None] = mapped_column(DateTime, nullable=True)
+    created_at: Mapped[datetime] = mapped_column(DateTime, nullable=False, server_default=func.current_timestamp())
+
+
+class AgentDebugConversation(DefaultFieldsMixin, Base):
+    """Current console Conversation pointer for one account and draft surface.
+
+    This row owns no Binding or runtime. A Preview Conversation holds its
+    CONVERSATION Binding pointer, while a DEBUG_BUILD AgentConfigDraft holds its
+    BUILD_DRAFT Binding pointer.
     """
 
     __tablename__ = "agent_debug_conversations"
@@ -259,7 +281,11 @@ class AgentDebugConversation(DefaultFieldsMixin, Base):
 
 
 class AgentConfigDraft(DefaultFieldsMixin, Base):
-    """Editable Agent Soul draft separated from immutable published snapshots."""
+    """Editable Agent Soul draft separated from immutable published snapshots.
+
+    A DEBUG_BUILD draft owns its materialized participant through
+    ``agent_workspace_binding_id``. Normal drafts leave that pointer unset.
+    """
 
     __tablename__ = "agent_config_drafts"
     __table_args__ = (
@@ -281,6 +307,8 @@ class AgentConfigDraft(DefaultFieldsMixin, Base):
     account_id: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
     draft_owner_key: Mapped[str] = mapped_column(String(255), nullable=False, default="")
     base_snapshot_id: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
+    home_snapshot_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    agent_workspace_binding_id: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
     config_snapshot: Mapped[Any] = mapped_column(JSONModelColumn(AgentSoulConfig), nullable=False)
     created_by: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
     updated_by: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
@@ -316,6 +344,7 @@ class AgentConfigSnapshot(DefaultFieldsMixin, Base):
     agent_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
     version: Mapped[int] = mapped_column(sa.Integer, nullable=False)
     config_snapshot: Mapped[Any] = mapped_column(JSONModelColumn(AgentSoulConfig), nullable=False)
+    home_snapshot_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
     summary: Mapped[str | None] = mapped_column(LongText, nullable=True)
     version_note: Mapped[str | None] = mapped_column(LongText, nullable=True)
     created_by: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
@@ -432,102 +461,83 @@ class WorkflowAgentNodeBinding(DefaultFieldsMixin, Base):
         return dict(self.node_job_config)
 
 
-class AgentRuntimeSession(DefaultFieldsMixin, Base):
-    """Persisted Agent backend session snapshot, owner-agnostic.
+class AgentWorkspace(DefaultFieldsMixin, Base):
+    """Mutable Workspace owned by one product scope, independent of Agents."""
 
-    One unified table serves both owners (decision Q2):
-    - workflow Agent Node runs: ``owner_type = workflow_run``; the
-      ``workflow_id / workflow_run_id / node_id / binding_id /
-      agent_config_snapshot_id / composition_layer_specs`` columns are set.
-    - Agent App conversations: ``owner_type = conversation``; the
-      ``conversation_id`` column is set and the workflow columns stay NULL.
-      Runtime state is scoped by ``agent_config_snapshot_id``. For published
-      web/API runs this points to an immutable AgentConfigSnapshot; for console
-      debugger/build runs it points to the editable AgentConfigDraft row.
-
-    The snapshot is runtime state returned by Agent backend, kept separate from
-    Agent Soul snapshots and workflow node-job config.
-    """
-
-    __tablename__ = "agent_runtime_sessions"
+    __tablename__ = "agent_workspaces"
     __table_args__ = (
-        sa.PrimaryKeyConstraint("id", name="agent_runtime_session_pkey"),
-        # Workflow owner uniqueness (partial: only rows with a workflow_run_id).
+        sa.PrimaryKeyConstraint("id", name="agent_workspace_pkey"),
         Index(
-            "agent_runtime_session_workflow_scope_unique",
+            "agent_workspace_owner_active_unique",
             "tenant_id",
-            "workflow_run_id",
-            "node_id",
-            "binding_id",
-            "agent_id",
+            "owner_type",
+            "owner_id",
+            "owner_scope_key",
+            "active_guard",
             unique=True,
-            postgresql_where=sa.text("workflow_run_id IS NOT NULL"),
         ),
-        # Conversation owner uniqueness (partial: only rows with a conversation_id).
-        Index(
-            "agent_runtime_session_conversation_scope_unique",
-            "tenant_id",
-            "conversation_id",
-            "agent_id",
-            "agent_config_snapshot_id",
-            unique=True,
-            postgresql_where=sa.text("conversation_id IS NOT NULL"),
-        ),
-        Index(
-            "agent_runtime_session_workflow_lookup_idx",
-            "tenant_id",
-            "workflow_run_id",
-            "node_id",
-            "status",
-        ),
-        Index(
-            "agent_runtime_session_conversation_lookup_idx",
-            "tenant_id",
-            "conversation_id",
-            "status",
-        ),
-        Index("agent_runtime_session_backend_run_idx", "backend_run_id"),
+        Index("agent_workspace_tenant_status_idx", "tenant_id", "status"),
+        Index("agent_workspace_tenant_app_status_idx", "tenant_id", "app_id", "status"),
+        Index("agent_workspace_status_retired_idx", "status", "retired_at"),
     )
 
     tenant_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
     app_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
-    owner_type: Mapped[AgentRuntimeSessionOwnerType] = mapped_column(
-        EnumText(AgentRuntimeSessionOwnerType, length=32), nullable=False
+    owner_type: Mapped[AgentWorkspaceOwnerType] = mapped_column(
+        EnumText(AgentWorkspaceOwnerType, length=32), nullable=False
     )
-    agent_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
-    backend_run_id: Mapped[str | None] = mapped_column(String(255), nullable=True)
-    session_snapshot: Mapped[str] = mapped_column(LongText, nullable=False)
-    # Workflow-owner columns (NULL for conversation owner).
-    workflow_id: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
-    workflow_run_id: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
-    node_id: Mapped[str | None] = mapped_column(String(255), nullable=True)
-    node_execution_id: Mapped[str | None] = mapped_column(String(255), nullable=True)
-    binding_id: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
-    agent_config_snapshot_id: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
-    # JSON-encoded list of non-sensitive runtime layer specs ({name, type, deps,
-    # config}). The persisted schema keeps its original name because the sandbox
-    # refactor intentionally avoids a storage migration.
-    composition_layer_specs: Mapped[str] = mapped_column(LongText, nullable=False, server_default="[]")
-    # Conversation-owner column (NULL for workflow owner).
-    conversation_id: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
-    status: Mapped[AgentRuntimeSessionStatus] = mapped_column(
-        EnumText(AgentRuntimeSessionStatus, length=32),
+    owner_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    owner_scope_key: Mapped[str] = mapped_column(String(255), nullable=False)
+    backend_workspace_ref: Mapped[str] = mapped_column(String(255), nullable=False)
+    status: Mapped[AgentWorkingResourceStatus] = mapped_column(
+        EnumText(AgentWorkingResourceStatus, length=32),
         nullable=False,
-        default=AgentRuntimeSessionStatus.ACTIVE,
+        default=AgentWorkingResourceStatus.ACTIVE,
+        server_default=AgentWorkingResourceStatus.ACTIVE.value,
     )
-    cleaned_at: Mapped[datetime | None] = mapped_column(DateTime, nullable=True)
-    # ENG-637: when a run pauses for a dify.ask_human deferred call, these link
-    # the session to the awaiting HITL form and the deferred tool_call_id, so a
-    # resumed node can map the submitted form back into deferred_tool_results.
-    # Both NULL whenever the session is not paused on human input.
+    active_guard: Mapped[int | None] = mapped_column(sa.SmallInteger, nullable=True, default=1, server_default="1")
+    retired_at: Mapped[datetime | None] = mapped_column(DateTime, nullable=True)
+
+
+class AgentWorkspaceBinding(DefaultFieldsMixin, Base):
+    """One materialized Agent participant and session attached to a Workspace.
+
+    All resource IDs are logical associations rather than database foreign
+    keys, so RETIRED rows can outlive their Workspace or base Home Snapshot.
+    ``agent_id`` identifies the source Agent Soul; this row's ``id`` identifies
+    the participant and its private Materialized Home.
+    """
+
+    __tablename__ = "agent_workspace_bindings"
+    __table_args__ = (
+        sa.PrimaryKeyConstraint("id", name="agent_workspace_binding_pkey"),
+        Index("agent_workspace_binding_workspace_status_idx", "tenant_id", "workspace_id", "status"),
+        Index("agent_workspace_binding_agent_status_idx", "tenant_id", "agent_id", "status"),
+        Index("agent_workspace_binding_status_retired_idx", "status", "retired_at"),
+    )
+
+    tenant_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    app_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    workspace_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    agent_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    base_home_snapshot_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    agent_config_version_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    agent_config_version_kind: Mapped[AgentConfigVersionKind] = mapped_column(
+        EnumText(AgentConfigVersionKind, length=32), nullable=False
+    )
+    backend_binding_ref: Mapped[str] = mapped_column(String(255), nullable=False)
+    session_snapshot: Mapped[str | None] = mapped_column(LongText, nullable=True)
+    status: Mapped[AgentWorkingResourceStatus] = mapped_column(
+        EnumText(AgentWorkingResourceStatus, length=32),
+        nullable=False,
+        default=AgentWorkingResourceStatus.ACTIVE,
+        server_default=AgentWorkingResourceStatus.ACTIVE.value,
+    )
+    retired_at: Mapped[datetime | None] = mapped_column(DateTime, nullable=True)
     pending_form_id: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
     pending_tool_call_id: Mapped[str | None] = mapped_column(String(255), nullable=True)
 
 
-# Back-compat alias for the shipped workflow lifecycle code (PR #36724).
-WorkflowAgentRuntimeSession = AgentRuntimeSession
-
-
 class AgentDriveFileKind(StrEnum):
     """Kind of existing file record an agent-drive KV entry points at."""
 

api/models/model.py

@@ -1160,6 +1160,13 @@ class OAuthProviderApp(TypeBase):
 
 
 class Conversation(Base):
+    """Conversation state, including the exact Agent participant when applicable.
+
+    ``agent_workspace_binding_id`` is a logical pointer rather than a foreign
+    key because retired Binding ledger rows may be collected before the
+    conversation history is deleted.
+    """
+
     __tablename__ = "conversations"
     __table_args__ = (
         sa.PrimaryKeyConstraint("id", name="conversation_pkey"),
@@ -1181,6 +1188,7 @@ class Conversation(Base):
     id: Mapped[str] = mapped_column(StringUUID, default=lambda: str(uuid4()))
     app_id = mapped_column(StringUUID, nullable=False)
     app_model_config_id = mapped_column(StringUUID, nullable=True)
+    agent_workspace_binding_id: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
     model_provider = mapped_column(String(255), nullable=True)
     override_model_configs = mapped_column(LongText)
     model_id = mapped_column(String(255), nullable=True)

api/models/workflow.py

@@ -1031,6 +1031,7 @@ class WorkflowNodeExecutionModel(Base):  # This model is expected to have `offlo
     node_id: Mapped[str] = mapped_column(String(255))
     node_type: Mapped[str] = mapped_column(String(255))
     title: Mapped[str] = mapped_column(String(255))
+    agent_workspace_binding_id: Mapped[str | None] = mapped_column(StringUUID, nullable=True)
     inputs: Mapped[str | None] = mapped_column(LongText)
     process_data: Mapped[str | None] = mapped_column(LongText)
     outputs: Mapped[str | None] = mapped_column(LongText)

api/openapi/markdown/console-openapi.md

@@ -964,12 +964,6 @@ Stop a running Agent App chat message generation
 | ---- | ---------- | ----------- | -------- | ------ |
 | agent_id | path |  | Yes | string (uuid) |
 
-#### Request Body
-
-| Required | Schema |
-| -------- | ------ |
-|  No | **application/json**: [AgentDebugConversationRefreshPayload](#agentdebugconversationrefreshpayload)<br> |
-
 #### Responses
 
 | Code | Description | Schema |
@@ -1261,7 +1255,8 @@ Get basic information for an Agent App conversation sandbox
 | Name | Located in | Description | Required | Schema |
 | ---- | ---------- | ----------- | -------- | ------ |
 | agent_id | path | Agent ID | Yes | string (uuid) |
-| conversation_id | query | Agent App conversation ID | Yes | string |
+| caller_id | query | Agent App caller ID | Yes | string |
+| caller_type | query |  | Yes | string, <br>**Available values:** "build_draft", "conversation" |
 
 #### Responses
 
@@ -1277,7 +1272,8 @@ List a directory in an Agent App conversation sandbox
 | Name | Located in | Description | Required | Schema |
 | ---- | ---------- | ----------- | -------- | ------ |
 | agent_id | path | Agent ID | Yes | string (uuid) |
-| conversation_id | query | Agent App conversation ID | Yes | string |
+| caller_id | query | Agent App caller ID | Yes | string |
+| caller_type | query |  | Yes | string, <br>**Available values:** "build_draft", "conversation" |
 | path | query | Directory path relative to the sandbox workspace | No | string, <br>**Default:** . |
 
 #### Responses
@@ -1294,7 +1290,8 @@ Read a text/binary preview file in an Agent App conversation sandbox
 | Name | Located in | Description | Required | Schema |
 | ---- | ---------- | ----------- | -------- | ------ |
 | agent_id | path | Agent ID | Yes | string (uuid) |
-| conversation_id | query | Agent App conversation ID | Yes | string |
+| caller_id | query | Agent App caller ID | Yes | string |
+| caller_type | query |  | Yes | string, <br>**Available values:** "build_draft", "conversation" |
 | path | query | File path relative to the sandbox workspace | Yes | string |
 
 #### Responses
@@ -3807,7 +3804,7 @@ List a directory in a workflow Agent node sandbox
 | app_id | path | Application ID | Yes | string (uuid) |
 | node_id | path | Workflow Agent node ID | Yes | string |
 | workflow_run_id | path | Workflow run ID | Yes | string (uuid) |
-| node_execution_id | query | Optional workflow node execution ID. When omitted, the latest active session for the node is used. | No | string |
+| node_execution_id | query | Workflow node execution ID | Yes | string |
 | path | query | Directory path relative to the sandbox workspace | No | string, <br>**Default:** . |
 
 #### Responses
@@ -3826,7 +3823,7 @@ Read a text/binary preview file in a workflow Agent node sandbox
 | app_id | path | Application ID | Yes | string (uuid) |
 | node_id | path | Workflow Agent node ID | Yes | string |
 | workflow_run_id | path | Workflow run ID | Yes | string (uuid) |
-| node_execution_id | query | Optional workflow node execution ID. When omitted, the latest active session for the node is used. | No | string |
+| node_execution_id | query | Workflow node execution ID | Yes | string |
 | path | query | File path relative to the sandbox workspace | Yes | string |
 
 #### Responses
@@ -13943,12 +13940,6 @@ Stable Agent Soul reference to one normalized skill archive.
 | date | string |  | Yes |
 | message_count | integer |  | Yes |
 
-#### AgentDebugConversationRefreshPayload
-
-| Name | Type | Description | Required |
-| ---- | ---- | ----------- | -------- |
-| draft_type | [AgentConfigDraftType](#agentconfigdrafttype) | Agent draft surface whose conversation should be refreshed | No |
-
 #### AgentDebugConversationRefreshResponse
 
 | Name | Type | Description | Required |
@@ -14666,7 +14657,8 @@ section may be empty, which is how callers express "no knowledge layer".
 
 | Name | Type | Description | Required |
 | ---- | ---- | ----------- | -------- |
-| conversation_id | string | Agent App conversation ID | Yes |
+| caller_id | string | Agent App caller ID | Yes |
+| caller_type | string, <br>**Available values:** "build_draft", "conversation" | *Enum:* `"build_draft"`, `"conversation"` | Yes |
 | path | string | File path relative to the sandbox workspace | Yes |
 
 #### AgentScope
@@ -21347,7 +21339,6 @@ Whitelist scopes accepted by RBAC app and dataset access config APIs.
 
 | Name | Type | Description | Required |
 | ---- | ---- | ----------- | -------- |
-| session_id | string |  | Yes |
 | workspace_cwd | string |  | Yes |
 
 #### SandboxListResponse
@@ -23167,7 +23158,7 @@ How a workflow node is bound to an Agent.
 
 | Name | Type | Description | Required |
 | ---- | ---- | ----------- | -------- |
-| node_execution_id | string | Optional workflow node execution ID. When omitted, the latest active session for the node is used. | No |
+| node_execution_id | string | Workflow node execution ID | Yes |
 | path | string | File path relative to the sandbox workspace | Yes |
 
 #### WorkflowAppLogPaginationResponse

api/providers/vdb/vdb-oracle/src/dify_vdb_oracle/oraclevector.py

@@ -316,10 +316,10 @@ class OracleVector(BaseVector):
                     entities.append(current_entity)
             else:
                 try:
-                    nltk.data.find("tokenizers/punkt")
+                    nltk.data.find("tokenizers/punkt_tab")
                     nltk.data.find("corpora/stopwords")
                 except LookupError:
-                    raise LookupError("Unable to find the required NLTK data package: punkt and stopwords")
+                    raise LookupError("Unable to find the required NLTK data package: punkt_tab and stopwords")
                 e_str = re.sub(r"[^\w ]", "", query)
                 all_tokens = nltk.word_tokenize(e_str)
                 stop_words = stopwords.words("english")

api/pyproject.toml

@@ -206,7 +206,7 @@ storage = [
 ############################################################
 # [ Tools ] dependency group
 ############################################################
-tools = ["cloudscraper>=1.2.71,<2.0.0", "nltk>=3.9.1,<4.0.0"]
+tools = ["cloudscraper>=1.2.71,<2.0.0", "nltk>=3.10.0,<4.0.0"]
 
 ############################################################
 # [ VDB ] workspace plugins — hollow packages under providers/vdb/*

api/services/agent/composer_service.py

@@ -9,7 +9,7 @@ from sqlalchemy.sql.elements import ColumnElement
 
 from core.agent.publish_visibility import agent_has_workflow_callable_active_snapshot
 from libs.helper import to_timestamp
-from models import Account
+from models import Account, Conversation
 from models.agent import (
     APP_BACKED_AGENT_SOURCES,
     Agent,
@@ -18,12 +18,15 @@ from models.agent import (
     AgentConfigRevision,
     AgentConfigRevisionOperation,
     AgentConfigSnapshot,
+    AgentConfigVersionKind,
+    AgentDebugConversation,
     AgentDriveFile,
     AgentIconType,
     AgentKind,
     AgentScope,
     AgentSource,
     AgentStatus,
+    AgentWorkspaceOwnerType,
     WorkflowAgentBindingType,
     WorkflowAgentNodeBinding,
 )
@@ -35,6 +38,7 @@ from models.workflow import Workflow
 from services.agent.agent_soul_state import agent_soul_has_model
 from services.agent.composer_validator import ComposerConfigValidator
 from services.agent.errors import (
+    AgentBuildSandboxNotFoundError,
     AgentModelNotConfiguredError,
     AgentNameConflictError,
     AgentNotFoundError,
@@ -42,11 +46,17 @@ from services.agent.errors import (
     AgentVersionNotFoundError,
     InvalidComposerConfigError,
 )
+from services.agent.home_snapshot_service import (
+    AgentHomeSnapshotService,
+    validate_home_snapshot_binding,
+)
 from services.agent.knowledge_datasets import (
     get_tenant_knowledge_dataset_rows,
     list_missing_tenant_knowledge_dataset_ids,
 )
+from services.agent.retirement_service import WorkflowAgentRetirementService
 from services.agent.roster_service import AgentRosterService
+from services.agent.workspace_service import AgentWorkspaceNotFoundError, AgentWorkspaceService, WorkspaceOwnerScope
 from services.app_service import AppService, CreateAppParams
 from services.entities.agent_entities import (
     AgentSoulConfig,
@@ -56,6 +66,7 @@ from services.entities.agent_entities import (
     ComposerVariant,
     WorkflowNodeJobConfig,
 )
+from tasks.collect_agent_resources_task import enqueue_agent_resource_collection
 
 # WorkflowAgentNodeBinding.workflow_version tag for the draft workflow row.
 # Mirrors Workflow.version when it is "draft" (see models/workflow.py).
@@ -198,6 +209,18 @@ class AgentComposerService:
         binding = cls._get_workflow_binding(
             session=session, tenant_id=tenant_id, workflow_id=workflow.id, node_id=node_id
         )
+        retirement_candidates = (
+            {binding.agent_id}
+            if binding is not None
+            and binding.binding_type == WorkflowAgentBindingType.INLINE_AGENT
+            and binding.agent_id
+            and payload.save_strategy
+            in {
+                ComposerSaveStrategy.SAVE_AS_NEW_AGENT,
+                ComposerSaveStrategy.SAVE_TO_ROSTER,
+            }
+            else set()
+        )
 
         match payload.save_strategy:
             case ComposerSaveStrategy.NODE_JOB_ONLY:
@@ -232,7 +255,11 @@ class AgentComposerService:
                 )
             case ComposerSaveStrategy.SAVE_TO_ROSTER:
                 binding = cls._save_to_roster(
-                    session=session, tenant_id=tenant_id, account_id=account_id, binding=binding, payload=payload
+                    session=session,
+                    tenant_id=tenant_id,
+                    account_id=account_id,
+                    binding=binding,
+                    payload=payload,
                 )
 
         session.flush()
@@ -257,6 +284,17 @@ class AgentComposerService:
             payload=payload,
             agent_id=binding.agent_id,
         )
+        session.commit()
+        binding_ids, home_snapshot_ids = WorkflowAgentRetirementService.retire_unowned(
+            tenant_id=tenant_id,
+            agent_ids=retirement_candidates,
+            account_id=account_id,
+        )
+        enqueue_agent_resource_collection(
+            tenant_id=tenant_id,
+            binding_ids=binding_ids,
+            home_snapshot_ids=home_snapshot_ids,
+        )
         return state
 
     @classmethod
@@ -390,12 +428,10 @@ class AgentComposerService:
 
     @classmethod
     def _load_agent_composer_for_agent(cls, *, session: Session, tenant_id: str, agent: Agent) -> dict[str, Any]:
-        draft = cls._get_or_create_agent_draft(
+        draft = cls.get_or_create_normal_agent_draft(
             session=session,
             tenant_id=tenant_id,
             agent=agent,
-            draft_type=AgentConfigDraftType.DRAFT,
-            account_id=None,
             created_by=agent.updated_by or agent.created_by,
         )
         version = cls._get_version_if_present(
@@ -417,7 +453,34 @@ class AgentComposerService:
 
     @classmethod
     def save_agent_app_composer(
-        cls, *, session: Session, tenant_id: str, app_id: str, account_id: str, payload: ComposerSavePayload
+        cls,
+        *,
+        session: Session,
+        tenant_id: str,
+        app_id: str,
+        account_id: str,
+        payload: ComposerSavePayload,
+    ) -> dict[str, Any]:
+        try:
+            return cls._save_agent_app_composer_impl(
+                session=session,
+                tenant_id=tenant_id,
+                app_id=app_id,
+                account_id=account_id,
+                payload=payload,
+            )
+        except IntegrityError as exc:
+            raise AgentNameConflictError() from exc
+
+    @classmethod
+    def _save_agent_app_composer_impl(
+        cls,
+        *,
+        session: Session,
+        tenant_id: str,
+        app_id: str,
+        account_id: str,
+        payload: ComposerSavePayload,
     ) -> dict[str, Any]:
         if payload.variant != ComposerVariant.AGENT_APP:
             raise ValueError("Agent App composer endpoint only accepts agent_app variant")
@@ -446,11 +509,25 @@ class AgentComposerService:
                 updated_by=account_id,
             )
             session.add(agent)
-            try:
-                session.flush()
-            except IntegrityError as exc:
-                session.rollback()
-                raise AgentNameConflictError() from exc
+            session.flush()
+            home_snapshot = AgentHomeSnapshotService.create_initial(
+                session=session,
+                tenant_id=tenant_id,
+                agent_id=agent.id,
+            )
+            initial_version = cls._create_config_version(
+                session=session,
+                tenant_id=tenant_id,
+                agent_id=agent.id,
+                account_id=account_id,
+                agent_soul=AgentSoulConfig(),
+                operation=AgentConfigRevisionOperation.CREATE_VERSION,
+                version_note=None,
+                home_snapshot_id=home_snapshot.id,
+            )
+            agent.active_config_snapshot_id = initial_version.id
+            agent.active_config_has_model = False
+            agent.active_config_is_published = False
         return cls._save_agent_composer_for_agent(
             session=session,
             tenant_id=tenant_id,
@@ -488,7 +565,7 @@ class AgentComposerService:
     ) -> dict[str, Any]:
         if payload.agent_soul is None:
             raise ValueError("agent_soul is required")
-        cls._save_agent_draft(
+        draft = cls._save_agent_draft(
             session=session,
             tenant_id=tenant_id,
             agent=agent,
@@ -503,6 +580,7 @@ class AgentComposerService:
             tenant_id=tenant_id,
             agent=agent,
             agent_soul=payload.agent_soul,
+            home_snapshot_id=draft.home_snapshot_id,
         )
 
         session.flush()
@@ -523,6 +601,7 @@ class AgentComposerService:
         tenant_id: str,
         agent: Agent,
         agent_soul: AgentSoulConfig,
+        home_snapshot_id: str,
     ) -> bool:
         if not agent.active_config_snapshot_id:
             return False
@@ -538,7 +617,9 @@ class AgentComposerService:
         if not agent_has_workflow_callable_active_snapshot(session=session, agent=agent):
             return False
 
-        return _agent_soul_config_json(agent_soul) == _agent_soul_config_json(active_version.config_snapshot_dict)
+        return home_snapshot_id == active_version.home_snapshot_id and _agent_soul_config_json(
+            agent_soul
+        ) == _agent_soul_config_json(active_version.config_snapshot_dict)
 
     @classmethod
     def publish_agent_app_draft(
@@ -567,6 +648,11 @@ class AgentComposerService:
         if not agent_soul_has_model(agent_soul):
             raise AgentModelNotConfiguredError()
         cls.validate_knowledge_datasets(session=session, tenant_id=tenant_id, agent_soul=agent_soul)
+        validate_home_snapshot_binding(
+            session=session,
+            agent=agent,
+            home_snapshot_id=draft.home_snapshot_id,
+        )
         version = cls._create_config_version(
             session=session,
             tenant_id=tenant_id,
@@ -576,6 +662,7 @@ class AgentComposerService:
             operation=AgentConfigRevisionOperation.PUBLISH_DRAFT,
             version_note=version_note,
             previous_snapshot_id=agent.active_config_snapshot_id,
+            home_snapshot_id=draft.home_snapshot_id,
         )
         agent.active_config_snapshot_id = version.id
         agent.active_config_has_model = agent_soul_has_model(agent_soul)
@@ -595,6 +682,29 @@ class AgentComposerService:
     def checkout_agent_app_build_draft(
         cls, *, session: Session, tenant_id: str, agent_id: str, account_id: str, force: bool = False
     ) -> dict[str, Any]:
+        try:
+            result, retired_binding_id = cls._checkout_agent_app_build_draft_in_transaction(
+                session=session,
+                tenant_id=tenant_id,
+                agent_id=agent_id,
+                account_id=account_id,
+                force=force,
+            )
+            session.commit()
+        except Exception:
+            session.rollback()
+            raise
+        if retired_binding_id is not None:
+            enqueue_agent_resource_collection(
+                tenant_id=tenant_id,
+                binding_ids=(retired_binding_id,),
+            )
+        return result
+
+    @classmethod
+    def _checkout_agent_app_build_draft_in_transaction(
+        cls, *, session: Session, tenant_id: str, agent_id: str, account_id: str, force: bool
+    ) -> tuple[dict[str, Any], str | None]:
         agent = cls._require_agent(session=session, tenant_id=tenant_id, agent_id=agent_id)
         normal_draft = cls._get_or_create_agent_draft(
             session=session,
@@ -612,7 +722,23 @@ class AgentComposerService:
             account_id=account_id,
         )
         if build_draft is not None and not force:
-            return cls._serialize_build_draft_state(build_draft)
+            return cls._serialize_build_draft_state(build_draft), None
+        retired_binding_id: str | None = None
+        if build_draft is not None and build_draft.agent_workspace_binding_id is not None:
+            cls._validate_active_build_draft_binding(
+                session=session,
+                tenant_id=tenant_id,
+                agent=agent,
+                build_draft=build_draft,
+            )
+            retired_binding_id = AgentWorkspaceService.retire_binding(
+                session=session,
+                tenant_id=tenant_id,
+                binding_id=build_draft.agent_workspace_binding_id,
+            )
+            if retired_binding_id is None:
+                raise AgentBuildSandboxNotFoundError()
+            build_draft.agent_workspace_binding_id = None
         if build_draft is None:
             build_draft = AgentConfigDraft(
                 tenant_id=tenant_id,
@@ -624,10 +750,44 @@ class AgentComposerService:
             )
             session.add(build_draft)
         build_draft.base_snapshot_id = normal_draft.base_snapshot_id
+        build_draft.home_snapshot_id = normal_draft.home_snapshot_id
         build_draft.config_snapshot = AgentSoulConfig.model_validate(normal_draft.config_snapshot_dict)
         build_draft.updated_by = account_id
         session.flush()
-        return cls._serialize_build_draft_state(build_draft)
+        return cls._serialize_build_draft_state(build_draft), retired_binding_id
+
+    @classmethod
+    def _validate_active_build_draft_binding(
+        cls,
+        *,
+        session: Session,
+        tenant_id: str,
+        agent: Agent,
+        build_draft: AgentConfigDraft,
+    ) -> None:
+        binding_id = build_draft.agent_workspace_binding_id
+        runtime_app_id = AgentRosterService.runtime_backing_app_id(agent)
+        if binding_id is None or runtime_app_id is None:
+            raise AgentBuildSandboxNotFoundError()
+        binding = AgentWorkspaceService.get_active_binding(
+            session=session,
+            tenant_id=tenant_id,
+            binding_id=binding_id,
+            expected_owner_scope=WorkspaceOwnerScope(
+                tenant_id=tenant_id,
+                app_id=runtime_app_id,
+                owner_type=AgentWorkspaceOwnerType.BUILD_DRAFT,
+                owner_id=build_draft.id,
+            ),
+        )
+        if binding is None or binding.agent_id != agent.id:
+            raise AgentBuildSandboxNotFoundError()
+        AgentWorkspaceService.validate_binding_generation(
+            binding,
+            base_home_snapshot_id=build_draft.home_snapshot_id,
+            agent_config_version_id=build_draft.id,
+            agent_config_version_kind=AgentConfigVersionKind.BUILD_DRAFT,
+        )
 
     @classmethod
     def load_agent_app_build_draft(
@@ -669,6 +829,29 @@ class AgentComposerService:
     def apply_agent_app_build_draft(
         cls, *, session: Session, tenant_id: str, agent_id: str, account_id: str
     ) -> dict[str, Any]:
+        try:
+            result, retired_binding_ids = cls._apply_agent_app_build_draft_in_transaction(
+                session=session,
+                tenant_id=tenant_id,
+                agent_id=agent_id,
+                account_id=account_id,
+            )
+            session.commit()
+        except Exception:
+            session.rollback()
+            raise
+        enqueue_agent_resource_collection(tenant_id=tenant_id, binding_ids=retired_binding_ids)
+        return result
+
+    @classmethod
+    def _apply_agent_app_build_draft_in_transaction(
+        cls,
+        *,
+        session: Session,
+        tenant_id: str,
+        agent_id: str,
+        account_id: str,
+    ) -> tuple[dict[str, Any], list[str]]:
         agent = cls._require_agent(session=session, tenant_id=tenant_id, agent_id=agent_id)
         build_draft = cls._get_agent_draft(
             session=session,
@@ -680,6 +863,21 @@ class AgentComposerService:
         if build_draft is None:
             raise AgentVersionNotFoundError()
         applied_agent_soul = AgentSoulConfig.model_validate(build_draft.config_snapshot_dict)
+        ComposerConfigValidator.validate_publish_payload(
+            ComposerSavePayload(
+                variant=ComposerVariant.AGENT_APP,
+                agent_soul=applied_agent_soul,
+                save_strategy=ComposerSaveStrategy.SAVE_AS_NEW_VERSION,
+            )
+        )
+        cls.validate_knowledge_datasets(session=session, tenant_id=tenant_id, agent_soul=applied_agent_soul)
+        source_binding_id = build_draft.agent_workspace_binding_id
+        if source_binding_id is None:
+            raise AgentBuildSandboxNotFoundError()
+        home_snapshot = AgentHomeSnapshotService.create_for_build_apply(
+            session=session,
+            build_draft=build_draft,
+        )
         normal_draft = cls._save_agent_draft(
             session=session,
             tenant_id=tenant_id,
@@ -690,32 +888,145 @@ class AgentComposerService:
             account_id_for_audit=account_id,
             base_snapshot_id=build_draft.base_snapshot_id,
         )
+        retired_binding_ids = cls._retire_normal_preview_bindings(
+            session=session,
+            tenant_id=tenant_id,
+            agent=agent,
+            normal_draft=normal_draft,
+        )
+        normal_draft.home_snapshot_id = home_snapshot.id
         agent.active_config_is_published = cls._agent_soul_matches_active_config(
             session=session,
             tenant_id=tenant_id,
             agent=agent,
             agent_soul=applied_agent_soul,
+            home_snapshot_id=home_snapshot.id,
         )
         agent.updated_by = account_id
+        retired_binding_id = AgentWorkspaceService.retire_binding(
+            session=session,
+            tenant_id=tenant_id,
+            binding_id=source_binding_id,
+        )
+        if retired_binding_id is None:
+            raise AgentBuildSandboxNotFoundError()
+        retired_binding_ids.append(source_binding_id)
         session.delete(build_draft)
-        session.flush()
-        return {"result": "success", "draft": cls._serialize_draft(normal_draft)}
+        return {"result": "success", "draft": cls._serialize_draft(normal_draft)}, retired_binding_ids
+
+    @classmethod
+    def _retire_normal_preview_bindings(
+        cls,
+        *,
+        session: Session,
+        tenant_id: str,
+        agent: Agent,
+        normal_draft: AgentConfigDraft,
+    ) -> list[str]:
+        """Retire Preview participants before Build Apply replaces the shared Draft Home."""
+
+        mappings = session.scalars(
+            select(AgentDebugConversation).where(
+                AgentDebugConversation.tenant_id == tenant_id,
+                AgentDebugConversation.agent_id == agent.id,
+                AgentDebugConversation.draft_type == AgentConfigDraftType.DRAFT,
+            )
+        ).all()
+        retired_binding_ids: list[str] = []
+        for mapping in mappings:
+            conversation = session.scalar(
+                select(Conversation).where(
+                    Conversation.id == mapping.conversation_id,
+                    Conversation.app_id == mapping.app_id,
+                    Conversation.is_deleted.is_(False),
+                )
+            )
+            if conversation is None or conversation.agent_workspace_binding_id is None:
+                continue
+            binding_id = conversation.agent_workspace_binding_id
+            binding = AgentWorkspaceService.get_active_binding(
+                session=session,
+                tenant_id=tenant_id,
+                binding_id=binding_id,
+                expected_owner_scope=WorkspaceOwnerScope(
+                    tenant_id=tenant_id,
+                    app_id=mapping.app_id,
+                    owner_type=AgentWorkspaceOwnerType.CONVERSATION,
+                    owner_id=conversation.id,
+                ),
+            )
+            if binding is None or binding.agent_id != agent.id:
+                raise AgentWorkspaceNotFoundError("Agent Preview participant Binding is unavailable")
+            AgentWorkspaceService.validate_binding_generation(
+                binding,
+                base_home_snapshot_id=normal_draft.home_snapshot_id,
+                agent_config_version_id=normal_draft.id,
+                agent_config_version_kind=AgentConfigVersionKind.DRAFT,
+            )
+            retired_binding_id = AgentWorkspaceService.retire_binding(
+                session=session,
+                tenant_id=tenant_id,
+                binding_id=binding_id,
+            )
+            if retired_binding_id is None:
+                raise AgentWorkspaceNotFoundError("Agent Preview participant Binding is unavailable")
+            conversation.agent_workspace_binding_id = None
+            retired_binding_ids.append(binding_id)
+        return retired_binding_ids
 
     @classmethod
     def discard_agent_app_build_draft(
         cls, *, session: Session, tenant_id: str, agent_id: str, account_id: str
     ) -> dict[str, Any]:
+        try:
+            result, retired_binding_id = cls._discard_agent_app_build_draft_in_transaction(
+                session=session,
+                tenant_id=tenant_id,
+                agent_id=agent_id,
+                account_id=account_id,
+            )
+            session.commit()
+        except Exception:
+            session.rollback()
+            raise
+        if retired_binding_id is not None:
+            enqueue_agent_resource_collection(
+                tenant_id=tenant_id,
+                binding_ids=(retired_binding_id,),
+            )
+        return result
+
+    @classmethod
+    def _discard_agent_app_build_draft_in_transaction(
+        cls, *, session: Session, tenant_id: str, agent_id: str, account_id: str
+    ) -> tuple[dict[str, Any], str | None]:
+        agent = cls._require_agent(session=session, tenant_id=tenant_id, agent_id=agent_id)
         build_draft = cls._get_agent_draft(
             session=session,
             tenant_id=tenant_id,
-            agent_id=agent_id,
+            agent_id=agent.id,
             draft_type=AgentConfigDraftType.DEBUG_BUILD,
             account_id=account_id,
         )
-        if build_draft is not None:
-            session.delete(build_draft)
-            session.flush()
-        return {"result": "success"}
+        if build_draft is None:
+            return {"result": "success"}, None
+        retired_binding_id: str | None = None
+        if build_draft.agent_workspace_binding_id is not None:
+            cls._validate_active_build_draft_binding(
+                session=session,
+                tenant_id=tenant_id,
+                agent=agent,
+                build_draft=build_draft,
+            )
+            retired_binding_id = AgentWorkspaceService.retire_binding(
+                session=session,
+                tenant_id=tenant_id,
+                binding_id=build_draft.agent_workspace_binding_id,
+            )
+            if retired_binding_id is None:
+                raise AgentBuildSandboxNotFoundError()
+        session.delete(build_draft)
+        return {"result": "success"}, retired_binding_id
 
     @classmethod
     def collect_validation_findings(
@@ -1279,6 +1590,12 @@ class AgentComposerService:
         binding = cls._require_binding(binding)
         if not binding.agent_id or payload.agent_soul is None:
             raise ValueError("agent_id and agent_soul are required")
+        current_snapshot = cls._require_version(
+            session=session,
+            tenant_id=tenant_id,
+            agent_id=binding.agent_id,
+            version_id=binding.current_snapshot_id,
+        )
         version = cls._create_config_version(
             session=session,
             tenant_id=tenant_id,
@@ -1287,6 +1604,7 @@ class AgentComposerService:
             agent_soul=payload.agent_soul,
             operation=AgentConfigRevisionOperation.SAVE_NEW_VERSION,
             version_note=payload.version_note,
+            home_snapshot_id=current_snapshot.home_snapshot_id,
         )
         agent = cls._require_agent(session=session, tenant_id=tenant_id, agent_id=binding.agent_id)
         agent.active_config_snapshot_id = version.id
@@ -1449,6 +1767,11 @@ class AgentComposerService:
         )
         session.add(agent)
         session.flush()
+        home_snapshot = AgentHomeSnapshotService.create_initial(
+            session=session,
+            tenant_id=tenant_id,
+            agent_id=agent.id,
+        )
         version = cls._create_config_version(
             session=session,
             tenant_id=tenant_id,
@@ -1457,6 +1780,7 @@ class AgentComposerService:
             agent_soul=agent_soul,
             operation=AgentConfigRevisionOperation.CREATE_VERSION,
             version_note=None,
+            home_snapshot_id=home_snapshot.id,
         )
         agent.active_config_snapshot_id = version.id
         agent.active_config_has_model = agent_soul_has_model(agent_soul)
@@ -1590,7 +1914,6 @@ class AgentComposerService:
                 session=session,
             )
         except IntegrityError as exc:
-            session.rollback()
             raise AgentNameConflictError() from exc
 
         agent = AgentRosterService(session).get_app_backing_agent(tenant_id=tenant_id, app_id=app.id)
@@ -1628,6 +1951,7 @@ class AgentComposerService:
         agent_soul: AgentSoulConfig,
         operation: AgentConfigRevisionOperation,
         version_note: str | None,
+        home_snapshot_id: str,
         previous_snapshot_id: str | None = None,
     ) -> AgentConfigSnapshot:
         next_version = (
@@ -1644,6 +1968,7 @@ class AgentComposerService:
             agent_id=agent_id,
             version=next_version,
             config_snapshot=agent_soul,
+            home_snapshot_id=home_snapshot_id,
             version_note=version_note,
             created_by=account_id,
         )
@@ -1683,6 +2008,7 @@ class AgentComposerService:
             operation=operation,
             version_note=version_note,
             previous_snapshot_id=current_snapshot.id,
+            home_snapshot_id=current_snapshot.home_snapshot_id,
         )
 
     @classmethod
@@ -1749,7 +2075,10 @@ class AgentComposerService:
         agent: Agent,
         created_by: str | None,
     ) -> AgentConfigDraft:
-        """Resolve the shared Preview draft, rebasing inline agents when needed."""
+        """Resolve the normal Draft, rebasing only stale WORKFLOW_ONLY DRAFT rows whose account_id is None.
+
+        Roster and DEBUG_BUILD Drafts are never rebased.
+        """
         return cls._get_or_create_agent_draft(
             session=session,
             tenant_id=tenant_id,
@@ -1767,6 +2096,8 @@ class AgentComposerService:
         snapshot: AgentConfigSnapshot,
         updated_by: str | None,
     ) -> bool:
+        """Sync a stale normal Draft's base_snapshot_id, home_snapshot_id, config_snapshot, and updated_by."""
+
         if (
             agent.scope != AgentScope.WORKFLOW_ONLY
             or draft.draft_type != AgentConfigDraftType.DRAFT
@@ -1777,6 +2108,7 @@ class AgentComposerService:
         ):
             return False
         draft.base_snapshot_id = snapshot.id
+        draft.home_snapshot_id = snapshot.home_snapshot_id
         draft.config_snapshot = AgentSoulConfig.model_validate(snapshot.config_snapshot_dict)
         draft.updated_by = updated_by
         return True
@@ -1813,7 +2145,9 @@ class AgentComposerService:
                     agent_id=agent.id,
                     version_id=agent.active_config_snapshot_id,
                 )
-                if active_snapshot is not None and cls._rebase_workflow_only_normal_draft(
+                if active_snapshot is None:
+                    raise AgentVersionNotFoundError()
+                if cls._rebase_workflow_only_normal_draft(
                     agent=agent,
                     draft=draft,
                     snapshot=active_snapshot,
@@ -1827,18 +2161,17 @@ class AgentComposerService:
             agent_id=agent.id,
             version_id=agent.active_config_snapshot_id,
         )
-        agent_soul = (
-            AgentSoulConfig.model_validate(base_snapshot.config_snapshot_dict)
-            if base_snapshot is not None
-            else AgentSoulConfig()
-        )
+        if base_snapshot is None:
+            raise AgentVersionNotFoundError()
+        agent_soul = AgentSoulConfig.model_validate(base_snapshot.config_snapshot_dict)
         draft = AgentConfigDraft(
             tenant_id=tenant_id,
             agent_id=agent.id,
             draft_type=draft_type,
             account_id=account_id if draft_type == AgentConfigDraftType.DEBUG_BUILD else None,
             draft_owner_key=account_id if draft_type == AgentConfigDraftType.DEBUG_BUILD and account_id else "",
-            base_snapshot_id=base_snapshot.id if base_snapshot else None,
+            base_snapshot_id=base_snapshot.id,
+            home_snapshot_id=base_snapshot.home_snapshot_id,
             config_snapshot=agent_soul,
             created_by=created_by,
             updated_by=created_by,
@@ -2140,7 +2473,7 @@ class AgentComposerService:
 
         from services.agent.roster_service import AgentRosterService
 
-        return AgentRosterService(session).get_or_create_agent_app_debug_conversation_id(
+        return AgentRosterService(session).get_or_create_build_conversation(
             tenant_id=tenant_id,
             agent_id=agent.id,
             account_id=account_id,

api/services/agent/dsl_service.py

@@ -49,6 +49,7 @@ from services.agent.dsl_entities import (
     make_portable_agent_package,
     portable_ref,
 )
+from services.agent.home_snapshot_service import AgentHomeSnapshotService
 from services.agent.knowledge_datasets import get_tenant_knowledge_dataset_rows
 from services.agent.roster_service import AgentRosterService
 from services.entities.dsl_entities import DslImportWarning
@@ -224,6 +225,7 @@ class AgentDslService:
                 account_id=None,
                 draft_owner_key="",
                 base_snapshot_id=snapshot.id,
+                home_snapshot_id=snapshot.home_snapshot_id,
                 config_snapshot=soul,
                 created_by=account.id,
                 updated_by=account.id,
@@ -243,7 +245,7 @@ class AgentDslService:
         portable_graph: Mapping[str, Any],
         raw_packages: Mapping[str, Any],
         account: Account,
-    ) -> tuple[dict[str, Any], list[DslImportWarning]]:
+    ) -> tuple[dict[str, Any], list[DslImportWarning], set[str]]:
         """Materialize every packaged Agent as a node-owned inline Agent."""
 
         graph = copy.deepcopy(dict(portable_graph))
@@ -256,6 +258,11 @@ class AgentDslService:
                 WorkflowAgentNodeBinding.workflow_version == Workflow.VERSION_DRAFT,
             )
         ).all()
+        retirement_candidates = {
+            binding.agent_id
+            for binding in previous_bindings
+            if binding.binding_type == WorkflowAgentBindingType.INLINE_AGENT and binding.agent_id
+        }
         for binding in previous_bindings:
             self.session.delete(binding)
         self.session.flush()
@@ -312,7 +319,7 @@ class AgentDslService:
 
         workflow.graph = json.dumps(graph)
         self.session.flush()
-        return graph, warnings
+        return graph, warnings, retirement_candidates
 
     def clone_inline_binding_for_node(
         self,
@@ -562,11 +569,17 @@ class AgentDslService:
             )
             or 0
         ) + 1
+        home_snapshot = AgentHomeSnapshotService.create_initial(
+            session=self.session,
+            tenant_id=tenant_id,
+            agent_id=agent.id,
+        )
         snapshot = AgentConfigSnapshot(
             tenant_id=tenant_id,
             agent_id=agent.id,
             version=next_version,
             config_snapshot=soul,
+            home_snapshot_id=home_snapshot.id,
             created_by=account_id,
         )
         self.session.add(snapshot)

api/services/agent/errors.py

@@ -29,6 +29,12 @@ class AgentModelNotConfiguredError(BaseHTTPException):
     code = 400
 
 
+class AgentBuildSandboxNotFoundError(BaseHTTPException):
+    error_code = "agent_build_sandbox_not_found"
+    description = "The retained Build Sandbox is no longer available."
+    code = 404
+
+
 class AgentSoulLockedError(BadRequest):
     description = "Agent Soul is locked for this workflow node."
 

api/services/agent/home_snapshot_service.py

@@ -0,0 +1,238 @@
+"""Own immutable Agent Home Snapshot ledger rows and physical collection."""
+
+from __future__ import annotations
+
+import logging
+
+from dify_agent.client import Client, DifyAgentNotFoundError
+from dify_agent.protocol import CreateHomeSnapshotFromBindingRequest, InitializeHomeSnapshotRequest
+from sqlalchemy import select
+from sqlalchemy.orm import Session
+
+from configs import dify_config
+from core.db.session_factory import session_factory
+from libs.datetime_utils import naive_utc_now
+from libs.uuid_utils import uuidv7
+from models.agent import (
+    Agent,
+    AgentConfigDraft,
+    AgentConfigSnapshot,
+    AgentConfigVersionKind,
+    AgentHomeSnapshot,
+    AgentStatus,
+    AgentWorkingResourceStatus,
+    AgentWorkspaceOwnerType,
+)
+from services.agent.errors import AgentBuildSandboxNotFoundError
+from services.agent.workspace_service import AgentWorkspaceService, WorkspaceOwnerScope
+
+logger = logging.getLogger(__name__)
+
+
+class AgentHomeSnapshotUnavailableError(RuntimeError):
+    """The requested owner-scoped Home Snapshot cannot be used."""
+
+
+class AgentHomeSnapshotService:
+    """Create, retire, and collect Agent-owned immutable Home Snapshots."""
+
+    @classmethod
+    def create_initial(
+        cls,
+        *,
+        session: Session,
+        tenant_id: str,
+        agent_id: str,
+    ) -> AgentHomeSnapshot:
+        home_snapshot_id = str(uuidv7())
+        with cls._client() as client:
+            response = client.initialize_home_snapshot_sync(
+                InitializeHomeSnapshotRequest(
+                    tenant_id=tenant_id,
+                    agent_id=agent_id,
+                    home_snapshot_id=home_snapshot_id,
+                )
+            )
+        home_snapshot = AgentHomeSnapshot(
+            id=home_snapshot_id,
+            tenant_id=tenant_id,
+            agent_id=agent_id,
+            snapshot_ref=response.snapshot_ref,
+            status=AgentWorkingResourceStatus.ACTIVE,
+        )
+        session.add(home_snapshot)
+        session.flush()
+        return home_snapshot
+
+    @classmethod
+    def create_for_build_apply(
+        cls,
+        *,
+        session: Session,
+        build_draft: AgentConfigDraft,
+    ) -> AgentHomeSnapshot:
+        """Checkpoint the exact participant owned by ``build_draft``."""
+
+        source_binding_id = build_draft.agent_workspace_binding_id
+        if source_binding_id is None:
+            raise AgentBuildSandboxNotFoundError()
+        agent = session.scalar(
+            select(Agent).where(
+                Agent.id == build_draft.agent_id,
+                Agent.tenant_id == build_draft.tenant_id,
+            )
+        )
+        if agent is None:
+            raise AgentBuildSandboxNotFoundError()
+        from services.agent.roster_service import AgentRosterService
+
+        runtime_app_id = AgentRosterService.runtime_backing_app_id(agent)
+        if runtime_app_id is None:
+            raise AgentBuildSandboxNotFoundError()
+        binding = AgentWorkspaceService.get_active_binding(
+            session=session,
+            tenant_id=build_draft.tenant_id,
+            binding_id=source_binding_id,
+            expected_owner_scope=WorkspaceOwnerScope(
+                tenant_id=build_draft.tenant_id,
+                app_id=runtime_app_id,
+                owner_type=AgentWorkspaceOwnerType.BUILD_DRAFT,
+                owner_id=build_draft.id,
+            ),
+        )
+        if binding is None or binding.agent_id != build_draft.agent_id:
+            raise AgentBuildSandboxNotFoundError()
+        AgentWorkspaceService.validate_binding_generation(
+            binding,
+            base_home_snapshot_id=build_draft.home_snapshot_id,
+            agent_config_version_id=build_draft.id,
+            agent_config_version_kind=AgentConfigVersionKind.BUILD_DRAFT,
+        )
+
+        home_snapshot_id = str(uuidv7())
+        try:
+            with cls._client() as client:
+                response = client.create_home_snapshot_from_binding_sync(
+                    CreateHomeSnapshotFromBindingRequest(
+                        tenant_id=build_draft.tenant_id,
+                        agent_id=build_draft.agent_id,
+                        home_snapshot_id=home_snapshot_id,
+                        backend_binding_ref=binding.backend_binding_ref,
+                    )
+                )
+        except DifyAgentNotFoundError as exc:
+            raise AgentBuildSandboxNotFoundError() from exc
+
+        home_snapshot = AgentHomeSnapshot(
+            id=home_snapshot_id,
+            tenant_id=build_draft.tenant_id,
+            agent_id=build_draft.agent_id,
+            snapshot_ref=response.snapshot_ref,
+            status=AgentWorkingResourceStatus.ACTIVE,
+        )
+        session.add(home_snapshot)
+        return home_snapshot
+
+    @classmethod
+    def retire_all_for_agent(cls, *, session: Session, tenant_id: str, agent_id: str) -> list[str]:
+        rows = session.scalars(
+            select(AgentHomeSnapshot).where(
+                AgentHomeSnapshot.tenant_id == tenant_id,
+                AgentHomeSnapshot.agent_id == agent_id,
+                AgentHomeSnapshot.status == AgentWorkingResourceStatus.ACTIVE,
+            )
+        ).all()
+        now = naive_utc_now()
+        for row in rows:
+            row.status = AgentWorkingResourceStatus.RETIRED
+            row.retired_at = now
+        return [row.id for row in rows]
+
+    @classmethod
+    def collect_retired_home_snapshot(cls, *, tenant_id: str, home_snapshot_id: str) -> None:
+        try:
+            cls._collect_retired_home_snapshot(tenant_id=tenant_id, home_snapshot_id=home_snapshot_id)
+        except Exception:
+            logger.exception(
+                "Failed to collect retired Agent Home Snapshot",
+                extra={"tenant_id": tenant_id, "home_snapshot_id": home_snapshot_id},
+            )
+
+    @classmethod
+    def _collect_retired_home_snapshot(cls, *, tenant_id: str, home_snapshot_id: str) -> None:
+        with session_factory.create_session() as session:
+            snapshot = session.scalar(
+                select(AgentHomeSnapshot).where(
+                    AgentHomeSnapshot.id == home_snapshot_id,
+                    AgentHomeSnapshot.tenant_id == tenant_id,
+                    AgentHomeSnapshot.status == AgentWorkingResourceStatus.RETIRED,
+                )
+            )
+            if snapshot is None:
+                return
+            referenced = session.scalar(
+                select(AgentConfigDraft.id).where(AgentConfigDraft.home_snapshot_id == home_snapshot_id).limit(1)
+            ) or session.scalar(
+                select(AgentConfigSnapshot.id).where(AgentConfigSnapshot.home_snapshot_id == home_snapshot_id).limit(1)
+            )
+            if referenced is not None:
+                return
+            snapshot_ref = snapshot.snapshot_ref
+        try:
+            cls.delete(snapshot_ref=snapshot_ref)
+        except Exception:
+            logger.exception(
+                "Failed to collect retired Agent Home Snapshot",
+                extra={"tenant_id": tenant_id, "home_snapshot_id": home_snapshot_id},
+            )
+            return
+        with session_factory.create_session() as session:
+            snapshot = session.scalar(
+                select(AgentHomeSnapshot).where(
+                    AgentHomeSnapshot.id == home_snapshot_id,
+                    AgentHomeSnapshot.tenant_id == tenant_id,
+                    AgentHomeSnapshot.status == AgentWorkingResourceStatus.RETIRED,
+                )
+            )
+            if snapshot is not None:
+                session.delete(snapshot)
+                session.commit()
+
+    @classmethod
+    def delete(cls, *, snapshot_ref: str) -> None:
+        with cls._client() as client:
+            client.delete_home_snapshot_sync(snapshot_ref)
+
+    @staticmethod
+    def _client() -> Client:
+        base_url = dify_config.AGENT_BACKEND_BASE_URL
+        if not base_url:
+            raise AgentHomeSnapshotUnavailableError("Dify Agent backend is required for Home Snapshot operations")
+        return Client(base_url=base_url)
+
+
+def validate_home_snapshot_binding(*, session: Session, agent: Agent, home_snapshot_id: str) -> None:
+    _require_owned_home_snapshot(session=session, agent=agent, home_snapshot_id=home_snapshot_id)
+
+
+def _require_owned_home_snapshot(*, session: Session, agent: Agent, home_snapshot_id: str) -> AgentHomeSnapshot:
+    if agent.status != AgentStatus.ACTIVE:
+        raise AgentHomeSnapshotUnavailableError(f"Agent {agent.id} is not active")
+    home_snapshot = session.scalar(
+        select(AgentHomeSnapshot).where(
+            AgentHomeSnapshot.id == home_snapshot_id,
+            AgentHomeSnapshot.tenant_id == agent.tenant_id,
+            AgentHomeSnapshot.agent_id == agent.id,
+            AgentHomeSnapshot.status == AgentWorkingResourceStatus.ACTIVE,
+        )
+    )
+    if home_snapshot is None:
+        raise AgentHomeSnapshotUnavailableError(f"Home Snapshot {home_snapshot_id} is unavailable for Agent {agent.id}")
+    return home_snapshot
+
+
+__all__ = [
+    "AgentHomeSnapshotService",
+    "AgentHomeSnapshotUnavailableError",
+    "validate_home_snapshot_binding",
+]

api/services/agent/retirement_service.py

@@ -0,0 +1,166 @@
+"""Workflow-only Agent ownership retirement after product transactions commit."""
+
+from __future__ import annotations
+
+import logging
+from collections.abc import Iterable
+
+from sqlalchemy import or_, select
+from sqlalchemy.orm import Session
+
+from core.db.session_factory import session_factory
+from libs.datetime_utils import naive_utc_now
+from models.agent import (
+    Agent,
+    AgentScope,
+    AgentStatus,
+    AgentWorkingResourceStatus,
+    AgentWorkspaceBinding,
+    WorkflowAgentNodeBinding,
+)
+from models.enums import AppStatus
+from models.model import App
+from models.workflow import Workflow
+from services.agent.home_snapshot_service import AgentHomeSnapshotService
+from services.agent.workspace_service import AgentWorkspaceService
+
+logger = logging.getLogger(__name__)
+
+
+class WorkflowAgentRetirementService:
+    """Archive workflow-only Agents once no effective binding owns them."""
+
+    @classmethod
+    def retire_unowned(
+        cls,
+        *,
+        tenant_id: str,
+        agent_ids: Iterable[str],
+        account_id: str | None,
+    ) -> tuple[list[str], list[str]]:
+        """Re-check ownership, archive orphans, and commit their resource retirement."""
+
+        candidates = tuple(sorted({agent_id for agent_id in agent_ids if agent_id}))
+        if not candidates:
+            return [], []
+        retired_bindings: list[str] = []
+        retired_snapshots: list[str] = []
+        try:
+            with session_factory.create_session() as session:
+                retired_agent_ids = cls.archive_unowned(
+                    session=session,
+                    tenant_id=tenant_id,
+                    agent_ids=candidates,
+                    account_id=account_id,
+                )
+                for agent_id in retired_agent_ids:
+                    bindings = session.scalars(
+                        select(AgentWorkspaceBinding).where(
+                            AgentWorkspaceBinding.tenant_id == tenant_id,
+                            AgentWorkspaceBinding.agent_id == agent_id,
+                            AgentWorkspaceBinding.status == AgentWorkingResourceStatus.ACTIVE,
+                        )
+                    ).all()
+                    for binding in bindings:
+                        binding_id = AgentWorkspaceService.retire_binding(
+                            session=session,
+                            tenant_id=tenant_id,
+                            binding_id=binding.id,
+                        )
+                        if binding_id is not None:
+                            retired_bindings.append(binding_id)
+                    retired_snapshots.extend(
+                        AgentHomeSnapshotService.retire_all_for_agent(
+                            session=session,
+                            tenant_id=tenant_id,
+                            agent_id=agent_id,
+                        )
+                    )
+                session.commit()
+        except Exception:
+            logger.exception(
+                "Failed to retire unowned Workflow Agents",
+                extra={
+                    "tenant_id": tenant_id,
+                    "agent_ids": candidates,
+                },
+            )
+            return [], []
+        return retired_bindings, retired_snapshots
+
+    @classmethod
+    def archive_unowned(
+        cls,
+        *,
+        session: Session,
+        tenant_id: str,
+        agent_ids: Iterable[str],
+        account_id: str | None,
+    ) -> list[str]:
+        """Archive active orphans and return every orphan eligible for Home cleanup."""
+        candidates = tuple(sorted({agent_id for agent_id in agent_ids if agent_id}))
+        if not candidates:
+            return []
+        agents = session.scalars(
+            select(Agent).where(
+                Agent.tenant_id == tenant_id,
+                Agent.id.in_(candidates),
+                Agent.scope == AgentScope.WORKFLOW_ONLY,
+                Agent.status.in_((AgentStatus.ACTIVE, AgentStatus.ARCHIVED)),
+            )
+        ).all()
+        effective_agent_ids = cls._effective_agent_ids(
+            session=session,
+            tenant_id=tenant_id,
+            agent_ids=[agent.id for agent in agents],
+        )
+        now = naive_utc_now()
+        cleanup_candidates: list[str] = []
+        for agent in agents:
+            if agent.id in effective_agent_ids:
+                continue
+            if agent.status == AgentStatus.ACTIVE:
+                agent.status = AgentStatus.ARCHIVED
+                agent.archived_by = account_id
+                agent.archived_at = now
+                agent.updated_by = account_id or agent.updated_by
+                agent.updated_at = now
+            cleanup_candidates.append(agent.id)
+        session.flush()
+        return cleanup_candidates
+
+    @staticmethod
+    def _effective_agent_ids(
+        *,
+        session: Session,
+        tenant_id: str,
+        agent_ids: list[str],
+    ) -> set[str]:
+        if not agent_ids:
+            return set()
+        values = session.scalars(
+            select(WorkflowAgentNodeBinding.agent_id)
+            .join(
+                Workflow,
+                Workflow.id == WorkflowAgentNodeBinding.workflow_id,
+            )
+            .join(App, App.id == WorkflowAgentNodeBinding.app_id)
+            .where(
+                WorkflowAgentNodeBinding.tenant_id == tenant_id,
+                WorkflowAgentNodeBinding.agent_id.in_(agent_ids),
+                Workflow.tenant_id == tenant_id,
+                Workflow.app_id == WorkflowAgentNodeBinding.app_id,
+                Workflow.version == WorkflowAgentNodeBinding.workflow_version,
+                App.tenant_id == tenant_id,
+                App.status == AppStatus.NORMAL,
+                or_(
+                    Workflow.version == Workflow.VERSION_DRAFT,
+                    App.workflow_id == Workflow.id,
+                ),
+            )
+            .distinct()
+        ).all()
+        return {agent_id for agent_id in values if agent_id}
+
+
+__all__ = ["WorkflowAgentRetirementService"]

api/services/agent/roster_service.py

@@ -4,10 +4,8 @@ from typing import Any, TypedDict
 from sqlalchemy import and_, func, or_, select
 from sqlalchemy.exc import IntegrityError
 
-from clients.agent_backend.session_cleanup import AgentBackendSessionCleanupPayload
 from constants.model_template import default_app_templates
 from core.agent.publish_visibility import workflow_callable_active_snapshot_filter
-from core.app.apps.agent_app.session_store import AgentAppRuntimeSessionStore
 from core.app.entities.app_invoke_entities import InvokeFrom
 from libs.datetime_utils import naive_utc_now
 from libs.helper import to_timestamp
@@ -25,6 +23,9 @@ from models.agent import (
     AgentScope,
     AgentSource,
     AgentStatus,
+    AgentWorkingResourceStatus,
+    AgentWorkspaceBinding,
+    AgentWorkspaceOwnerType,
     WorkflowAgentBindingType,
     WorkflowAgentNodeBinding,
 )
@@ -36,15 +37,18 @@ from services.agent.agent_soul_state import agent_soul_has_model
 from services.agent.composer_validator import ComposerConfigValidator
 from services.agent.errors import (
     AgentArchivedError,
+    AgentBuildSandboxNotFoundError,
     AgentNameConflictError,
     AgentNotFoundError,
     AgentVersionNotFoundError,
 )
+from services.agent.home_snapshot_service import AgentHomeSnapshotService
+from services.agent.workspace_service import AgentWorkspaceNotFoundError, AgentWorkspaceService, WorkspaceOwnerScope
 from services.app_service import AppService, CreateAppParams
 from services.enterprise.enterprise_service import EnterpriseService
 from services.entities.agent_entities import RosterAgentCreatePayload, RosterAgentUpdatePayload
 from services.feature_service import FeatureService
-from tasks.agent_backend_session_cleanup_task import cleanup_conversation_agent_runtime_session
+from tasks.collect_agent_resources_task import enqueue_agent_resource_collection
 
 logger = logging.getLogger(__name__)
 
@@ -304,6 +308,27 @@ class AgentRosterService:
         account_id: str,
         payload: RosterAgentCreatePayload,
         source: AgentSource = AgentSource.ROSTER,
+    ) -> Agent:
+        try:
+            agent = self._create_roster_agent_in_transaction(
+                tenant_id=tenant_id,
+                account_id=account_id,
+                payload=payload,
+                source=source,
+            )
+            self._session.commit()
+            return agent
+        except IntegrityError as exc:
+            self._session.rollback()
+            raise AgentNameConflictError() from exc
+
+    def _create_roster_agent_in_transaction(
+        self,
+        *,
+        tenant_id: str,
+        account_id: str,
+        payload: RosterAgentCreatePayload,
+        source: AgentSource,
     ) -> Agent:
         ComposerConfigValidator.validate_agent_soul(payload.agent_soul)
 
@@ -323,17 +348,19 @@ class AgentRosterService:
             updated_by=account_id,
         )
         self._session.add(agent)
-        try:
-            self._session.flush()
-        except IntegrityError as exc:
-            self._session.rollback()
-            raise AgentNameConflictError() from exc
+        self._session.flush()
 
+        home_snapshot = AgentHomeSnapshotService.create_initial(
+            session=self._session,
+            tenant_id=tenant_id,
+            agent_id=agent.id,
+        )
         version = AgentConfigSnapshot(
             tenant_id=tenant_id,
             agent_id=agent.id,
             version=1,
             config_snapshot=payload.agent_soul,
+            home_snapshot_id=home_snapshot.id,
             version_note=payload.version_note,
             created_by=account_id,
         )
@@ -354,11 +381,6 @@ class AgentRosterService:
         agent.active_config_has_model = agent_soul_has_model(payload.agent_soul)
         agent.active_config_is_published = True
 
-        try:
-            self._session.commit()
-        except IntegrityError as exc:
-            self._session.rollback()
-            raise AgentNameConflictError() from exc
         return agent
 
     def create_backing_agent_for_app(
@@ -405,17 +427,19 @@ class AgentRosterService:
             updated_by=account_id,
         )
         self._session.add(agent)
-        try:
-            self._session.flush()
-        except IntegrityError as exc:
-            self._session.rollback()
-            raise AgentNameConflictError() from exc
+        self._session.flush()
 
+        home_snapshot = AgentHomeSnapshotService.create_initial(
+            session=self._session,
+            tenant_id=tenant_id,
+            agent_id=agent.id,
+        )
         version = AgentConfigSnapshot(
             tenant_id=tenant_id,
             agent_id=agent.id,
             version=1,
             config_snapshot=soul,
+            home_snapshot_id=home_snapshot.id,
             created_by=account_id,
         )
         self._session.add(version)
@@ -590,16 +614,15 @@ class AgentRosterService:
         self._session.flush()
         return conversation_id
 
-    def get_or_create_agent_app_debug_conversation_id(
+    def get_or_create_build_conversation(
         self,
         *,
         tenant_id: str,
         agent_id: str,
         account_id: str,
-        draft_type: AgentConfigDraftType = AgentConfigDraftType.DEBUG_BUILD,
         commit: bool = True,
     ) -> str:
-        """Return the current editor's Build or Preview conversation for an Agent App."""
+        """Return the current editor's stable Build conversation."""
 
         agent = self._session.scalar(
             select(Agent).where(
@@ -614,21 +637,20 @@ class AgentRosterService:
         conversation_id = self._get_or_create_agent_app_debug_conversation(
             agent=agent,
             account_id=account_id,
-            draft_type=draft_type,
+            draft_type=AgentConfigDraftType.DEBUG_BUILD,
         )
         if commit:
             self._session.commit()
         return conversation_id
 
-    def load_agent_app_debug_conversation_id(
+    def get_current_preview_conversation(
         self,
         *,
         tenant_id: str,
         agent_id: str,
         account_id: str,
-        draft_type: AgentConfigDraftType = AgentConfigDraftType.DEBUG_BUILD,
     ) -> str | None:
-        """Return the editor's existing scoped conversation without creating or repairing rows."""
+        """Return the editor's current Preview conversation without creating one."""
 
         return self._session.scalar(
             select(Conversation.id)
@@ -637,7 +659,7 @@ class AgentRosterService:
                 AgentDebugConversation.tenant_id == tenant_id,
                 AgentDebugConversation.agent_id == agent_id,
                 AgentDebugConversation.account_id == account_id,
-                AgentDebugConversation.draft_type == draft_type,
+                AgentDebugConversation.draft_type == AgentConfigDraftType.DRAFT,
                 AgentDebugConversation.app_id == Conversation.app_id,
                 Conversation.from_source == ConversationFromSource.CONSOLE,
                 Conversation.from_account_id == account_id,
@@ -657,25 +679,12 @@ class AgentRosterService:
             or 0
         )
 
-    def refresh_agent_app_debug_conversation_id(
-        self,
-        *,
-        tenant_id: str,
-        agent_id: str,
-        account_id: str,
-        draft_type: AgentConfigDraftType = AgentConfigDraftType.DEBUG_BUILD,
-    ) -> str:
-        """Start a new scoped console conversation for the current Agent App editor.
+    def rotate_preview_conversation(self, *, tenant_id: str, agent_id: str, account_id: str) -> str:
+        """Rotate Preview and retire its exact Conversation-owned Binding.
 
-        If this account already has a mapping for the requested draft surface, the previous
-        conversation is abandoned after the replacement mapping is committed: any ACTIVE
-        conversation-owned Agent runtime sessions for that old conversation are sent through
-        best-effort backend cleanup and then retired locally even when enqueueing fails. This
-        order prevents a failed database commit from retiring the still-current runtime session.
-        The other draft surface is left untouched.
-
-        A user and draft surface own one current mapping. If new-conversation requests overlap,
-        the last committed rotation becomes current and earlier response IDs cannot be continued.
+        The mapping update and exact CONVERSATION Binding retirement commit in
+        one transaction. Validation failures fail fast; collection is enqueued
+        only after commit.
         """
 
         agent = self._session.scalar(
@@ -694,142 +703,194 @@ class AgentRosterService:
         if not backing_app_id:
             raise AgentNotFoundError()
 
-        conversation_id = self._create_agent_app_debug_conversation(
-            app_id=backing_app_id,
-            account_id=account_id,
-        )
-        previous_conversation: tuple[str, str] | None = None
-        mapping = self._session.scalar(
-            select(AgentDebugConversation).where(
-                AgentDebugConversation.tenant_id == tenant_id,
-                AgentDebugConversation.agent_id == agent_id,
-                AgentDebugConversation.account_id == account_id,
-                AgentDebugConversation.draft_type == draft_type,
+        retired_binding_id: str | None = None
+        try:
+            conversation_id = self._create_agent_app_debug_conversation(
+                app_id=backing_app_id,
+                account_id=account_id,
             )
-        )
-        if mapping is None:
-            self._session.add(
-                AgentDebugConversation(
-                    tenant_id=tenant_id,
-                    agent_id=agent_id,
-                    app_id=backing_app_id,
-                    account_id=account_id,
-                    draft_type=draft_type,
-                    conversation_id=conversation_id,
+            mapping = self._session.scalar(
+                select(AgentDebugConversation).where(
+                    AgentDebugConversation.tenant_id == tenant_id,
+                    AgentDebugConversation.agent_id == agent_id,
+                    AgentDebugConversation.account_id == account_id,
+                    AgentDebugConversation.draft_type == AgentConfigDraftType.DRAFT,
                 )
             )
-        else:
-            previous_app_id = mapping.app_id
-            previous_conversation_id = mapping.conversation_id
-            if previous_conversation_id:
-                previous_conversation = (previous_app_id or backing_app_id, previous_conversation_id)
-            mapping.app_id = backing_app_id
-            mapping.conversation_id = conversation_id
-        self._session.flush()
-        self._session.commit()
-
-        if previous_conversation:
-            previous_app_id, previous_conversation_id = previous_conversation
-            self._cleanup_debug_conversation_runtime_sessions(
+            if mapping is None:
+                self._session.add(
+                    AgentDebugConversation(
+                        tenant_id=tenant_id,
+                        agent_id=agent_id,
+                        app_id=backing_app_id,
+                        account_id=account_id,
+                        draft_type=AgentConfigDraftType.DRAFT,
+                        conversation_id=conversation_id,
+                    )
+                )
+            else:
+                previous_app_id = mapping.app_id or backing_app_id
+                previous_conversation_id = mapping.conversation_id
+                if previous_conversation_id:
+                    previous_conversation = self._session.scalar(
+                        select(Conversation).where(
+                            Conversation.id == previous_conversation_id,
+                            Conversation.app_id == previous_app_id,
+                            Conversation.from_source == ConversationFromSource.CONSOLE,
+                            Conversation.from_account_id == account_id,
+                            Conversation.is_deleted.is_(False),
+                        )
+                    )
+                    if (
+                        previous_conversation is not None
+                        and previous_conversation.agent_workspace_binding_id is not None
+                    ):
+                        binding_id = previous_conversation.agent_workspace_binding_id
+                        binding = AgentWorkspaceService.get_active_binding(
+                            session=self._session,
+                            tenant_id=tenant_id,
+                            binding_id=binding_id,
+                            expected_owner_scope=WorkspaceOwnerScope(
+                                tenant_id=tenant_id,
+                                app_id=previous_app_id,
+                                owner_type=AgentWorkspaceOwnerType.CONVERSATION,
+                                owner_id=previous_conversation.id,
+                            ),
+                        )
+                        if binding is None or binding.agent_id != agent_id:
+                            raise AgentWorkspaceNotFoundError(
+                                "Agent debug Conversation participant Binding is unavailable"
+                            )
+                        retired_binding_id = AgentWorkspaceService.retire_binding(
+                            session=self._session,
+                            tenant_id=tenant_id,
+                            binding_id=binding_id,
+                        )
+                        if retired_binding_id is None:
+                            raise AgentWorkspaceNotFoundError(
+                                "Agent debug Conversation participant Binding is unavailable"
+                            )
+                mapping.app_id = backing_app_id
+                mapping.conversation_id = conversation_id
+            self._session.flush()
+            self._session.commit()
+        except Exception:
+            self._session.rollback()
+            raise
+        if retired_binding_id is not None:
+            enqueue_agent_resource_collection(
                 tenant_id=tenant_id,
-                agent_id=agent_id,
-                account_id=account_id,
-                draft_type=draft_type,
-                app_id=previous_app_id,
-                conversation_id=previous_conversation_id,
+                binding_ids=(retired_binding_id,),
             )
         return conversation_id
 
-    def _cleanup_debug_conversation_runtime_sessions(
-        self,
-        *,
-        tenant_id: str,
-        agent_id: str,
-        account_id: str,
-        draft_type: AgentConfigDraftType,
-        app_id: str,
-        conversation_id: str,
-    ) -> None:
+    def reset_build_conversation(self, *, tenant_id: str, agent_id: str, account_id: str) -> str:
+        """Reset Build and retire its exact DEBUG_BUILD Draft-owned Binding.
+
+        The mapping update, exact BUILD_DRAFT Binding retirement, and Draft
+        pointer clear commit in one transaction. Validation failures fail fast;
+        collection is enqueued only after commit.
+        """
+
+        agent = self._session.scalar(
+            select(Agent).where(
+                Agent.tenant_id == tenant_id,
+                Agent.id == agent_id,
+                Agent.status == AgentStatus.ACTIVE,
+            )
+        )
+        if agent is None:
+            raise AgentNotFoundError()
+        backing_app_id = self._ensure_workflow_agent_backing_app(
+            agent=agent,
+            account_id=agent.updated_by or agent.created_by,
+        )
+        if not backing_app_id:
+            raise AgentNotFoundError()
+
+        retired_binding_id: str | None = None
         try:
-            session_store = AgentAppRuntimeSessionStore()
-            stored_sessions = session_store.list_active_sessions_for_conversation(
-                tenant_id=tenant_id,
-                app_id=app_id,
-                conversation_id=conversation_id,
+            conversation_id = self._create_agent_app_debug_conversation(
+                app_id=backing_app_id,
+                account_id=account_id,
             )
-        except Exception:
-            logger.warning(
-                "Failed to load Agent App runtime sessions for debug conversation refresh: "
-                "tenant_id=%s app_id=%s conversation_id=%s",
-                tenant_id,
-                app_id,
-                conversation_id,
-                exc_info=True,
-            )
-            return
-
-        for stored_session in stored_sessions:
-            try:
-                if stored_session.runtime_layer_specs:
-                    payload = AgentBackendSessionCleanupPayload(
-                        session_snapshot=stored_session.session_snapshot,
-                        runtime_layer_specs=stored_session.runtime_layer_specs,
-                        idempotency_key=(
-                            f"{tenant_id}:{agent_id}:{account_id}:{draft_type.value}:{conversation_id}:"
-                            "debug-session-cleanup:"
-                            f"{stored_session.scope.agent_id}:"
-                            f"{stored_session.scope.agent_config_snapshot_id or 'no-config'}:"
-                            f"{stored_session.backend_run_id or 'no-run'}"
-                        ),
-                        metadata={
-                            "tenant_id": stored_session.scope.tenant_id,
-                            "app_id": stored_session.scope.app_id,
-                            "conversation_id": stored_session.scope.conversation_id,
-                            "agent_id": stored_session.scope.agent_id,
-                            "agent_config_snapshot_id": stored_session.scope.agent_config_snapshot_id,
-                            "draft_type": draft_type.value,
-                            "previous_agent_backend_run_id": stored_session.backend_run_id,
-                        },
-                    )
-                    cleanup_conversation_agent_runtime_session.delay(payload.model_dump(mode="json"))
-            except Exception:
-                logger.warning(
-                    "Failed to enqueue Agent backend cleanup for debug conversation refresh: "
-                    "tenant_id=%s app_id=%s conversation_id=%s agent_id=%s backend_run_id=%s",
-                    stored_session.scope.tenant_id,
-                    stored_session.scope.app_id,
-                    stored_session.scope.conversation_id,
-                    stored_session.scope.agent_id,
-                    stored_session.backend_run_id,
-                    exc_info=True,
+            mapping = self._session.scalar(
+                select(AgentDebugConversation).where(
+                    AgentDebugConversation.tenant_id == tenant_id,
+                    AgentDebugConversation.agent_id == agent_id,
+                    AgentDebugConversation.account_id == account_id,
+                    AgentDebugConversation.draft_type == AgentConfigDraftType.DEBUG_BUILD,
                 )
-            finally:
-                try:
-                    session_store.mark_cleaned(
-                        scope=stored_session.scope,
-                        backend_run_id=stored_session.backend_run_id,
-                    )
-                except Exception:
-                    logger.warning(
-                        "Failed to retire Agent App runtime session for debug conversation refresh: "
-                        "tenant_id=%s app_id=%s conversation_id=%s agent_id=%s backend_run_id=%s",
-                        stored_session.scope.tenant_id,
-                        stored_session.scope.app_id,
-                        stored_session.scope.conversation_id,
-                        stored_session.scope.agent_id,
-                        stored_session.backend_run_id,
-                        exc_info=True,
-                    )
+            )
+            build_draft = self._session.scalar(
+                select(AgentConfigDraft)
+                .where(
+                    AgentConfigDraft.tenant_id == tenant_id,
+                    AgentConfigDraft.agent_id == agent_id,
+                    AgentConfigDraft.draft_type == AgentConfigDraftType.DEBUG_BUILD,
+                    AgentConfigDraft.account_id == account_id,
+                )
+                .order_by(AgentConfigDraft.updated_at.desc())
+                .limit(1)
+            )
+            if build_draft is not None and build_draft.agent_workspace_binding_id is not None:
+                binding_id = build_draft.agent_workspace_binding_id
+                binding = AgentWorkspaceService.get_active_binding(
+                    session=self._session,
+                    tenant_id=tenant_id,
+                    binding_id=binding_id,
+                    expected_owner_scope=WorkspaceOwnerScope(
+                        tenant_id=tenant_id,
+                        app_id=backing_app_id,
+                        owner_type=AgentWorkspaceOwnerType.BUILD_DRAFT,
+                        owner_id=build_draft.id,
+                    ),
+                )
+                if binding is None or binding.agent_id != agent_id:
+                    raise AgentBuildSandboxNotFoundError()
+                retired_binding_id = AgentWorkspaceService.retire_binding(
+                    session=self._session,
+                    tenant_id=tenant_id,
+                    binding_id=binding_id,
+                )
+                if retired_binding_id is None:
+                    raise AgentBuildSandboxNotFoundError()
+                build_draft.agent_workspace_binding_id = None
 
-    def load_or_create_agent_app_debug_conversation_ids_by_agent_id(
+            if mapping is None:
+                self._session.add(
+                    AgentDebugConversation(
+                        tenant_id=tenant_id,
+                        agent_id=agent_id,
+                        app_id=backing_app_id,
+                        account_id=account_id,
+                        draft_type=AgentConfigDraftType.DEBUG_BUILD,
+                        conversation_id=conversation_id,
+                    )
+                )
+            else:
+                mapping.app_id = backing_app_id
+                mapping.conversation_id = conversation_id
+            self._session.flush()
+            self._session.commit()
+        except Exception:
+            self._session.rollback()
+            raise
+        if retired_binding_id is not None:
+            enqueue_agent_resource_collection(
+                tenant_id=tenant_id,
+                binding_ids=(retired_binding_id,),
+            )
+        return conversation_id
+
+    def load_or_create_build_conversation_ids_by_agent_id(
         self,
         *,
         tenant_id: str,
         agents: list[Agent],
         account_id: str,
-        draft_type: AgentConfigDraftType = AgentConfigDraftType.DEBUG_BUILD,
     ) -> dict[str, str]:
-        """Return per-account scoped conversations for a page of Agent Apps."""
+        """Return per-account Build conversations for a page of Agent Apps."""
 
         conversation_ids_by_agent_id: dict[str, str] = {}
         changed = False
@@ -839,7 +900,7 @@ class AgentRosterService:
             conversation_ids_by_agent_id[agent.id] = self._get_or_create_agent_app_debug_conversation(
                 agent=agent,
                 account_id=account_id,
-                draft_type=draft_type,
+                draft_type=AgentConfigDraftType.DEBUG_BUILD,
             )
             changed = True
         if changed:
@@ -1057,7 +1118,6 @@ class AgentRosterService:
             account_id=account.id,
         )
         self._session.commit()
-
         if FeatureService.get_system_features().webapp_auth.enabled:
             try:
                 original_settings = EnterpriseService.WebAppAuth.get_app_access_mode_by_id(source_app.id)
@@ -1211,13 +1271,38 @@ class AgentRosterService:
 
     def archive_roster_agent(self, *, tenant_id: str, agent_id: str, account_id: str) -> None:
         agent = self._get_agent(tenant_id=tenant_id, agent_id=agent_id, roster_only=True)
-        if agent.status == AgentStatus.ARCHIVED:
-            return
-        agent.status = AgentStatus.ARCHIVED
-        agent.archived_by = account_id
-        agent.archived_at = naive_utc_now()
-        agent.updated_by = account_id
+        retired_binding_ids: list[str] = []
+        if agent.status != AgentStatus.ARCHIVED:
+            agent.status = AgentStatus.ARCHIVED
+            agent.archived_by = account_id
+            agent.archived_at = naive_utc_now()
+            agent.updated_by = account_id
+        bindings = self._session.scalars(
+            select(AgentWorkspaceBinding).where(
+                AgentWorkspaceBinding.tenant_id == tenant_id,
+                AgentWorkspaceBinding.agent_id == agent_id,
+                AgentWorkspaceBinding.status == AgentWorkingResourceStatus.ACTIVE,
+            )
+        ).all()
+        for binding in bindings:
+            retired_id = AgentWorkspaceService.retire_binding(
+                session=self._session,
+                tenant_id=tenant_id,
+                binding_id=binding.id,
+            )
+            if retired_id is not None:
+                retired_binding_ids.append(retired_id)
+        retired_snapshot_ids = AgentHomeSnapshotService.retire_all_for_agent(
+            session=self._session,
+            tenant_id=tenant_id,
+            agent_id=agent_id,
+        )
         self._session.commit()
+        enqueue_agent_resource_collection(
+            tenant_id=tenant_id,
+            binding_ids=retired_binding_ids,
+            home_snapshot_ids=retired_snapshot_ids,
+        )
 
     @staticmethod
     def _visible_version_operations(agent: Agent) -> set[AgentConfigRevisionOperation]:
@@ -1382,9 +1467,11 @@ class AgentRosterService:
                 account_id=None,
                 draft_owner_key="",
                 created_by=account_id,
+                home_snapshot_id=version.home_snapshot_id,
             )
             self._session.add(draft)
         draft.base_snapshot_id = version.id
+        draft.home_snapshot_id = version.home_snapshot_id
         draft.config_snapshot = AgentSoulConfig.model_validate(version.config_snapshot_dict)
         draft.updated_by = account_id
         agent.active_config_is_published = version.id == agent.active_config_snapshot_id

api/services/agent/workflow_publish_service.py

@@ -24,6 +24,7 @@ from models.agent_config_entities import (
     WorkflowNodeJobConfig,
     WorkflowPreviousNodeOutputRef,
 )
+from models.model import App
 from models.workflow import Workflow
 from services.agent.composer_validator import ComposerConfigValidator
 from services.agent.prompt_mentions import (
@@ -224,7 +225,7 @@ class WorkflowAgentPublishService:
         session: Session,
         draft_workflow: Workflow,
         account_id: str,
-    ) -> None:
+    ) -> set[str]:
         agent_nodes = dict(WorkflowAgentNodeValidator.iter_agent_v2_nodes(draft_workflow.graph_dict))
         existing_bindings = list(
             session.scalars(
@@ -237,9 +238,12 @@ class WorkflowAgentPublishService:
             ).all()
         )
         existing_by_node_id = {binding.node_id: binding for binding in existing_bindings}
+        retirement_candidates: set[str] = set()
 
         for binding in existing_bindings:
             if binding.node_id not in agent_nodes:
+                if binding.binding_type == WorkflowAgentBindingType.INLINE_AGENT and binding.agent_id:
+                    retirement_candidates.add(binding.agent_id)
                 session.delete(binding)
 
         for node_id, node_data in agent_nodes.items():
@@ -252,16 +256,34 @@ class WorkflowAgentPublishService:
                 not binding_payload.get("agent_id") or not binding_payload.get("current_snapshot_id")
             ):
                 continue
+            existing_binding = existing_by_node_id.get(node_id)
+            replaced_inline_agent_id = (
+                existing_binding.agent_id
+                if existing_binding is not None
+                and existing_binding.binding_type == WorkflowAgentBindingType.INLINE_AGENT
+                and existing_binding.agent_id
+                else None
+            )
             cls._sync_agent_binding_for_node(
                 session=session,
                 draft_workflow=draft_workflow,
                 node_id=node_id,
                 node_data=node_data,
                 node_binding=binding_payload,
-                existing_binding=existing_by_node_id.get(node_id),
+                existing_binding=existing_binding,
                 account_id=account_id,
             )
+            if (
+                replaced_inline_agent_id
+                and existing_binding is not None
+                and (
+                    existing_binding.binding_type != WorkflowAgentBindingType.INLINE_AGENT
+                    or existing_binding.agent_id != replaced_inline_agent_id
+                )
+            ):
+                retirement_candidates.add(replaced_inline_agent_id)
         session.flush()
+        return retirement_candidates
 
     @classmethod
     def sync_roster_agent_bindings_for_draft(
@@ -270,8 +292,8 @@ class WorkflowAgentPublishService:
         session: Session,
         draft_workflow: Workflow,
         account_id: str,
-    ) -> None:
-        cls.sync_agent_bindings_for_draft(
+    ) -> set[str]:
+        return cls.sync_agent_bindings_for_draft(
             session=session,
             draft_workflow=draft_workflow,
             account_id=account_id,
@@ -561,12 +583,32 @@ class WorkflowAgentPublishService:
         session: Session,
         draft_workflow: Workflow,
         published_workflow: Workflow,
-    ) -> None:
+    ) -> set[str]:
+        current_workflow_id = session.scalar(
+            select(App.workflow_id).where(
+                App.tenant_id == draft_workflow.tenant_id,
+                App.id == draft_workflow.app_id,
+            )
+        )
+        retirement_candidates: set[str] = set()
+        if current_workflow_id:
+            retirement_candidates = {
+                agent_id
+                for agent_id in session.scalars(
+                    select(WorkflowAgentNodeBinding.agent_id).where(
+                        WorkflowAgentNodeBinding.tenant_id == draft_workflow.tenant_id,
+                        WorkflowAgentNodeBinding.app_id == draft_workflow.app_id,
+                        WorkflowAgentNodeBinding.workflow_id == current_workflow_id,
+                        WorkflowAgentNodeBinding.binding_type == WorkflowAgentBindingType.INLINE_AGENT,
+                    )
+                ).all()
+                if agent_id
+            }
         node_ids = {
             node_id for node_id, _node_data in WorkflowAgentNodeValidator.iter_agent_v2_nodes(draft_workflow.graph_dict)
         }
         if not node_ids:
-            return
+            return retirement_candidates
 
         bindings = session.scalars(
             select(WorkflowAgentNodeBinding).where(
@@ -578,7 +620,7 @@ class WorkflowAgentPublishService:
             )
         ).all()
         if not bindings:
-            return
+            return retirement_candidates
 
         agents_by_id = {
             agent.id: agent
@@ -611,6 +653,7 @@ class WorkflowAgentPublishService:
                 updated_by=binding.updated_by,
             )
             session.add(copied)
+        return retirement_candidates
 
     @classmethod
     def restore_agent_node_bindings_to_draft(
@@ -620,7 +663,7 @@ class WorkflowAgentPublishService:
         source_workflow: Workflow,
         draft_workflow: Workflow,
         account_id: str,
-    ) -> None:
+    ) -> set[str]:
         """Replace draft bindings with the frozen bindings of a published workflow."""
 
         existing = session.scalars(
@@ -631,6 +674,11 @@ class WorkflowAgentPublishService:
                 WorkflowAgentNodeBinding.workflow_version == cls._DRAFT_WORKFLOW_VERSION,
             )
         ).all()
+        retirement_candidates = {
+            binding.agent_id
+            for binding in existing
+            if binding.binding_type == WorkflowAgentBindingType.INLINE_AGENT and binding.agent_id
+        }
         for binding in existing:
             session.delete(binding)
 
@@ -681,3 +729,4 @@ class WorkflowAgentPublishService:
                 )
             )
         session.flush()
+        return retirement_candidates

api/services/agent/workspace_service.py

@@ -0,0 +1,469 @@
+"""Own Workspace and AgentWorkspaceBinding product lifecycle.
+
+Dify API is the lifecycle ledger. Dify Agent only executes physical create,
+acquire, and destroy operations selected by this service. Retire methods only
+mutate the caller's transaction; collection performs network I/O after commit
+and deletes ledger rows only after idempotent physical cleanup succeeds.
+"""
+
+from __future__ import annotations
+
+import logging
+from dataclasses import dataclass
+
+from dify_agent.client import Client
+from dify_agent.protocol import CreateExecutionBindingRequest, DestroyExecutionBindingRequest
+from sqlalchemy import select
+from sqlalchemy.orm import Session
+
+from configs import dify_config
+from core.db.session_factory import session_factory
+from libs.datetime_utils import naive_utc_now
+from libs.uuid_utils import uuidv7
+from models.agent import (
+    AgentConfigVersionKind,
+    AgentHomeSnapshot,
+    AgentWorkingResourceStatus,
+    AgentWorkspace,
+    AgentWorkspaceBinding,
+    AgentWorkspaceOwnerType,
+)
+
+logger = logging.getLogger(__name__)
+
+
+class AgentWorkspaceError(RuntimeError):
+    pass
+
+
+class AgentWorkspaceNotFoundError(AgentWorkspaceError):
+    pass
+
+
+class AgentWorkspaceBindingGenerationMismatchError(AgentWorkspaceError):
+    pass
+
+
+@dataclass(frozen=True, slots=True)
+class WorkspaceOwnerScope:
+    tenant_id: str
+    app_id: str
+    owner_type: AgentWorkspaceOwnerType
+    owner_id: str
+    owner_scope_key: str = "root"
+
+
+class AgentWorkspaceService:
+    """Allocate and manage working-environment resources.
+
+    A Binding ID is the participant identity. Product callers persist that ID
+    and use :meth:`get_active_binding`; Agent and Workspace attributes are not
+    participant lookup keys.
+    """
+
+    @classmethod
+    def resolve_active_workspace(cls, *, session: Session, scope: WorkspaceOwnerScope) -> AgentWorkspace | None:
+        return session.scalar(
+            select(AgentWorkspace).where(
+                AgentWorkspace.tenant_id == scope.tenant_id,
+                AgentWorkspace.app_id == scope.app_id,
+                AgentWorkspace.owner_type == scope.owner_type,
+                AgentWorkspace.owner_id == scope.owner_id,
+                AgentWorkspace.owner_scope_key == scope.owner_scope_key,
+                AgentWorkspace.status == AgentWorkingResourceStatus.ACTIVE,
+            )
+        )
+
+    @classmethod
+    def get_active_binding(
+        cls,
+        *,
+        session: Session,
+        tenant_id: str,
+        binding_id: str,
+        expected_owner_scope: WorkspaceOwnerScope,
+    ) -> AgentWorkspaceBinding | None:
+        return session.scalar(
+            select(AgentWorkspaceBinding)
+            .join(
+                AgentWorkspace,
+                (AgentWorkspace.tenant_id == AgentWorkspaceBinding.tenant_id)
+                & (AgentWorkspace.id == AgentWorkspaceBinding.workspace_id),
+            )
+            .where(
+                AgentWorkspaceBinding.id == binding_id,
+                AgentWorkspaceBinding.tenant_id == tenant_id,
+                AgentWorkspaceBinding.status == AgentWorkingResourceStatus.ACTIVE,
+                AgentWorkspace.tenant_id == expected_owner_scope.tenant_id,
+                AgentWorkspace.app_id == expected_owner_scope.app_id,
+                AgentWorkspace.owner_type == expected_owner_scope.owner_type,
+                AgentWorkspace.owner_id == expected_owner_scope.owner_id,
+                AgentWorkspace.owner_scope_key == expected_owner_scope.owner_scope_key,
+                AgentWorkspace.status == AgentWorkingResourceStatus.ACTIVE,
+            )
+        )
+
+    @classmethod
+    def create_binding(
+        cls,
+        *,
+        session: Session,
+        scope: WorkspaceOwnerScope,
+        agent_id: str,
+        base_home_snapshot_id: str,
+        agent_config_version_id: str,
+        agent_config_version_kind: AgentConfigVersionKind,
+    ) -> AgentWorkspaceBinding:
+        """Allocate one new participant in the caller-owned transaction.
+
+        After backend creation returns successfully, any later Python, flush,
+        or commit failure may leave an orphan. Dify API does not perform
+        cross-system compensation; a future global reconciler is responsible
+        for those orphans. Backend-local cleanup applies only when creation
+        fails before the backend returns success.
+        """
+
+        home_snapshot = session.scalar(
+            select(AgentHomeSnapshot).where(
+                AgentHomeSnapshot.id == base_home_snapshot_id,
+                AgentHomeSnapshot.tenant_id == scope.tenant_id,
+                AgentHomeSnapshot.agent_id == agent_id,
+                AgentHomeSnapshot.status == AgentWorkingResourceStatus.ACTIVE,
+            )
+        )
+        if home_snapshot is None:
+            raise AgentWorkspaceNotFoundError("base Home Snapshot is unavailable")
+        workspace = cls.resolve_active_workspace(session=session, scope=scope)
+        workspace_id = workspace.id if workspace is not None else str(uuidv7())
+        binding_id = str(uuidv7())
+        with cls._client() as client:
+            allocation = client.create_execution_binding_sync(
+                CreateExecutionBindingRequest(
+                    tenant_id=scope.tenant_id,
+                    agent_id=agent_id,
+                    binding_id=binding_id,
+                    workspace_id=workspace_id,
+                    existing_workspace_ref=workspace.backend_workspace_ref if workspace is not None else None,
+                    home_snapshot_ref=home_snapshot.snapshot_ref,
+                )
+            )
+        if workspace is not None and allocation.workspace_ref != workspace.backend_workspace_ref:
+            raise AgentWorkspaceError("backend changed the existing Workspace ref")
+        if workspace is None:
+            workspace = AgentWorkspace(
+                id=workspace_id,
+                tenant_id=scope.tenant_id,
+                app_id=scope.app_id,
+                owner_type=scope.owner_type,
+                owner_id=scope.owner_id,
+                owner_scope_key=scope.owner_scope_key,
+                backend_workspace_ref=allocation.workspace_ref,
+                status=AgentWorkingResourceStatus.ACTIVE,
+                active_guard=1,
+            )
+            session.add(workspace)
+        binding = AgentWorkspaceBinding(
+            id=binding_id,
+            tenant_id=scope.tenant_id,
+            app_id=scope.app_id,
+            workspace_id=workspace_id,
+            agent_id=agent_id,
+            base_home_snapshot_id=base_home_snapshot_id,
+            agent_config_version_id=agent_config_version_id,
+            agent_config_version_kind=agent_config_version_kind,
+            backend_binding_ref=allocation.binding_ref,
+            status=AgentWorkingResourceStatus.ACTIVE,
+        )
+        session.add(binding)
+        return binding
+
+    @classmethod
+    def save_binding_session_snapshot(
+        cls,
+        *,
+        tenant_id: str,
+        binding_id: str,
+        session_snapshot: str,
+        pending_form_id: str | None = None,
+        pending_tool_call_id: str | None = None,
+    ) -> None:
+        with session_factory.create_session() as session:
+            binding = session.scalar(
+                select(AgentWorkspaceBinding).where(
+                    AgentWorkspaceBinding.id == binding_id,
+                    AgentWorkspaceBinding.tenant_id == tenant_id,
+                    AgentWorkspaceBinding.status == AgentWorkingResourceStatus.ACTIVE,
+                )
+            )
+            if binding is None:
+                raise AgentWorkspaceNotFoundError("ACTIVE Binding is unavailable")
+            binding.session_snapshot = session_snapshot
+            binding.pending_form_id = pending_form_id
+            binding.pending_tool_call_id = pending_tool_call_id
+            session.commit()
+
+    @classmethod
+    def retire_binding(cls, *, session: Session, tenant_id: str, binding_id: str) -> str | None:
+        binding = session.scalar(
+            select(AgentWorkspaceBinding)
+            .where(
+                AgentWorkspaceBinding.id == binding_id,
+                AgentWorkspaceBinding.tenant_id == tenant_id,
+                AgentWorkspaceBinding.status == AgentWorkingResourceStatus.ACTIVE,
+            )
+            .with_for_update()
+        )
+        if binding is None:
+            return None
+        workspace = session.scalar(
+            select(AgentWorkspace)
+            .where(
+                AgentWorkspace.id == binding.workspace_id,
+                AgentWorkspace.tenant_id == tenant_id,
+                AgentWorkspace.status == AgentWorkingResourceStatus.ACTIVE,
+            )
+            .with_for_update()
+        )
+        now = naive_utc_now()
+        binding.status = AgentWorkingResourceStatus.RETIRED
+        binding.retired_at = now
+        if workspace is not None:
+            other_binding = session.scalar(
+                select(AgentWorkspaceBinding.id).where(
+                    AgentWorkspaceBinding.tenant_id == tenant_id,
+                    AgentWorkspaceBinding.workspace_id == workspace.id,
+                    AgentWorkspaceBinding.status == AgentWorkingResourceStatus.ACTIVE,
+                    AgentWorkspaceBinding.id != binding.id,
+                )
+            )
+            if other_binding is None:
+                workspace.status = AgentWorkingResourceStatus.RETIRED
+                workspace.active_guard = None
+                workspace.retired_at = now
+        return binding.id
+
+    @classmethod
+    def retire_workspace(cls, *, session: Session, tenant_id: str, workspace_id: str) -> str | None:
+        workspace = session.scalar(
+            select(AgentWorkspace)
+            .where(
+                AgentWorkspace.id == workspace_id,
+                AgentWorkspace.tenant_id == tenant_id,
+                AgentWorkspace.status == AgentWorkingResourceStatus.ACTIVE,
+            )
+            .with_for_update()
+        )
+        if workspace is None:
+            return None
+        now = naive_utc_now()
+        workspace.status = AgentWorkingResourceStatus.RETIRED
+        workspace.active_guard = None
+        workspace.retired_at = now
+        bindings = session.scalars(
+            select(AgentWorkspaceBinding).where(
+                AgentWorkspaceBinding.tenant_id == tenant_id,
+                AgentWorkspaceBinding.workspace_id == workspace.id,
+                AgentWorkspaceBinding.status == AgentWorkingResourceStatus.ACTIVE,
+            )
+        ).all()
+        for binding in bindings:
+            binding.status = AgentWorkingResourceStatus.RETIRED
+            binding.retired_at = now
+        return workspace.id
+
+    @classmethod
+    def retire_all_for_app(cls, *, session: Session, tenant_id: str, app_id: str) -> list[str]:
+        """Retire all ACTIVE Workspaces owned by an App in the caller's transaction."""
+
+        workspaces = session.scalars(
+            select(AgentWorkspace).where(
+                AgentWorkspace.tenant_id == tenant_id,
+                AgentWorkspace.app_id == app_id,
+                AgentWorkspace.status == AgentWorkingResourceStatus.ACTIVE,
+            )
+        ).all()
+        retired: list[str] = []
+        for workspace in workspaces:
+            workspace_id = cls.retire_workspace(
+                session=session,
+                tenant_id=tenant_id,
+                workspace_id=workspace.id,
+            )
+            if workspace_id is not None:
+                retired.append(workspace_id)
+        return retired
+
+    @classmethod
+    def collect_retired_binding(cls, *, tenant_id: str, binding_id: str) -> None:
+        try:
+            cls._collect_retired_binding(tenant_id=tenant_id, binding_id=binding_id)
+        except Exception:
+            logger.exception(
+                "Failed to collect retired Agent Workspace Binding",
+                extra={"tenant_id": tenant_id, "binding_id": binding_id},
+            )
+
+    @classmethod
+    def _collect_retired_binding(cls, *, tenant_id: str, binding_id: str) -> None:
+        with session_factory.create_session() as session:
+            binding = session.scalar(
+                select(AgentWorkspaceBinding).where(
+                    AgentWorkspaceBinding.id == binding_id,
+                    AgentWorkspaceBinding.tenant_id == tenant_id,
+                    AgentWorkspaceBinding.status == AgentWorkingResourceStatus.RETIRED,
+                )
+            )
+            if binding is None:
+                return
+            backend_binding_ref = binding.backend_binding_ref
+            workspace = session.scalar(
+                select(AgentWorkspace).where(
+                    AgentWorkspace.id == binding.workspace_id,
+                    AgentWorkspace.tenant_id == tenant_id,
+                )
+            )
+            if workspace is not None and workspace.status == AgentWorkingResourceStatus.RETIRED:
+                workspace_id = workspace.id
+            else:
+                workspace_id = None
+        if workspace_id is not None:
+            cls.collect_retired_workspace(tenant_id=tenant_id, workspace_id=workspace_id)
+            return
+        try:
+            with cls._client() as client:
+                client.destroy_execution_binding_sync(
+                    DestroyExecutionBindingRequest(
+                        binding_ref=backend_binding_ref,
+                        destroy_workspace=False,
+                    )
+                )
+        except Exception:
+            logger.exception(
+                "Failed to collect retired Agent Workspace Binding",
+                extra={"tenant_id": tenant_id, "binding_id": binding_id},
+            )
+            return
+        with session_factory.create_session() as session:
+            binding = session.scalar(
+                select(AgentWorkspaceBinding).where(
+                    AgentWorkspaceBinding.id == binding_id,
+                    AgentWorkspaceBinding.tenant_id == tenant_id,
+                    AgentWorkspaceBinding.status == AgentWorkingResourceStatus.RETIRED,
+                )
+            )
+            if binding is not None:
+                session.delete(binding)
+                session.commit()
+
+    @classmethod
+    def collect_retired_workspace(cls, *, tenant_id: str, workspace_id: str) -> None:
+        try:
+            cls._collect_retired_workspace(tenant_id=tenant_id, workspace_id=workspace_id)
+        except Exception:
+            logger.exception(
+                "Failed to collect retired Agent Workspace",
+                extra={"tenant_id": tenant_id, "workspace_id": workspace_id},
+            )
+
+    @classmethod
+    def _collect_retired_workspace(cls, *, tenant_id: str, workspace_id: str) -> None:
+        with session_factory.create_session() as session:
+            workspace = session.scalar(
+                select(AgentWorkspace).where(
+                    AgentWorkspace.id == workspace_id,
+                    AgentWorkspace.tenant_id == tenant_id,
+                    AgentWorkspace.status == AgentWorkingResourceStatus.RETIRED,
+                )
+            )
+            if workspace is None:
+                return
+            bindings = session.scalars(
+                select(AgentWorkspaceBinding)
+                .where(
+                    AgentWorkspaceBinding.tenant_id == tenant_id,
+                    AgentWorkspaceBinding.workspace_id == workspace_id,
+                    AgentWorkspaceBinding.status == AgentWorkingResourceStatus.RETIRED,
+                )
+                .order_by(AgentWorkspaceBinding.created_at)
+            ).all()
+            if not bindings:
+                logger.error(
+                    "RETIRED Workspace has no Binding available for physical collection",
+                    extra={"tenant_id": tenant_id, "workspace_id": workspace_id},
+                )
+                return
+            anchor = bindings[0]
+            remaining_ids = [binding.id for binding in bindings[1:]]
+            workspace_ref = workspace.backend_workspace_ref
+            binding_ref = anchor.backend_binding_ref
+            anchor_id = anchor.id
+        try:
+            with cls._client() as client:
+                client.destroy_execution_binding_sync(
+                    DestroyExecutionBindingRequest(
+                        binding_ref=binding_ref,
+                        workspace_ref=workspace_ref,
+                        destroy_workspace=True,
+                    )
+                )
+        except Exception:
+            logger.exception(
+                "Failed to collect retired Agent Workspace",
+                extra={"tenant_id": tenant_id, "workspace_id": workspace_id, "binding_id": anchor_id},
+            )
+            return
+        with session_factory.create_session() as session:
+            stored_workspace = session.scalar(
+                select(AgentWorkspace).where(
+                    AgentWorkspace.id == workspace_id,
+                    AgentWorkspace.tenant_id == tenant_id,
+                    AgentWorkspace.status == AgentWorkingResourceStatus.RETIRED,
+                )
+            )
+            stored_anchor = session.scalar(
+                select(AgentWorkspaceBinding).where(
+                    AgentWorkspaceBinding.id == anchor_id,
+                    AgentWorkspaceBinding.tenant_id == tenant_id,
+                    AgentWorkspaceBinding.status == AgentWorkingResourceStatus.RETIRED,
+                )
+            )
+            if stored_workspace is not None:
+                session.delete(stored_workspace)
+            if stored_anchor is not None:
+                session.delete(stored_anchor)
+            session.commit()
+        for remaining_id in remaining_ids:
+            cls.collect_retired_binding(tenant_id=tenant_id, binding_id=remaining_id)
+
+    @staticmethod
+    def validate_binding_generation(
+        binding: AgentWorkspaceBinding,
+        *,
+        base_home_snapshot_id: str,
+        agent_config_version_id: str,
+        agent_config_version_kind: AgentConfigVersionKind,
+    ) -> None:
+        if (
+            binding.base_home_snapshot_id != base_home_snapshot_id
+            or binding.agent_config_version_id != agent_config_version_id
+            or binding.agent_config_version_kind != agent_config_version_kind
+        ):
+            raise AgentWorkspaceBindingGenerationMismatchError(
+                "ACTIVE Binding belongs to a different Agent config/Home generation"
+            )
+
+    @staticmethod
+    def _client() -> Client:
+        base_url = dify_config.AGENT_BACKEND_BASE_URL
+        if not base_url:
+            raise AgentWorkspaceError("Dify Agent backend is required for Workspace operations")
+        return Client(base_url=base_url)
+
+
+__all__ = [
+    "AgentWorkspaceBindingGenerationMismatchError",
+    "AgentWorkspaceError",
+    "AgentWorkspaceNotFoundError",
+    "AgentWorkspaceService",
+    "WorkspaceOwnerScope",
+]

api/services/agent_app_sandbox_service.py

@@ -1,39 +1,40 @@
-"""Resolve and proxy sandbox file access for Agent App and workflow Agent sessions.
-
-These services keep product-facing locators (conversation, workflow run, node)
-on the API boundary and translate them into the agent backend's
-``SandboxLocator`` using persisted non-sensitive runtime layer specs plus the
-saved Agenton session snapshot. Upload responses stay console-facing here: the
-agent backend still returns a canonical ToolFile mapping, while this API layer
-re-resolves that mapping into a signed browser download URL.
-"""
+"""Resolve product locators to ACTIVE Workspace Bindings and proxy file access."""
 
 from __future__ import annotations
 
 import urllib.parse
 from collections.abc import Callable
-from typing import Any
+from typing import Any, Literal, cast
 
-from agenton.compositor import CompositorSessionSnapshot
 from dify_agent.client import Client
-from dify_agent.protocol import RuntimeLayerSpec, SandboxLocator, build_sandbox_locator_from_layer_specs
-from pydantic import BaseModel, TypeAdapter
+from dify_agent.layers.execution_context import (
+    DifyExecutionContextAgentConfigVersionKind,
+    DifyExecutionContextLayerConfig,
+)
+from dify_agent.protocol import WorkspaceListResponse, WorkspaceReadResponse, WorkspaceUploadRequest
+from pydantic import BaseModel
 from sqlalchemy import select
 from sqlalchemy.orm import Session
 
 from configs import dify_config
-from core.app.apps.agent_app.session_store import AgentAppRuntimeSessionStore
 from core.app.file_access import DatabaseFileAccessController
 from core.app.workflow.file_runtime import DifyWorkflowFileRuntime
+from core.db.session_factory import session_factory
 from factories import file_factory
-from models.agent import AgentRuntimeSessionOwnerType, WorkflowAgentRuntimeSession, WorkflowAgentRuntimeSessionStatus
-
-_RUNTIME_LAYER_SPECS_ADAPTER: TypeAdapter[list[RuntimeLayerSpec]] = TypeAdapter(list[RuntimeLayerSpec])
+from models.agent import (
+    Agent,
+    AgentConfigDraft,
+    AgentConfigDraftType,
+    AgentWorkspaceBinding,
+    AgentWorkspaceOwnerType,
+)
+from models.model import App, Conversation
+from models.workflow import WorkflowNodeExecutionModel
+from services.agent.roster_service import AgentRosterService
+from services.agent.workspace_service import AgentWorkspaceService, WorkspaceOwnerScope
 
 
 class AgentSandboxInspectorError(Exception):
-    """A sandbox inspection failure mapped to an HTTP status by the controller."""
-
     code: str
     message: str
     status_code: int
@@ -46,82 +47,200 @@ class AgentSandboxInspectorError(Exception):
 
 
 class AgentSandboxInfo(BaseModel):
-    """Basic Agent App sandbox metadata returned after a successful availability probe."""
-
-    session_id: str
     workspace_cwd: str
 
 
 class AgentSandboxUploadDownload(BaseModel):
-    """Signed browser download URL for one sandbox upload result."""
-
     url: str
 
 
 class AgentAppSandboxService:
-    """Inspect and proxy file access for an Agent App conversation sandbox."""
-
-    def __init__(
-        self,
-        *,
-        session_store: AgentAppRuntimeSessionStore | None = None,
-        client_factory: Callable[[], Client] | None = None,
-    ) -> None:
-        self._session_store = session_store or AgentAppRuntimeSessionStore()
+    def __init__(self, *, client_factory: Callable[[], Client] | None = None) -> None:
         self._client_factory = client_factory or _default_client_factory
 
-    def get_info(self, *, tenant_id: str, app_id: str, conversation_id: str) -> AgentSandboxInfo:
-        locator = self._resolve_locator(tenant_id=tenant_id, app_id=app_id, conversation_id=conversation_id)
-        session_id, workspace_cwd = _extract_shell_workspace_or_raise(
-            snapshot=locator.session_snapshot,
-            not_found_message="this conversation's agent has no sandbox workspace",
-        )
-
-        return AgentSandboxInfo(
-            session_id=session_id,
-            workspace_cwd=workspace_cwd,
-        )
-
-    def list_files(self, *, tenant_id: str, app_id: str, conversation_id: str, path: str):
-        locator = self._resolve_locator(tenant_id=tenant_id, app_id=app_id, conversation_id=conversation_id)
-        return self._client_factory().list_sandbox_files_sync(locator, path)
-
-    def read_file(self, *, tenant_id: str, app_id: str, conversation_id: str, path: str):
-        locator = self._resolve_locator(tenant_id=tenant_id, app_id=app_id, conversation_id=conversation_id)
-        return self._client_factory().read_sandbox_file_sync(locator, path)
-
-    def upload_file(
-        self, *, tenant_id: str, app_id: str, conversation_id: str, path: str
-    ) -> AgentSandboxUploadDownload:
-        locator = self._resolve_locator(tenant_id=tenant_id, app_id=app_id, conversation_id=conversation_id)
-        uploaded = self._client_factory().upload_sandbox_file_sync(locator, path)
-        return _upload_download_response(
-            tenant_id=tenant_id,
-            file_mapping=uploaded.file.model_dump(mode="python"),
-        )
-
-    def _resolve_locator(self, *, tenant_id: str, app_id: str, conversation_id: str) -> SandboxLocator:
-        stored = self._session_store.load_active_session_for_conversation(
+    def get_info(
+        self,
+        *,
+        tenant_id: str,
+        app_id: str,
+        agent_id: str,
+        caller_type: Literal["conversation", "build_draft"],
+        caller_id: str,
+        account_id: str,
+    ) -> AgentSandboxInfo:
+        self._resolve_binding(
             tenant_id=tenant_id,
             app_id=app_id,
-            conversation_id=conversation_id,
+            agent_id=agent_id,
+            caller_type=caller_type,
+            caller_id=caller_id,
+            account_id=account_id,
         )
-        if stored is None:
-            raise AgentSandboxInspectorError(
-                "no_active_session",
-                "this conversation has no active sandbox session yet",
-                status_code=404,
+        return AgentSandboxInfo(workspace_cwd=".")
+
+    def list_files(
+        self,
+        *,
+        tenant_id: str,
+        app_id: str,
+        agent_id: str,
+        caller_type: Literal["conversation", "build_draft"],
+        caller_id: str,
+        account_id: str,
+        path: str,
+    ) -> WorkspaceListResponse:
+        binding = self._resolve_binding(
+            tenant_id=tenant_id,
+            app_id=app_id,
+            agent_id=agent_id,
+            caller_type=caller_type,
+            caller_id=caller_id,
+            account_id=account_id,
+        )
+        with self._client_factory() as client:
+            return client.list_workspace_files_sync(binding.backend_binding_ref, path)
+
+    def read_file(
+        self,
+        *,
+        tenant_id: str,
+        app_id: str,
+        agent_id: str,
+        caller_type: Literal["conversation", "build_draft"],
+        caller_id: str,
+        account_id: str,
+        path: str,
+    ) -> WorkspaceReadResponse:
+        binding = self._resolve_binding(
+            tenant_id=tenant_id,
+            app_id=app_id,
+            agent_id=agent_id,
+            caller_type=caller_type,
+            caller_id=caller_id,
+            account_id=account_id,
+        )
+        with self._client_factory() as client:
+            return client.read_workspace_file_sync(binding.backend_binding_ref, path)
+
+    def upload_file(
+        self,
+        *,
+        tenant_id: str,
+        app_id: str,
+        agent_id: str,
+        caller_type: Literal["conversation", "build_draft"],
+        caller_id: str,
+        account_id: str,
+        path: str,
+    ) -> AgentSandboxUploadDownload:
+        binding = self._resolve_binding(
+            tenant_id=tenant_id,
+            app_id=app_id,
+            agent_id=agent_id,
+            caller_type=caller_type,
+            caller_id=caller_id,
+            account_id=account_id,
+        )
+        with self._client_factory() as client:
+            uploaded = client.upload_workspace_file_sync(
+                WorkspaceUploadRequest(
+                    backend_binding_ref=binding.backend_binding_ref,
+                    path=path,
+                    execution_context=DifyExecutionContextLayerConfig(
+                        tenant_id=tenant_id,
+                        app_id=app_id,
+                        conversation_id=caller_id if caller_type == "conversation" else None,
+                        agent_id=agent_id,
+                        agent_config_version_id=binding.agent_config_version_id,
+                        agent_config_version_kind=cast(
+                            DifyExecutionContextAgentConfigVersionKind,
+                            binding.agent_config_version_kind.value,
+                        ),
+                        agent_mode="agent_app",
+                        invoke_from="debugger",
+                    ),
+                )
             )
-        return _build_locator_or_raise(
-            snapshot=stored.session_snapshot,
-            runtime_layer_specs=stored.runtime_layer_specs,
-            not_found_message="this conversation's agent has no sandbox workspace",
-        )
+        return _upload_download_response(tenant_id=tenant_id, file_mapping=uploaded.file.model_dump(mode="python"))
+
+    @staticmethod
+    def _resolve_binding(
+        *,
+        tenant_id: str,
+        app_id: str,
+        agent_id: str,
+        caller_type: Literal["conversation", "build_draft"],
+        caller_id: str,
+        account_id: str,
+    ) -> AgentWorkspaceBinding:
+        with session_factory.create_session() as session:
+            caller: AgentConfigDraft | Conversation | None
+            if caller_type == "build_draft":
+                agent = session.scalar(
+                    select(Agent).where(
+                        Agent.id == agent_id,
+                        Agent.tenant_id == tenant_id,
+                    )
+                )
+                if agent is None or AgentRosterService.runtime_backing_app_id(agent) != app_id:
+                    caller = None
+                else:
+                    caller = session.scalar(
+                        select(AgentConfigDraft).where(
+                            AgentConfigDraft.id == caller_id,
+                            AgentConfigDraft.tenant_id == tenant_id,
+                            AgentConfigDraft.agent_id == agent_id,
+                            AgentConfigDraft.account_id == account_id,
+                            AgentConfigDraft.draft_type == AgentConfigDraftType.DEBUG_BUILD,
+                        )
+                    )
+                owner_scope = WorkspaceOwnerScope(
+                    tenant_id=tenant_id,
+                    app_id=app_id,
+                    owner_type=AgentWorkspaceOwnerType.BUILD_DRAFT,
+                    owner_id=caller_id,
+                )
+            else:
+                caller = session.scalar(
+                    select(Conversation)
+                    .join(App, App.id == Conversation.app_id)
+                    .where(
+                        App.tenant_id == tenant_id,
+                        Conversation.app_id == app_id,
+                        Conversation.id == caller_id,
+                        Conversation.from_account_id == account_id,
+                        Conversation.is_deleted.is_(False),
+                    )
+                )
+                owner_scope = WorkspaceOwnerScope(
+                    tenant_id=tenant_id,
+                    app_id=app_id,
+                    owner_type=AgentWorkspaceOwnerType.CONVERSATION,
+                    owner_id=caller_id,
+                )
+            if caller is None or caller.agent_workspace_binding_id is None:
+                raise AgentSandboxInspectorError(
+                    "no_active_binding",
+                    "this caller has no active Agent Workspace Binding",
+                    status_code=404,
+                )
+            binding = AgentWorkspaceService.get_active_binding(
+                session=session,
+                tenant_id=tenant_id,
+                binding_id=caller.agent_workspace_binding_id,
+                expected_owner_scope=owner_scope,
+            )
+            if binding is None or binding.agent_id != agent_id:
+                raise AgentSandboxInspectorError(
+                    "no_active_binding",
+                    "this caller has no active Agent Workspace Binding",
+                    status_code=404,
+                )
+            session.expunge(binding)
+            return binding
 
 
 class WorkflowAgentSandboxService:
-    """List/read/upload files in a workflow Agent node sandbox."""
-
     def __init__(self, *, client_factory: Callable[[], Client] | None = None) -> None:
         self._client_factory = client_factory or _default_client_factory
 
@@ -132,11 +251,11 @@ class WorkflowAgentSandboxService:
         app_id: str,
         workflow_run_id: str,
         node_id: str,
-        node_execution_id: str | None,
+        node_execution_id: str,
         path: str,
         session: Session,
-    ):
-        locator = self._resolve_locator(
+    ) -> WorkspaceListResponse:
+        binding = self._resolve_binding(
             tenant_id=tenant_id,
             app_id=app_id,
             workflow_run_id=workflow_run_id,
@@ -144,7 +263,8 @@ class WorkflowAgentSandboxService:
             node_execution_id=node_execution_id,
             session=session,
         )
-        return self._client_factory().list_sandbox_files_sync(locator, path)
+        with self._client_factory() as client:
+            return client.list_workspace_files_sync(binding.backend_binding_ref, path)
 
     def read_file(
         self,
@@ -153,11 +273,11 @@ class WorkflowAgentSandboxService:
         app_id: str,
         workflow_run_id: str,
         node_id: str,
-        node_execution_id: str | None,
+        node_execution_id: str,
         path: str,
         session: Session,
-    ):
-        locator = self._resolve_locator(
+    ) -> WorkspaceReadResponse:
+        binding = self._resolve_binding(
             tenant_id=tenant_id,
             app_id=app_id,
             workflow_run_id=workflow_run_id,
@@ -165,7 +285,8 @@ class WorkflowAgentSandboxService:
             node_execution_id=node_execution_id,
             session=session,
         )
-        return self._client_factory().read_sandbox_file_sync(locator, path)
+        with self._client_factory() as client:
+            return client.read_workspace_file_sync(binding.backend_binding_ref, path)
 
     def upload_file(
         self,
@@ -174,11 +295,11 @@ class WorkflowAgentSandboxService:
         app_id: str,
         workflow_run_id: str,
         node_id: str,
-        node_execution_id: str | None,
+        node_execution_id: str,
         path: str,
         session: Session,
     ) -> AgentSandboxUploadDownload:
-        locator = self._resolve_locator(
+        binding = self._resolve_binding(
             tenant_id=tenant_id,
             app_id=app_id,
             workflow_run_id=workflow_run_id,
@@ -186,118 +307,97 @@ class WorkflowAgentSandboxService:
             node_execution_id=node_execution_id,
             session=session,
         )
-        uploaded = self._client_factory().upload_sandbox_file_sync(locator, path)
-        return _upload_download_response(
-            tenant_id=tenant_id,
-            file_mapping=uploaded.file.model_dump(mode="python"),
-        )
+        with self._client_factory() as client:
+            uploaded = client.upload_workspace_file_sync(
+                WorkspaceUploadRequest(
+                    backend_binding_ref=binding.backend_binding_ref,
+                    path=path,
+                    execution_context=DifyExecutionContextLayerConfig(
+                        tenant_id=tenant_id,
+                        app_id=app_id,
+                        workflow_run_id=workflow_run_id,
+                        node_id=node_id,
+                        agent_id=binding.agent_id,
+                        agent_config_version_id=binding.agent_config_version_id,
+                        agent_config_version_kind=cast(
+                            DifyExecutionContextAgentConfigVersionKind,
+                            binding.agent_config_version_kind.value,
+                        ),
+                        agent_mode="workflow_run",
+                        invoke_from="debugger",
+                    ),
+                )
+            )
+        return _upload_download_response(tenant_id=tenant_id, file_mapping=uploaded.file.model_dump(mode="python"))
 
-    def _resolve_locator(
-        self,
+    @staticmethod
+    def _resolve_binding(
         *,
         tenant_id: str,
         app_id: str,
         workflow_run_id: str,
         node_id: str,
-        node_execution_id: str | None,
+        node_execution_id: str,
         session: Session,
-    ) -> SandboxLocator:
-        """Resolve one workflow Agent sandbox from product-facing identifiers.
-
-        Callers may target either a specific node execution or the current node
-        as a whole. When ``node_execution_id`` is provided, lookup narrows to
-        that execution's ACTIVE runtime-session row. When it is omitted, the
-        service falls back to the most recently updated ACTIVE session for the
-        same ``workflow_run_id + node_id`` pair so console sandbox inspection can
-        still work from the broader workflow/node locator.
-        """
-        stmt = select(WorkflowAgentRuntimeSession).where(
-            WorkflowAgentRuntimeSession.owner_type == AgentRuntimeSessionOwnerType.WORKFLOW_RUN,
-            WorkflowAgentRuntimeSession.tenant_id == tenant_id,
-            WorkflowAgentRuntimeSession.app_id == app_id,
-            WorkflowAgentRuntimeSession.workflow_run_id == workflow_run_id,
-            WorkflowAgentRuntimeSession.node_id == node_id,
-            WorkflowAgentRuntimeSession.status == WorkflowAgentRuntimeSessionStatus.ACTIVE,
+    ) -> AgentWorkspaceBinding:
+        execution = session.scalar(
+            select(WorkflowNodeExecutionModel).where(
+                WorkflowNodeExecutionModel.id == node_execution_id,
+                WorkflowNodeExecutionModel.tenant_id == tenant_id,
+                WorkflowNodeExecutionModel.app_id == app_id,
+                WorkflowNodeExecutionModel.workflow_run_id == workflow_run_id,
+                WorkflowNodeExecutionModel.node_id == node_id,
+            )
         )
-        if node_execution_id:
-            stmt = stmt.where(WorkflowAgentRuntimeSession.node_execution_id == node_execution_id)
-        stmt = stmt.order_by(WorkflowAgentRuntimeSession.updated_at.desc()).limit(1)
-
-        row = session.scalar(stmt)
-
-        if row is None:
+        process_data = execution.process_data_dict if execution is not None else None
+        workflow_agent_binding_id = process_data.get("workflow_agent_binding_id") if process_data is not None else None
+        if (
+            execution is None
+            or execution.agent_workspace_binding_id is None
+            or not isinstance(workflow_agent_binding_id, str)
+        ):
             raise AgentSandboxInspectorError(
-                "no_active_session",
-                "this workflow Agent node has no active sandbox session yet",
+                "no_active_binding",
+                "this Workflow Agent node execution has no active Workspace Binding",
                 status_code=404,
             )
-        return _build_locator_or_raise(
-            snapshot=CompositorSessionSnapshot.model_validate_json(row.session_snapshot),
-            runtime_layer_specs=_deserialize_runtime_layer_specs(row.composition_layer_specs),
-            not_found_message="this workflow Agent node has no sandbox workspace",
+        binding = AgentWorkspaceService.get_active_binding(
+            session=session,
+            tenant_id=tenant_id,
+            binding_id=execution.agent_workspace_binding_id,
+            expected_owner_scope=WorkspaceOwnerScope(
+                tenant_id=tenant_id,
+                app_id=app_id,
+                owner_type=AgentWorkspaceOwnerType.WORKFLOW_RUN,
+                owner_id=workflow_run_id,
+                owner_scope_key=f"{node_id}:{workflow_agent_binding_id}",
+            ),
         )
-
-
-def _build_locator_or_raise(
-    *,
-    snapshot: CompositorSessionSnapshot,
-    runtime_layer_specs: list[RuntimeLayerSpec],
-    not_found_message: str,
-) -> SandboxLocator:
-    try:
-        return build_sandbox_locator_from_layer_specs(
-            layer_specs=runtime_layer_specs,
-            session_snapshot=snapshot,
-        )
-    except ValueError as exc:
-        raise AgentSandboxInspectorError("no_sandbox", not_found_message, status_code=404) from exc
-
-
-def _extract_shell_workspace_or_raise(
-    *,
-    snapshot: CompositorSessionSnapshot,
-    not_found_message: str,
-) -> tuple[str, str]:
-    shell_layer = next((layer for layer in snapshot.layers if layer.name == "shell"), None)
-    if shell_layer is None:
-        raise AgentSandboxInspectorError("no_sandbox", not_found_message, status_code=404)
-
-    session_id = shell_layer.runtime_state.get("session_id")
-    workspace_cwd = shell_layer.runtime_state.get("workspace_cwd")
-    if not isinstance(session_id, str) or not isinstance(workspace_cwd, str):
-        raise AgentSandboxInspectorError("no_sandbox", not_found_message, status_code=404)
-    return session_id, workspace_cwd
-
-
-def _deserialize_runtime_layer_specs(value: str | None) -> list[RuntimeLayerSpec]:
-    if not value:
-        return []
-    return _RUNTIME_LAYER_SPECS_ADAPTER.validate_json(value)
+        if binding is None:
+            raise AgentSandboxInspectorError(
+                "no_active_binding",
+                "this Workflow Agent node execution has no active Workspace Binding",
+                status_code=404,
+            )
+        return binding
 
 
 def _upload_download_response(*, tenant_id: str, file_mapping: dict[str, Any]) -> AgentSandboxUploadDownload:
-    """Resolve one uploaded ToolFile mapping into a signed external download URL."""
-
     controller = DatabaseFileAccessController()
     runtime = DifyWorkflowFileRuntime(file_access_controller=controller)
     try:
-        file = file_factory.build_from_mapping(
-            mapping=file_mapping,
-            tenant_id=tenant_id,
-            access_controller=controller,
-        )
+        file = file_factory.build_from_mapping(mapping=file_mapping, tenant_id=tenant_id, access_controller=controller)
         url = runtime.resolve_file_url(file=file, for_external=True)
     except ValueError as exc:
         raise AgentSandboxInspectorError(
-            "sandbox_upload_download_unavailable",
-            "uploaded sandbox file could not be converted to a download URL",
+            "workspace_upload_download_unavailable",
+            "uploaded Workspace file could not be converted to a download URL",
             status_code=502,
         ) from exc
-
     if not url:
         raise AgentSandboxInspectorError(
-            "sandbox_upload_download_unavailable",
-            "uploaded sandbox file does not support download URL generation",
+            "workspace_upload_download_unavailable",
+            "uploaded Workspace file does not support download URL generation",
             status_code=502,
         )
     return AgentSandboxUploadDownload(url=_with_as_attachment(url))
@@ -315,7 +415,7 @@ def _default_client_factory() -> Client:
     if not base_url:
         raise AgentSandboxInspectorError(
             "inspector_unavailable",
-            "the sandbox file inspector is not available (agent backend not configured)",
+            "the Workspace file inspector is not available (Agent backend not configured)",
             status_code=503,
         )
     return Client(base_url=base_url)

api/services/app_dsl_service.py

@@ -41,6 +41,7 @@ from models import Account, App, AppMode
 from models.model import AppModelConfig, AppModelConfigDict, IconType, load_annotation_reply_config
 from models.workflow import Workflow
 from services.agent.dsl_service import AgentDslService, AgentPackage
+from services.agent.retirement_service import WorkflowAgentRetirementService
 from services.agent.workflow_publish_service import WorkflowAgentPublishService
 from services.dsl_content import DSL_MAX_SIZE, dsl_content_size
 from services.dsl_version import check_version_compatibility
@@ -51,6 +52,7 @@ from services.errors.app import WorkflowNotFoundError
 from services.plugin.dependencies_analysis import DependenciesAnalysisService
 from services.workflow_draft_variable_service import WorkflowDraftVariableService
 from services.workflow_service import WorkflowService
+from tasks.collect_agent_resources_task import enqueue_agent_resource_collection
 
 logger = logging.getLogger(__name__)
 
@@ -546,7 +548,7 @@ class AppDslService:
                     sync_agent_bindings=not raw_agent_packages,
                 )
                 if raw_agent_packages:
-                    _, warnings = AgentDslService(self._session).import_workflow_packages(
+                    _, warnings, retirement_candidates = AgentDslService(self._session).import_workflow_packages(
                         workflow=draft_workflow,
                         portable_graph=graph,
                         raw_packages=raw_agent_packages,
@@ -557,6 +559,17 @@ class AppDslService:
                         session=self._session,
                         draft_workflow=draft_workflow,
                     )
+                    self._session.commit()
+                    binding_ids, home_snapshot_ids = WorkflowAgentRetirementService.retire_unowned(
+                        tenant_id=app.tenant_id,
+                        agent_ids=retirement_candidates,
+                        account_id=account.id,
+                    )
+                    enqueue_agent_resource_collection(
+                        tenant_id=app.tenant_id,
+                        binding_ids=binding_ids,
+                        home_snapshot_ids=home_snapshot_ids,
+                    )
             case AppMode.CHAT | AppMode.AGENT_CHAT | AppMode.COMPLETION:
                 # Initialize model config
                 model_config = data.get("model_config")

api/services/app_service.py

@@ -26,17 +26,29 @@ from libs.datetime_utils import naive_utc_now
 from libs.login import current_user
 from libs.pagination import PaginatedResult, paginate_query
 from models import Account, AppStar
-from models.agent import APP_BACKED_AGENT_SOURCES, Agent, AgentIconType, AgentScope, AgentStatus
+from models.agent import (
+    APP_BACKED_AGENT_SOURCES,
+    Agent,
+    AgentIconType,
+    AgentScope,
+    AgentStatus,
+    AgentWorkingResourceStatus,
+    AgentWorkspaceBinding,
+)
 from models.model import App, AppMode, AppModelConfig, IconType, Site, load_annotation_reply_config
 from models.tools import ApiToolProvider
 from models.workflow import Workflow
 from services.agent.errors import AgentNameConflictError
+from services.agent.home_snapshot_service import AgentHomeSnapshotService
+from services.agent.retirement_service import WorkflowAgentRetirementService
+from services.agent.workspace_service import AgentWorkspaceService
 from services.billing_service import BillingService
 from services.enterprise import rbac_service as enterprise_rbac_service
 from services.enterprise.enterprise_service import EnterpriseService
 from services.feature_service import FeatureService
 from services.openapi.visibility import apply_openapi_gate, is_openapi_visible
 from services.tag_service import TagService
+from tasks.collect_agent_resources_task import enqueue_agent_resource_collection
 from tasks.remove_app_and_related_data_task import remove_app_and_related_data_task
 
 logger = logging.getLogger(__name__)
@@ -114,7 +126,7 @@ class AppModelConfigResponseView:
         self._session = session
 
     def __getattr__(self, name: str) -> Any:
-        return getattr(self._app_model_config, name)  # noqa: no-new-getattr response adapter delegates model fields
+        return getattr(self._app_model_config, name)  # guard-ignore: no-new-getattr -- delegates model fields
 
     @property
     def annotation_reply_dict(self) -> Any:
@@ -129,7 +141,7 @@ class AppResponseView:
         self._session = session
 
     def __getattr__(self, name: str) -> Any:
-        return getattr(self._app, name)  # noqa: no-new-getattr response adapter delegates model fields
+        return getattr(self._app, name)  # guard-ignore: no-new-getattr -- delegates model fields
 
     @property
     def desc_or_prompt(self) -> str:
@@ -478,7 +490,14 @@ class AppService:
 
         session.delete(existing_star)
 
-    def create_app(self, tenant_id: str, params: CreateAppParams, account: Account, *, session: Session) -> App:
+    def create_app(
+        self,
+        tenant_id: str,
+        params: CreateAppParams,
+        account: Account,
+        *,
+        session: Session,
+    ) -> App:
         """
         Create app
         :param tenant_id: tenant id
@@ -943,18 +962,67 @@ class AppService:
         app_was_deleted.send(app)
 
         backing_agent = self._get_backing_agent_for_update(app, session=session)
+        workflow_agent_ids = session.scalars(
+            select(Agent.id).where(
+                Agent.tenant_id == app.tenant_id,
+                Agent.app_id == app.id,
+                Agent.scope == AgentScope.WORKFLOW_ONLY,
+                Agent.status == AgentStatus.ACTIVE,
+            )
+        ).all()
+        account_id = current_user.id if current_user else None
         if backing_agent is not None:
             now = naive_utc_now()
-            account_id = getattr(current_user, "id", None)
             backing_agent.status = AgentStatus.ARCHIVED
             backing_agent.archived_by = account_id
             backing_agent.archived_at = now
             backing_agent.updated_by = account_id
             backing_agent.updated_at = now
 
+        retired_binding_ids: list[str] = []
+        retired_snapshot_ids: list[str] = []
+        if backing_agent is not None:
+            bindings = session.scalars(
+                select(AgentWorkspaceBinding).where(
+                    AgentWorkspaceBinding.tenant_id == app.tenant_id,
+                    AgentWorkspaceBinding.agent_id == backing_agent.id,
+                    AgentWorkspaceBinding.status == AgentWorkingResourceStatus.ACTIVE,
+                )
+            ).all()
+            for binding in bindings:
+                binding_id = AgentWorkspaceService.retire_binding(
+                    session=session,
+                    tenant_id=app.tenant_id,
+                    binding_id=binding.id,
+                )
+                if binding_id is not None:
+                    retired_binding_ids.append(binding_id)
+            retired_snapshot_ids = AgentHomeSnapshotService.retire_all_for_agent(
+                session=session,
+                tenant_id=app.tenant_id,
+                agent_id=backing_agent.id,
+            )
+
+        retired_workspace_ids = AgentWorkspaceService.retire_all_for_app(
+            session=session,
+            tenant_id=app.tenant_id,
+            app_id=app.id,
+        )
         session.delete(app)
         session.commit()
 
+        workflow_binding_ids, workflow_snapshot_ids = WorkflowAgentRetirementService.retire_unowned(
+            tenant_id=app.tenant_id,
+            agent_ids=workflow_agent_ids,
+            account_id=account_id,
+        )
+        enqueue_agent_resource_collection(
+            tenant_id=app.tenant_id,
+            workspace_ids=retired_workspace_ids,
+            binding_ids=[*retired_binding_ids, *workflow_binding_ids],
+            home_snapshot_ids=[*retired_snapshot_ids, *workflow_snapshot_ids],
+        )
+
         # clean up web app settings
         if FeatureService.get_system_features().webapp_auth.enabled:
             EnterpriseService.WebAppAuth.cleanup_webapp(app.id)

api/services/conversation_service.py

@@ -6,9 +6,7 @@ from typing import Any
 from sqlalchemy import asc, desc, func, or_, select
 from sqlalchemy.orm import Session
 
-from clients.agent_backend import AgentBackendSessionCleanupPayload
 from configs import dify_config
-from core.app.apps.agent_app.session_store import AgentAppRuntimeSessionStore
 from core.app.entities.app_invoke_entities import InvokeFrom
 from core.llm_generator.llm_generator import LLMGenerator
 from factories import variable_factory
@@ -16,7 +14,9 @@ from graphon.variables.types import SegmentType
 from libs.datetime_utils import naive_utc_now
 from libs.infinite_scroll_pagination import InfiniteScrollPagination
 from models import Account, ConversationVariable
+from models.agent import AgentWorkspaceOwnerType
 from models.model import App, Conversation, EndUser, Message
+from services.agent.workspace_service import AgentWorkspaceNotFoundError, AgentWorkspaceService, WorkspaceOwnerScope
 from services.errors.conversation import (
     ConversationNotExistsError,
     ConversationVariableNotExistsError,
@@ -24,7 +24,7 @@ from services.errors.conversation import (
     LastConversationNotExistsError,
 )
 from services.errors.message import MessageNotExistsError
-from tasks.agent_backend_session_cleanup_task import cleanup_conversation_agent_runtime_session
+from tasks.collect_agent_resources_task import enqueue_agent_resource_collection
 from tasks.delete_conversation_task import delete_conversation_related_data
 
 logger = logging.getLogger(__name__)
@@ -189,23 +189,30 @@ class ConversationService:
         """
         Delete a conversation only if it belongs to the given user and app context.
 
-        Before removing the conversation row, this best-effort lifecycle path
-        enumerates any ACTIVE conversation-owned Agent backend runtime sessions,
-        enqueues asynchronous backend cleanup for rows with persisted runtime
-        layer specs, and then retires the local session rows even if enqueueing
-        fails. Conversation deletion and related-data cleanup scheduling still
-        proceed when that lifecycle bookkeeping only partially succeeds.
+        Conversation deletion is the product lifecycle boundary for its
+        Workspace. Physical collection happens only after the retire commit.
 
         Raises:
             ConversationNotExistsError: When the conversation is not visible to the current user.
         """
         conversation = cls.get_conversation(app_model, conversation_id, user, session=session)
-        session_store = AgentAppRuntimeSessionStore()
-        stored_sessions = session_store.list_active_sessions_for_conversation(
-            tenant_id=app_model.tenant_id,
-            app_id=app_model.id,
-            conversation_id=conversation.id,
-        )
+        binding_id = conversation.agent_workspace_binding_id
+        retired_binding_id: str | None = None
+        if binding_id is not None:
+            owner_scope = WorkspaceOwnerScope(
+                tenant_id=app_model.tenant_id,
+                app_id=app_model.id,
+                owner_type=AgentWorkspaceOwnerType.CONVERSATION,
+                owner_id=conversation.id,
+            )
+            binding = AgentWorkspaceService.get_active_binding(
+                session=session,
+                tenant_id=app_model.tenant_id,
+                binding_id=binding_id,
+                expected_owner_scope=owner_scope,
+            )
+            if binding is None:
+                raise AgentWorkspaceNotFoundError("Conversation participant Binding is unavailable")
 
         try:
             logger.info(
@@ -213,66 +220,25 @@ class ConversationService:
                 app_model.name,
                 conversation_id,
             )
-            for stored_session in stored_sessions:
-                try:
-                    if stored_session.runtime_layer_specs:
-                        payload = AgentBackendSessionCleanupPayload(
-                            session_snapshot=stored_session.session_snapshot,
-                            runtime_layer_specs=stored_session.runtime_layer_specs,
-                            idempotency_key=(
-                                f"{stored_session.scope.tenant_id}:{stored_session.scope.app_id}:"
-                                f"{stored_session.scope.conversation_id}:agent-runtime-session-cleanup:"
-                                f"{stored_session.scope.agent_id}:"
-                                f"{stored_session.scope.agent_config_snapshot_id or 'no-config'}:"
-                                f"{stored_session.backend_run_id or 'no-run'}"
-                            ),
-                            metadata={
-                                "tenant_id": stored_session.scope.tenant_id,
-                                "app_id": stored_session.scope.app_id,
-                                "conversation_id": stored_session.scope.conversation_id,
-                                "agent_id": stored_session.scope.agent_id,
-                                "agent_config_snapshot_id": stored_session.scope.agent_config_snapshot_id,
-                                "previous_agent_backend_run_id": stored_session.backend_run_id,
-                            },
-                        )
-                        cleanup_conversation_agent_runtime_session.delay(payload.model_dump(mode="json"))
-                except Exception:
-                    logger.warning(
-                        "Failed to enqueue Agent backend cleanup for conversation deletion: "
-                        "tenant_id=%s app_id=%s conversation_id=%s agent_id=%s backend_run_id=%s",
-                        stored_session.scope.tenant_id,
-                        stored_session.scope.app_id,
-                        stored_session.scope.conversation_id,
-                        stored_session.scope.agent_id,
-                        stored_session.backend_run_id,
-                        exc_info=True,
-                    )
-                finally:
-                    try:
-                        session_store.mark_cleaned(
-                            scope=stored_session.scope,
-                            backend_run_id=stored_session.backend_run_id,
-                        )
-                    except Exception:
-                        logger.warning(
-                            "Failed to retire Agent App runtime session for conversation deletion: "
-                            "tenant_id=%s app_id=%s conversation_id=%s agent_id=%s backend_run_id=%s",
-                            stored_session.scope.tenant_id,
-                            stored_session.scope.app_id,
-                            stored_session.scope.conversation_id,
-                            stored_session.scope.agent_id,
-                            stored_session.backend_run_id,
-                            exc_info=True,
-                        )
-
+            if binding_id is not None:
+                retired_binding_id = AgentWorkspaceService.retire_binding(
+                    session=session,
+                    tenant_id=app_model.tenant_id,
+                    binding_id=binding_id,
+                )
+                if retired_binding_id is None:
+                    raise AgentWorkspaceNotFoundError("Conversation participant Binding is unavailable")
             session.delete(conversation)
             session.commit()
-
-            delete_conversation_related_data.delay(conversation.id)
-
         except Exception:
             session.rollback()
             raise
+        if retired_binding_id is not None:
+            enqueue_agent_resource_collection(
+                tenant_id=app_model.tenant_id,
+                binding_ids=(retired_binding_id,),
+            )
+        delete_conversation_related_data.delay(conversation.id)
 
     @classmethod
     def get_conversational_variable(

api/services/data_migration/import_service.py

@@ -26,6 +26,7 @@ from models import Account, ApiToken, Tenant, TenantAccountJoin, TenantAccountRo
 from models.enums import ApiTokenType
 from models.model import App
 from models.tools import ApiToolProvider, MCPToolProvider, WorkflowToolProvider
+from services.agent.retirement_service import WorkflowAgentRetirementService
 from services.app_dsl_service import AppDslService
 from services.data_migration.dependency_discovery_service import DependencyDiscoveryService
 from services.data_migration.entities import (
@@ -47,6 +48,7 @@ from services.tools.api_tools_manage_service import ApiToolManageService
 from services.tools.mcp_tools_manage_service import MCPToolManageService
 from services.tools.workflow_tools_manage_service import WorkflowToolManageService
 from services.workflow_service import WorkflowService
+from tasks.collect_agent_resources_task import enqueue_agent_resource_collection
 
 
 @dataclass(frozen=True)
@@ -711,7 +713,7 @@ class MigrationImportService:
                 raise MigrationDataError(f"Referenced workflow app was not found in target tenant: {app_id}")
             if account_in_session is None:
                 raise MigrationDataError(f"Operator account not found: {account.id}")
-            workflow = workflow_service.publish_workflow(
+            workflow, retirement_candidates = workflow_service.publish_workflow(
                 session=session,
                 app_model=app_in_session,
                 account=account_in_session,
@@ -721,6 +723,16 @@ class MigrationImportService:
             app_in_session.workflow_id = workflow.id
             app_in_session.updated_by = account.id
             app_in_session.updated_at = naive_utc_now()
+        binding_ids, home_snapshot_ids = WorkflowAgentRetirementService.retire_unowned(
+            tenant_id=target.tenant_id,
+            agent_ids=retirement_candidates,
+            account_id=account.id,
+        )
+        enqueue_agent_resource_collection(
+            tenant_id=target.tenant_id,
+            binding_ids=binding_ids,
+            home_snapshot_ids=home_snapshot_ids,
+        )
 
     def _import_mcp_tools(
         self,

api/services/snippet_dsl_service.py

@@ -19,12 +19,14 @@ from models import Account
 from models.snippet import CustomizedSnippet, SnippetType
 from models.workflow import Workflow
 from services.agent.dsl_service import AgentDslService
+from services.agent.retirement_service import WorkflowAgentRetirementService
 from services.agent.workflow_publish_service import WorkflowAgentPublishService
 from services.dsl_content import DSL_MAX_SIZE, dsl_content_size
 from services.dsl_version import check_version_compatibility
 from services.entities.dsl_entities import CheckDependenciesResult, DslImportWarning, ImportMode, ImportStatus
 from services.plugin.dependencies_analysis import DependenciesAnalysisService
 from services.snippet_service import SNIPPET_FORBIDDEN_NODE_TYPES, SnippetService
+from tasks.collect_agent_resources_task import enqueue_agent_resource_collection
 
 logger = logging.getLogger(__name__)
 
@@ -426,6 +428,7 @@ class SnippetDslService:
             self._session.flush()
 
         # Create or update draft workflow
+        retirement_candidates: set[str] = set()
         if workflow_data:
             graph = workflow_data.get("graph", {})
             raw_agent_packages = data.get("agent_packages") or {}
@@ -444,10 +447,10 @@ class SnippetDslService:
                 unique_hash=unique_hash,
                 account=account,
                 input_fields=input_fields,
-                sync_agent_bindings=not raw_agent_packages,
+                sync_agent_bindings=False,
             )
             if raw_agent_packages:
-                _, warnings = AgentDslService(self._session).import_workflow_packages(
+                _, warnings, retirement_candidates = AgentDslService(self._session).import_workflow_packages(
                     workflow=draft_workflow,
                     portable_graph=graph,
                     raw_packages=raw_agent_packages,
@@ -458,8 +461,29 @@ class SnippetDslService:
                     session=self._session,
                     draft_workflow=draft_workflow,
                 )
+            else:
+                retirement_candidates = WorkflowAgentPublishService.sync_agent_bindings_for_draft(
+                    session=self._session,
+                    draft_workflow=draft_workflow,
+                    account_id=account.id,
+                )
+                WorkflowAgentPublishService.validate_agent_nodes_for_draft_sync(
+                    session=self._session,
+                    draft_workflow=draft_workflow,
+                )
 
         self._session.commit()
+        if workflow_data:
+            binding_ids, home_snapshot_ids = WorkflowAgentRetirementService.retire_unowned(
+                tenant_id=snippet.tenant_id,
+                agent_ids=retirement_candidates,
+                account_id=account.id,
+            )
+            enqueue_agent_resource_collection(
+                tenant_id=snippet.tenant_id,
+                binding_ids=binding_ids,
+                home_snapshot_ids=home_snapshot_ids,
+            )
         return snippet
 
     def export_snippet_dsl(self, snippet: CustomizedSnippet, include_secret: bool = False) -> str:

api/services/snippet_service.py

@@ -35,6 +35,7 @@ from models.workflow import (
     WorkflowType,
 )
 from repositories.factory import DifyAPIRepositoryFactory
+from services.agent.retirement_service import WorkflowAgentRetirementService
 from services.errors.app import IsDraftWorkflowError, WorkflowHashNotEqualError, WorkflowNotFoundError
 from services.tag_service import TagService
 from services.workflow_node_execution_trace_service import (
@@ -42,6 +43,7 @@ from services.workflow_node_execution_trace_service import (
     assemble_workflow_node_execution_traces,
 )
 from services.workflow_restore import apply_published_workflow_snapshot_to_draft
+from tasks.collect_agent_resources_task import enqueue_agent_resource_collection
 
 logger = logging.getLogger(__name__)
 
@@ -80,7 +82,7 @@ class SnippetService:
 
     @contextmanager
     def _session_scope(self) -> Generator[Session, None, None]:
-        current_session = getattr(self, "_session", None)
+        current_session = self._session
         if current_session is not None:
             yield current_session
             return
@@ -89,7 +91,7 @@ class SnippetService:
             yield session
 
     def _commit_if_owned(self, session: Session) -> None:
-        if getattr(self, "_session", None) is None:
+        if self._session is None:
             session.commit()
 
     @staticmethod
@@ -600,12 +602,13 @@ class SnippetService:
 
         from services.agent.workflow_publish_service import WorkflowAgentPublishService
 
+        retirement_candidates: set[str] = set()
         with self._session_scope() as session:
             session.add(workflow)
             session.add(snippet)
             if sync_agent_bindings:
                 session.flush()
-                WorkflowAgentPublishService.sync_agent_bindings_for_draft(
+                retirement_candidates = WorkflowAgentPublishService.sync_agent_bindings_for_draft(
                     session=session,
                     draft_workflow=workflow,
                     account_id=account.id,
@@ -615,6 +618,17 @@ class SnippetService:
                     draft_workflow=workflow,
                 )
             self._commit_if_owned(session)
+        if self._session is None:
+            binding_ids, home_snapshot_ids = WorkflowAgentRetirementService.retire_unowned(
+                tenant_id=snippet.tenant_id,
+                agent_ids=retirement_candidates,
+                account_id=account.id,
+            )
+            enqueue_agent_resource_collection(
+                tenant_id=snippet.tenant_id,
+                binding_ids=binding_ids,
+                home_snapshot_ids=home_snapshot_ids,
+            )
         return workflow
 
     def restore_published_workflow_to_draft(
@@ -656,13 +670,24 @@ class SnippetService:
             session.flush()
             from services.agent.workflow_publish_service import WorkflowAgentPublishService
 
-            WorkflowAgentPublishService.restore_agent_node_bindings_to_draft(
+            retirement_candidates = WorkflowAgentPublishService.restore_agent_node_bindings_to_draft(
                 session=session,
                 source_workflow=source_workflow,
                 draft_workflow=draft_workflow,
                 account_id=account.id,
             )
             self._commit_if_owned(session)
+        if self._session is None:
+            binding_ids, home_snapshot_ids = WorkflowAgentRetirementService.retire_unowned(
+                tenant_id=snippet.tenant_id,
+                agent_ids=retirement_candidates,
+                account_id=account.id,
+            )
+            enqueue_agent_resource_collection(
+                tenant_id=snippet.tenant_id,
+                binding_ids=binding_ids,
+                home_snapshot_ids=home_snapshot_ids,
+            )
         return draft_workflow
 
     def publish_workflow(
@@ -671,7 +696,7 @@ class SnippetService:
         session: Session,
         snippet: CustomizedSnippet,
         account: Account,
-    ) -> Workflow:
+    ) -> tuple[Workflow, set[str]]:
         """
         Publish the draft workflow as a new version.
 
@@ -715,7 +740,7 @@ class SnippetService:
             kind=WorkflowKind.SNIPPET.value,
         )
         session.add(workflow)
-        WorkflowAgentPublishService.copy_agent_node_bindings_to_published(
+        retirement_candidates = WorkflowAgentPublishService.copy_agent_node_bindings_to_published(
             session=session,
             draft_workflow=draft_workflow,
             published_workflow=workflow,
@@ -728,7 +753,7 @@ class SnippetService:
         snippet.updated_by = account.id
         session.add(snippet)
 
-        return workflow
+        return workflow, retirement_candidates
 
     def get_all_published_workflows(
         self,

api/services/workflow_service.py

@@ -76,6 +76,7 @@ from models.model import App, AppMode
 from models.tools import WorkflowToolProvider
 from models.workflow import Workflow, WorkflowNodeExecutionModel, WorkflowNodeExecutionTriggeredFrom, WorkflowType
 from repositories.factory import DifyAPIRepositoryFactory
+from services.agent.retirement_service import WorkflowAgentRetirementService
 from services.billing_service import BillingService
 from services.errors.app import (
     IsDraftWorkflowError,
@@ -83,6 +84,7 @@ from services.errors.app import (
     WorkflowHashNotEqualError,
     WorkflowNotFoundError,
 )
+from tasks.collect_agent_resources_task import enqueue_agent_resource_collection
 
 
 @dataclass(frozen=True)
@@ -374,8 +376,9 @@ class WorkflowService:
         from services.agent.workflow_publish_service import WorkflowAgentPublishService
 
         session.flush()
+        retirement_candidates: set[str] = set()
         if sync_agent_bindings:
-            WorkflowAgentPublishService.sync_agent_bindings_for_draft(
+            retirement_candidates = WorkflowAgentPublishService.sync_agent_bindings_for_draft(
                 session=session,
                 draft_workflow=workflow,
                 account_id=account.id,
@@ -388,6 +391,16 @@ class WorkflowService:
         # commit db session changes
         if commit:
             session.commit()
+            binding_ids, home_snapshot_ids = WorkflowAgentRetirementService.retire_unowned(
+                tenant_id=app_model.tenant_id,
+                agent_ids=retirement_candidates,
+                account_id=account.id,
+            )
+            enqueue_agent_resource_collection(
+                tenant_id=app_model.tenant_id,
+                binding_ids=binding_ids,
+                home_snapshot_ids=home_snapshot_ids,
+            )
 
         # trigger app workflow events
         if commit:
@@ -509,7 +522,7 @@ class WorkflowService:
         from services.agent.workflow_publish_service import WorkflowAgentPublishService
 
         session.flush()
-        WorkflowAgentPublishService.restore_agent_node_bindings_to_draft(
+        retirement_candidates = WorkflowAgentPublishService.restore_agent_node_bindings_to_draft(
             session=session,
             source_workflow=source_workflow,
             draft_workflow=draft_workflow,
@@ -517,6 +530,16 @@ class WorkflowService:
         )
 
         session.commit()
+        binding_ids, home_snapshot_ids = WorkflowAgentRetirementService.retire_unowned(
+            tenant_id=app_model.tenant_id,
+            agent_ids=retirement_candidates,
+            account_id=account.id,
+        )
+        enqueue_agent_resource_collection(
+            tenant_id=app_model.tenant_id,
+            binding_ids=binding_ids,
+            home_snapshot_ids=home_snapshot_ids,
+        )
         app_draft_workflow_was_synced.send(app_model, synced_draft_workflow=draft_workflow)
 
         return draft_workflow
@@ -529,7 +552,7 @@ class WorkflowService:
         account: Account,
         marked_name: str = "",
         marked_comment: str = "",
-    ) -> Workflow:
+    ) -> tuple[Workflow, set[str]]:
         draft_workflow_stmt = select(Workflow).where(
             Workflow.tenant_id == app_model.tenant_id,
             Workflow.app_id == app_model.id,
@@ -588,7 +611,7 @@ class WorkflowService:
 
         # commit db session changes
         session.add(workflow)
-        WorkflowAgentPublishService.copy_agent_node_bindings_to_published(
+        retirement_candidates = WorkflowAgentPublishService.copy_agent_node_bindings_to_published(
             session=session,
             draft_workflow=draft_workflow,
             published_workflow=workflow,
@@ -598,7 +621,7 @@ class WorkflowService:
         app_published_workflow_was_updated.send(app_model, published_workflow=workflow)
 
         # return new workflow
-        return workflow
+        return workflow, retirement_candidates
 
     def _validate_workflow_credentials(self, workflow: Workflow, *, session: Session) -> None:
         """

api/tasks/agent_backend_session_cleanup_task.py

@@ -1,72 +0,0 @@
-"""Celery tasks that execute Agent backend lifecycle-only session cleanup."""
-
-from __future__ import annotations
-
-import logging
-
-from celery import shared_task
-
-from clients.agent_backend.factory import create_agent_backend_run_client
-from clients.agent_backend.request_builder import AgentBackendRunRequestBuilder
-from clients.agent_backend.session_cleanup import (
-    AgentBackendSessionCleanupPayload,
-    cleanup_agent_backend_session,
-)
-from configs import dify_config
-
-logger = logging.getLogger(__name__)
-
-
-def _create_agent_backend_client():
-    if not (dify_config.AGENT_BACKEND_USE_FAKE or dify_config.AGENT_BACKEND_BASE_URL):
-        return None
-    return create_agent_backend_run_client(
-        base_url=dify_config.AGENT_BACKEND_BASE_URL,
-        api_token=dify_config.AGENT_BACKEND_API_TOKEN,
-        use_fake=dify_config.AGENT_BACKEND_USE_FAKE,
-        fake_scenario=dify_config.AGENT_BACKEND_FAKE_SCENARIO,
-        stream_read_timeout_seconds=dify_config.AGENT_BACKEND_STREAM_READ_TIMEOUT_SECONDS,
-        stream_max_reconnects=dify_config.AGENT_BACKEND_STREAM_MAX_RECONNECTS,
-        stream_run_timeout_seconds=dify_config.AGENT_BACKEND_RUN_TIMEOUT_SECONDS,
-    )
-
-
-def _run_cleanup_task(payload_dict: dict[str, object]) -> None:
-    payload = AgentBackendSessionCleanupPayload.model_validate(payload_dict)
-    result = cleanup_agent_backend_session(
-        payload=payload,
-        client=_create_agent_backend_client(),
-        request_builder=AgentBackendRunRequestBuilder(),
-    )
-    if result.status == "succeeded":
-        return
-
-    log_fields = {
-        "tenant_id": payload.metadata.get("tenant_id"),
-        "app_id": payload.metadata.get("app_id"),
-        "workflow_run_id": payload.metadata.get("workflow_run_id"),
-        "node_id": payload.metadata.get("node_id"),
-        "conversation_id": payload.metadata.get("conversation_id"),
-        "agent_id": payload.metadata.get("agent_id"),
-        "previous_agent_backend_run_id": payload.metadata.get("previous_agent_backend_run_id"),
-        "failed_agent_backend_run_id": payload.metadata.get("failed_agent_backend_run_id"),
-        "cleanup_run_id": result.cleanup_run_id,
-        "reason": result.reason,
-    }
-    if result.status == "skipped":
-        logger.info("Agent backend session cleanup skipped: %s", log_fields)
-        return
-
-    logger.warning("Agent backend session cleanup failed: %s", log_fields)
-
-
-@shared_task(queue="workflow_storage")
-def cleanup_workflow_agent_runtime_session(payload_dict: dict[str, object]) -> None:
-    """Run one workflow-owned Agent backend cleanup payload."""
-    _run_cleanup_task(payload_dict)
-
-
-@shared_task(queue="conversation")
-def cleanup_conversation_agent_runtime_session(payload_dict: dict[str, object]) -> None:
-    """Run one conversation-owned Agent backend cleanup payload."""
-    _run_cleanup_task(payload_dict)

api/tasks/app_generate/resume_agent_app_task.py

@@ -51,6 +51,7 @@ def resume_agent_app_execution(*, conversation_id: str, form_id: str) -> None:
             app_model=app_model,
             user=user,
             conversation_id=conversation_id,
+            form_id=form_id,
             invoke_from=_resolve_invoke_from(conversation),
             session=db.session(),
         )

api/tasks/collect_agent_resources_task.py

@@ -0,0 +1,75 @@
+"""Asynchronously collect retired Agent working resources."""
+
+from __future__ import annotations
+
+import logging
+from collections.abc import Iterable
+
+from celery import shared_task
+
+from services.agent.home_snapshot_service import AgentHomeSnapshotService
+from services.agent.workspace_service import AgentWorkspaceService
+
+logger = logging.getLogger(__name__)
+
+
+@shared_task(queue="retention")
+def collect_agent_resources(
+    *,
+    tenant_id: str,
+    binding_ids: list[str],
+    workspace_ids: list[str],
+    home_snapshot_ids: list[str],
+) -> None:
+    """Collect only the explicitly identified RETIRED resources."""
+
+    collectors = (
+        (workspace_ids, "workspace_id", AgentWorkspaceService.collect_retired_workspace),
+        (binding_ids, "binding_id", AgentWorkspaceService.collect_retired_binding),
+        (
+            home_snapshot_ids,
+            "home_snapshot_id",
+            AgentHomeSnapshotService.collect_retired_home_snapshot,
+        ),
+    )
+    for resource_ids, argument_name, collector in collectors:
+        for resource_id in resource_ids:
+            try:
+                collector(tenant_id=tenant_id, **{argument_name: resource_id})
+            except Exception:
+                logger.exception(
+                    "Failed to collect retired Agent resource",
+                    extra={
+                        "tenant_id": tenant_id,
+                        "resource_type": argument_name.removesuffix("_id"),
+                        "resource_id": resource_id,
+                    },
+                )
+
+
+def enqueue_agent_resource_collection(
+    *,
+    tenant_id: str,
+    binding_ids: Iterable[str] = (),
+    workspace_ids: Iterable[str] = (),
+    home_snapshot_ids: Iterable[str] = (),
+) -> None:
+    """Best-effort enqueue of physical collection after retire has committed."""
+
+    payload = {
+        "binding_ids": sorted({resource_id for resource_id in binding_ids if resource_id}),
+        "workspace_ids": sorted({resource_id for resource_id in workspace_ids if resource_id}),
+        "home_snapshot_ids": sorted({resource_id for resource_id in home_snapshot_ids if resource_id}),
+    }
+    if not any(payload.values()):
+        return
+    try:
+        collect_agent_resources.delay(tenant_id=tenant_id, **payload)
+    except Exception:
+        logger.exception(
+            "Failed to enqueue retired Agent resource collection",
+            extra={"tenant_id": tenant_id, **payload},
+        )
+
+
+__all__ = ["collect_agent_resources", "enqueue_agent_resource_collection"]

api/tasks/remove_app_and_related_data_task.py

@@ -5,25 +5,17 @@ from typing import Any, cast
 
 import click
 import sqlalchemy as sa
-from agenton.compositor import CompositorSessionSnapshot
 from celery import shared_task
-from dify_agent.protocol import RuntimeLayerSpec
-from pydantic import JsonValue, TypeAdapter
 from sqlalchemy import delete, select
 from sqlalchemy.engine import CursorResult
 from sqlalchemy.exc import SQLAlchemyError
 from sqlalchemy.orm import sessionmaker
 
-from clients.agent_backend.session_cleanup import AgentBackendSessionCleanupPayload
 from configs import dify_config
 from core.db.session_factory import session_factory
 from extensions.ext_database import db
 from libs.archive_storage import ArchiveStorageNotConfiguredError, get_archive_storage
-from libs.datetime_utils import naive_utc_now
 from models import (
-    AgentRuntimeSession,
-    AgentRuntimeSessionOwnerType,
-    AgentRuntimeSessionStatus,
     ApiToken,
     AppAnnotationHitHistory,
     AppAnnotationSetting,
@@ -58,13 +50,8 @@ from models.workflow import (
 )
 from repositories.factory import DifyAPIRepositoryFactory
 from services.api_token_service import ApiTokenCache
-from tasks.agent_backend_session_cleanup_task import (
-    cleanup_conversation_agent_runtime_session,
-    cleanup_workflow_agent_runtime_session,
-)
 
 logger = logging.getLogger(__name__)
-_RUNTIME_LAYER_SPECS_ADAPTER: TypeAdapter[list[RuntimeLayerSpec]] = TypeAdapter(list[RuntimeLayerSpec])
 
 
 @shared_task(queue="app_deletion", bind=True, max_retries=3)
@@ -72,7 +59,6 @@ def remove_app_and_related_data_task(self, tenant_id: str, app_id: str):
     logger.info(click.style(f"Start deleting app and related data: {tenant_id}:{app_id}", fg="green"))
     start_at = time.perf_counter()
     try:
-        _cleanup_active_agent_runtime_sessions_for_app(tenant_id, app_id)
         # Delete related data
         _delete_app_model_configs(tenant_id, app_id)
         _delete_app_site(tenant_id, app_id)
@@ -113,143 +99,6 @@ def remove_app_and_related_data_task(self, tenant_id: str, app_id: str):
         raise self.retry(exc=e, countdown=60)  # Retry after 60 seconds
 
 
-def _cleanup_active_agent_runtime_sessions_for_app(tenant_id: str, app_id: str, *, batch_size: int = 100) -> None:
-    """Best-effort fan-out for ACTIVE Agent runtime sessions during app deletion.
-
-    App deletion must not block on synchronous Agent backend lifecycle work, so
-    this helper scans ACTIVE ``agent_runtime_sessions`` rows in batches,
-    dispatches owner-specific cleanup tasks only when enough persisted data
-    exists to replay a lifecycle-only run, and then marks each visited row
-    ``CLEANED`` locally regardless of enqueue outcome. The local retirement is
-    the contract that lets the rest of app deletion continue even when backend
-    cleanup dispatch is skipped or fails.
-    """
-    if batch_size <= 0:
-        raise ValueError("batch_size must be positive")
-
-    while True:
-        with session_factory.create_session() as session:
-            row_ids = session.scalars(
-                select(AgentRuntimeSession.id)
-                .where(
-                    AgentRuntimeSession.tenant_id == tenant_id,
-                    AgentRuntimeSession.app_id == app_id,
-                    AgentRuntimeSession.status == AgentRuntimeSessionStatus.ACTIVE,
-                )
-                .order_by(AgentRuntimeSession.updated_at.asc())
-                .limit(batch_size)
-            ).all()
-
-        if not row_ids:
-            return
-
-        retired_count = 0
-        for row_id in row_ids:
-            with session_factory.create_session() as session:
-                row = session.get(AgentRuntimeSession, row_id)
-                if row is None or row.status != AgentRuntimeSessionStatus.ACTIVE:
-                    retired_count += 1
-                    continue
-
-                try:
-                    payload = _build_agent_runtime_session_cleanup_payload(row)
-                    if payload is not None:
-                        _enqueue_agent_runtime_session_cleanup(row=row, payload=payload)
-                except Exception:
-                    logger.warning(
-                        "Failed to enqueue Agent backend cleanup during app deletion: "
-                        "tenant_id=%s app_id=%s owner_type=%s conversation_id=%s workflow_run_id=%s "
-                        "node_id=%s agent_id=%s backend_run_id=%s",
-                        row.tenant_id,
-                        row.app_id,
-                        row.owner_type,
-                        row.conversation_id,
-                        row.workflow_run_id,
-                        row.node_id,
-                        row.agent_id,
-                        row.backend_run_id,
-                        exc_info=True,
-                    )
-                finally:
-                    try:
-                        row.status = AgentRuntimeSessionStatus.CLEANED
-                        row.cleaned_at = naive_utc_now()
-                        session.commit()
-                        retired_count += 1
-                    except Exception:
-                        session.rollback()
-                        logger.warning(
-                            "Failed to retire Agent runtime session during app deletion: "
-                            "tenant_id=%s app_id=%s owner_type=%s conversation_id=%s workflow_run_id=%s "
-                            "node_id=%s agent_id=%s backend_run_id=%s",
-                            row.tenant_id,
-                            row.app_id,
-                            row.owner_type,
-                            row.conversation_id,
-                            row.workflow_run_id,
-                            row.node_id,
-                            row.agent_id,
-                            row.backend_run_id,
-                            exc_info=True,
-                        )
-
-        if retired_count == 0:
-            logger.warning(
-                "Failed to retire any active Agent runtime sessions during app deletion: tenant_id=%s app_id=%s",
-                tenant_id,
-                app_id,
-            )
-            return
-
-
-def _build_agent_runtime_session_cleanup_payload(
-    row: AgentRuntimeSession,
-) -> AgentBackendSessionCleanupPayload | None:
-    runtime_layer_specs = _RUNTIME_LAYER_SPECS_ADAPTER.validate_json(row.composition_layer_specs or "[]")
-    if not runtime_layer_specs:
-        return None
-
-    metadata: dict[str, JsonValue] = {
-        "tenant_id": row.tenant_id,
-        "app_id": row.app_id,
-        "agent_id": row.agent_id,
-        "agent_config_snapshot_id": row.agent_config_snapshot_id,
-        "previous_agent_backend_run_id": row.backend_run_id,
-    }
-    if row.owner_type == AgentRuntimeSessionOwnerType.CONVERSATION:
-        metadata["conversation_id"] = row.conversation_id
-        idempotency_key = (
-            f"{row.tenant_id}:{row.app_id}:{row.conversation_id}:"
-            f"{row.agent_id}:app-delete-cleanup:{row.id or row.backend_run_id or 'no-session-id'}"
-        )
-    else:
-        metadata["workflow_run_id"] = row.workflow_run_id
-        metadata["node_id"] = row.node_id
-        idempotency_key = (
-            f"{row.tenant_id}:{row.app_id}:{row.workflow_run_id}:{row.node_id}:"
-            f"{row.agent_id}:app-delete-cleanup:{row.id or row.backend_run_id or 'no-session-id'}"
-        )
-
-    return AgentBackendSessionCleanupPayload(
-        session_snapshot=CompositorSessionSnapshot.model_validate_json(row.session_snapshot),
-        runtime_layer_specs=runtime_layer_specs,
-        idempotency_key=idempotency_key,
-        metadata=metadata,
-    )
-
-
-def _enqueue_agent_runtime_session_cleanup(
-    *,
-    row: AgentRuntimeSession,
-    payload: AgentBackendSessionCleanupPayload,
-) -> None:
-    payload_dict = payload.model_dump(mode="json")
-    if row.owner_type == AgentRuntimeSessionOwnerType.CONVERSATION:
-        cleanup_conversation_agent_runtime_session.delay(payload_dict)
-        return
-    cleanup_workflow_agent_runtime_session.delay(payload_dict)
-
-
 def _delete_app_model_configs(tenant_id: str, app_id: str):
     def del_model_config(session, model_config_id: str):
         session.execute(

api/tasks/workflow_node_execution_tasks.py

@@ -13,6 +13,7 @@ from celery import shared_task
 from sqlalchemy import select
 
 from core.db.session_factory import session_factory
+from core.workflow.node_execution_process_data import preserve_workflow_agent_binding_id
 from graphon.entities.workflow_node_execution import (
     WorkflowNodeExecution,
 )
@@ -144,8 +145,9 @@ def _update_node_execution_from_domain(node_execution: WorkflowNodeExecutionMode
     # Update serialized data
     json_converter = WorkflowRuntimeTypeConverter()
     node_execution.inputs = json.dumps(json_converter.to_json_encodable(execution.inputs)) if execution.inputs else "{}"
+    process_data = preserve_workflow_agent_binding_id(node_execution.process_data_dict, execution.process_data)
     node_execution.process_data = (
-        json.dumps(json_converter.to_json_encodable(execution.process_data)) if execution.process_data else "{}"
+        json.dumps(json_converter.to_json_encodable(process_data)) if process_data is not None else "{}"
     )
     node_execution.outputs = (
         json.dumps(json_converter.to_json_encodable(execution.outputs)) if execution.outputs else "{}"

api/tests/test_containers_integration_tests/.ruff.toml

@@ -2,114 +2,94 @@ extend = "../../.ruff.toml"
 src = ["../.."]
 
 [lint]
-extend-select = ["ANN401", "ARG", "TID251"]
+extend-select = ["ANN401", "ARG"]
 
+# Existing strict-mode debt. Remove a file entry when bringing it under strict checking.
 [lint.per-file-ignores]
-"core/rag/pipeline/test_queue_integration.py" = ["ANN401", "TID251", "ARG"]
+"controllers/console/test_apikey.py" = ["ARG002"]
+"controllers/openapi/test_app_dsl.py" = ["ARG002"]
+"controllers/service_api/dataset/test_dataset.py" = ["ARG002"]
+"controllers/web/test_conversation.py" = ["ARG002"]
+"controllers/web/test_human_input_form.py" = ["ARG001"]
+"controllers/web/test_wraps.py" = ["ARG002"]
+"core/app/layers/test_pause_state_persist_layer.py" = ["ARG002"]
+"core/rag/pipeline/test_queue_integration.py" = ["ARG002", "TID251"]
+"core/rag/retrieval/test_dataset_retrieval_integration.py" = ["ARG002"]
+"models/test_conversation_message_inputs.py" = ["ARG001"]
 "models/test_types_enum_text.py" = ["ANN401", "TID251"]
-"services/test_app_dsl_service.py" = ["ANN401", "TID251", "ARG"]
-"services/test_file_service_zip_and_lookup.py" = ["ANN401", "TID251", "ARG"]
-"trigger/conftest.py" = ["ANN401", "TID251"]
-"trigger/test_trigger_e2e.py" = ["ANN401", "TID251", "ARG"]
-"controllers/console/app/test_app_apis.py" = ["ARG"]
-"controllers/console/app/test_app_import_api.py" = ["ARG"]
-"controllers/console/auth/test_oauth.py" = ["ARG"]
-"controllers/console/auth/test_password_reset.py" = ["ARG"]
-"controllers/console/datasets/test_data_source.py" = ["ARG"]
-"controllers/console/test_apikey.py" = ["ARG"]
-"controllers/console/workspace/test_tool_provider.py" = ["ARG"]
-"controllers/mcp/test_mcp.py" = ["ARG"]
-"controllers/openapi/test_app_dsl.py" = ["ARG"]
-"controllers/openapi/test_workspaces.py" = ["ARG"]
-"controllers/service_api/dataset/test_dataset.py" = ["ARG"]
-"controllers/web/test_conversation.py" = ["ARG"]
-"controllers/web/test_human_input_form.py" = ["ARG"]
-"controllers/web/test_wraps.py" = ["ARG"]
-"core/app/layers/test_pause_state_persist_layer.py" = ["ARG"]
-"core/rag/retrieval/test_dataset_retrieval_integration.py" = ["ARG"]
-"models/test_conversation_message_inputs.py" = ["ARG"]
-"models/test_conversation_status_count.py" = ["ARG"]
-"repositories/test_sqlalchemy_api_workflow_run_repository.py" = ["ARG"]
-"repositories/test_workflow_run_repository.py" = ["ARG"]
-"services/auth/test_api_key_auth_service.py" = ["ARG"]
-"services/auth/test_auth_integration.py" = ["ARG"]
-"services/dataset_collection_binding.py" = ["ARG"]
-"services/dataset_service_update_delete.py" = ["ARG"]
-"services/document_service_status.py" = ["ARG"]
-"services/enterprise/test_account_deletion_sync.py" = ["ARG"]
-"services/plugin/test_plugin_parameter_service.py" = ["ARG"]
-"services/plugin/test_plugin_service.py" = ["ARG"]
-"services/rag_pipeline/test_rag_pipeline_service_db.py" = ["ARG"]
-"services/recommend_app/test_database_retrieval.py" = ["ARG"]
-"services/test_account_service.py" = ["ARG"]
-"services/test_advanced_prompt_template_service.py" = ["ARG"]
-"services/test_annotation_service.py" = ["ARG"]
-"services/test_api_based_extension_service.py" = ["ARG"]
-"services/test_api_token_service.py" = ["ARG"]
-"services/test_app_generate_service.py" = ["ARG"]
-"services/test_app_service.py" = ["ARG"]
-"services/test_attachment_service.py" = ["ARG"]
-"services/test_conversation_variable_updater.py" = ["ARG"]
-"services/test_dataset_permission_service.py" = ["ARG"]
-"services/test_dataset_service_batch_update_document_status.py" = ["ARG"]
-"services/test_dataset_service_retrieval.py" = ["ARG"]
-"services/test_delete_archived_workflow_run.py" = ["ARG"]
-"services/test_document_service_rename_document.py" = ["ARG"]
-"services/test_end_user_service.py" = ["ARG"]
-"services/test_feature_service.py" = ["ARG"]
-"services/test_feedback_service.py" = ["ARG"]
-"services/test_file_service.py" = ["ARG"]
-"services/test_human_input_delivery_test_service.py" = ["ARG"]
-"services/test_message_service.py" = ["ARG"]
-"services/test_messages_clean_service.py" = ["ARG", "S110"]
-"services/test_metadata_partial_update.py" = ["ARG"]
-"services/test_metadata_service.py" = ["ARG"]
-"services/test_model_load_balancing_service.py" = ["ARG"]
-"services/test_model_provider_service.py" = ["ARG"]
-"services/test_oauth_server_service.py" = ["ARG"]
-"services/test_ops_service.py" = ["ARG"]
-"services/test_saved_message_service.py" = ["ARG"]
-"services/test_web_conversation_service.py" = ["ARG"]
-"services/test_webapp_auth_service.py" = ["ARG"]
-"services/test_webhook_service.py" = ["ARG"]
-"services/test_workflow_app_service.py" = ["ARG"]
-"services/test_workflow_draft_variable_service.py" = ["ARG"]
-"services/test_workflow_run_service.py" = ["ARG"]
-"services/test_workflow_service.py" = ["ARG"]
-"services/test_workspace_service.py" = ["ARG"]
-"services/tools/test_api_tools_manage_service.py" = ["ARG"]
-"services/tools/test_mcp_tools_manage_service.py" = ["ARG"]
-"services/tools/test_tools_transform_service.py" = ["ARG"]
-"services/workflow/test_workflow_converter.py" = ["ARG"]
-"tasks/test_add_document_to_index_task.py" = ["ARG"]
-"tasks/test_batch_clean_document_task.py" = ["ARG"]
-"tasks/test_batch_create_segment_to_index_task.py" = ["ARG"]
+"repositories/test_sqlalchemy_api_workflow_run_repository.py" = ["ARG002", "ARG005"]
+"repositories/test_workflow_run_repository.py" = ["ARG002"]
+"services/auth/test_auth_integration.py" = ["ARG002"]
+"services/dataset_collection_binding.py" = ["ARG002"]
+"services/document_service_status.py" = ["ARG002"]
+"services/rag_pipeline/test_rag_pipeline_service_db.py" = ["ARG002"]
+"services/recommend_app/test_database_retrieval.py" = ["ARG002"]
+"services/test_account_service.py" = ["ARG002"]
+"services/test_advanced_prompt_template_service.py" = ["ARG002"]
+"services/test_app_dsl_service.py" = ["ANN401", "ARG001", "ARG002", "ARG005", "TID251"]
+"services/test_app_service.py" = ["ARG002"]
+"services/test_attachment_service.py" = ["ARG002"]
+"services/test_conversation_variable_updater.py" = ["ARG002"]
+"services/test_dataset_service_batch_update_document_status.py" = ["ARG002"]
+"services/test_delete_archived_workflow_run.py" = ["ARG002"]
+"services/test_document_service_rename_document.py" = ["ARG001"]
+"services/test_end_user_service.py" = ["ARG002"]
+"services/test_feature_service.py" = ["ARG002"]
+"services/test_file_service.py" = ["ARG002"]
+"services/test_file_service_zip_and_lookup.py" = ["TID251"]
+"services/test_messages_clean_service.py" = ["ARG002", "S110"]
+"services/test_metadata_partial_update.py" = ["ARG002"]
+"services/test_metadata_service.py" = ["ARG002"]
+"services/test_model_load_balancing_service.py" = ["ARG002"]
+"services/test_model_provider_service.py" = ["ARG002"]
+"services/test_ops_service.py" = ["ARG002"]
+"services/test_webapp_auth_service.py" = ["ARG002"]
+"services/test_webhook_service.py" = ["ARG002"]
+"services/test_workflow_draft_variable_service.py" = ["ARG002"]
+"services/test_workflow_run_service.py" = ["ARG002"]
+"services/test_workflow_service.py" = ["ARG002"]
+"services/test_workspace_service.py" = ["ARG002"]
+"services/tools/test_api_tools_manage_service.py" = ["ARG002"]
+"services/tools/test_mcp_tools_manage_service.py" = ["ARG002", "ARG005"]
+"services/tools/test_tools_transform_service.py" = ["ARG002"]
+"services/workflow/test_workflow_converter.py" = ["ARG002"]
+"tasks/test_add_document_to_index_task.py" = ["ARG002"]
+"tasks/test_batch_clean_document_task.py" = ["ARG002"]
+"tasks/test_batch_create_segment_to_index_task.py" = ["ARG001", "ARG002"]
 "tasks/test_clean_dataset_task.py" = ["T201"]
-"tasks/test_clean_notion_document_task.py" = ["ARG"]
-"tasks/test_create_segment_to_index_task.py" = ["ARG"]
-"tasks/test_dataset_indexing_task.py" = ["ARG"]
-"tasks/test_deal_dataset_vector_index_task.py" = ["ARG"]
-"tasks/test_delete_segment_from_index_task.py" = ["ARG"]
-"tasks/test_disable_segment_from_index_task.py" = ["ARG"]
-"tasks/test_disable_segments_from_index_task.py" = ["ARG"]
-"tasks/test_document_indexing_sync_task.py" = ["ARG"]
-"tasks/test_document_indexing_task.py" = ["ARG"]
-"tasks/test_document_indexing_update_task.py" = ["ARG"]
-"tasks/test_duplicate_document_indexing_task.py" = ["ARG"]
-"tasks/test_enable_segments_to_index_task.py" = ["ARG"]
-"tasks/test_mail_change_mail_task.py" = ["ARG"]
-"tasks/test_mail_email_code_login_task.py" = ["ARG"]
-"tasks/test_mail_human_input_delivery_task.py" = ["ARG"]
-"tasks/test_mail_inner_task.py" = ["ARG"]
-"tasks/test_mail_invite_member_task.py" = ["ARG"]
-"tasks/test_mail_owner_transfer_task.py" = ["ARG"]
-"tasks/test_mail_register_task.py" = ["ARG"]
-"tasks/test_rag_pipeline_run_tasks.py" = ["ARG"]
+"tasks/test_clean_notion_document_task.py" = ["ARG002"]
+"tasks/test_create_segment_to_index_task.py" = ["ARG002"]
+"tasks/test_dataset_indexing_task.py" = ["ARG002"]
+"tasks/test_deal_dataset_vector_index_task.py" = ["ARG002"]
+"tasks/test_delete_segment_from_index_task.py" = ["ARG002"]
+"tasks/test_disable_segment_from_index_task.py" = ["ARG002"]
+"tasks/test_disable_segments_from_index_task.py" = ["ARG002"]
+"tasks/test_document_indexing_sync_task.py" = ["ARG002"]
+"tasks/test_document_indexing_task.py" = ["ARG002"]
+"tasks/test_document_indexing_update_task.py" = ["ARG002"]
+"tasks/test_duplicate_document_indexing_task.py" = ["ARG002"]
+"tasks/test_enable_segments_to_index_task.py" = ["ARG002"]
+"tasks/test_mail_change_mail_task.py" = ["ARG002"]
+"tasks/test_mail_email_code_login_task.py" = ["ARG002"]
+"tasks/test_mail_human_input_delivery_task.py" = ["ARG001"]
+"tasks/test_mail_inner_task.py" = ["ARG002"]
+"tasks/test_mail_invite_member_task.py" = ["ARG002"]
+"tasks/test_mail_owner_transfer_task.py" = ["ARG002"]
+"tasks/test_mail_register_task.py" = ["ARG002"]
+"tasks/test_rag_pipeline_run_tasks.py" = ["ARG002"]
 "test_workflow_pause_integration.py" = ["T201"]
-"workflow/nodes/code_executor/test_code_javascript.py" = ["ARG"]
-"workflow/nodes/code_executor/test_code_jinja2.py" = ["ARG"]
-"workflow/nodes/code_executor/test_code_python3.py" = ["ARG"]
+"trigger/conftest.py" = ["ANN401", "TID251"]
+"trigger/test_trigger_e2e.py" = ["ANN401", "ARG001", "TID251"]
+"workflow/nodes/code_executor/test_code_javascript.py" = ["ARG002"]
+"workflow/nodes/code_executor/test_code_jinja2.py" = ["ARG002"]
+"workflow/nodes/code_executor/test_code_python3.py" = ["ARG002"]
 "workflow/nodes/code_executor/test_utils.py" = ["T201"]
 
+[lint.flake8-tidy-imports.banned-api."flask_restx.reqparse"]
+msg = "Use Pydantic payload/query models instead of reqparse."
+
+[lint.flake8-tidy-imports.banned-api."flask_restx.reqparse.RequestParser"]
+msg = "Use Pydantic payload/query models instead of reqparse."
+
 [lint.flake8-tidy-imports.banned-api."typing.Any"]
 msg = "Use object, Protocol, TypedDict, TypeVar, ParamSpec, or a localized cast instead."

api/tests/test_containers_integration_tests/pyrefly.toml

@@ -1,42 +1,24 @@
 preset = "strict"
-strict-callable-subtyping = true
 project-includes = ["."]
 search-path = ["../.."]
+python-platform = "linux"
+python-version = "3.12.0"
+infer-with-first-use = true
+min-severity = "warn"
 
-# Verify project-excludes from the repo root:
-#   tmp_config=$(mktemp --tmpdir=api/tests/test_containers_integration_tests pyrefly-no-excludes.XXXXXX.toml)
-#   awk 'BEGIN {skip=0} /^project-excludes = \[/ {skip=1; next} skip && /^\]/ {skip=0; next} !skip {print}' api/tests/test_containers_integration_tests/pyrefly.toml > "$tmp_config"
-#   tmp_name=$(basename "$tmp_config")
-#   comm -3 <(sed -n 's/^  "\(.*\)",$/\1/p' api/tests/test_containers_integration_tests/pyrefly.toml | sort) <(uv --directory api run pyrefly check --config "tests/test_containers_integration_tests/$tmp_name" --summary=none --output-format=min-text 2>/dev/null | rg '^ERROR ' | sed -E 's#^ERROR (tests/test_containers_integration_tests/[^:]+):.*#\1#' | sed 's#^tests/test_containers_integration_tests/##' | sort -u)
-#   rm --force "$tmp_config"
+# Existing strict-mode debt. Remove a file when bringing it under strict checking.
 project-excludes = [
   "commands/test_legacy_model_type_migration.py",
-  "controllers/console/app/test_app_apis.py",
-  "controllers/console/app/test_app_import_api.py",
   "controllers/console/app/test_chat_conversation_status_count_api.py",
   "controllers/console/app/test_conversation_read_timestamp.py",
   "controllers/console/app/test_workflow_draft_variable.py",
-  "controllers/console/auth/test_email_register.py",
-  "controllers/console/auth/test_forgot_password.py",
-  "controllers/console/auth/test_oauth.py",
-  "controllers/console/auth/test_password_reset.py",
-  "controllers/console/datasets/rag_pipeline/test_rag_pipeline.py",
-  "controllers/console/datasets/rag_pipeline/test_rag_pipeline_datasets.py",
-  "controllers/console/datasets/rag_pipeline/test_rag_pipeline_import.py",
-  "controllers/console/datasets/rag_pipeline/test_rag_pipeline_workflow.py",
-  "controllers/console/datasets/test_data_source.py",
-  "controllers/console/explore/test_conversation.py",
   "controllers/console/test_api_based_extension.py",
   "controllers/console/test_apikey.py",
-  "controllers/console/workspace/test_tool_provider.py",
-  "controllers/console/workspace/test_trigger_providers.py",
   "controllers/console/workspace/test_workspace_wraps.py",
-  "controllers/mcp/test_mcp.py",
   "controllers/service_api/dataset/test_dataset.py",
   "controllers/service_api/test_site.py",
   "controllers/web/test_conversation.py",
   "controllers/web/test_site.py",
-  "controllers/web/test_web_forgot_password.py",
   "controllers/web/test_wraps.py",
   "core/app/layers/test_pause_state_persist_layer.py",
   "core/rag/pipeline/test_queue_integration.py",
@@ -57,22 +39,16 @@ project-excludes = [
   "repositories/test_sqlalchemy_execution_extra_content_repository.py",
   "repositories/test_sqlalchemy_workflow_node_execution_repository.py",
   "repositories/test_workflow_run_repository.py",
-  "services/auth/test_api_key_auth_service.py",
   "services/auth/test_auth_integration.py",
   "services/dataset_collection_binding.py",
   "services/dataset_service_update_delete.py",
   "services/document_service_status.py",
-  "services/enterprise/test_account_deletion_sync.py",
-  "services/plugin/test_plugin_parameter_service.py",
-  "services/plugin/test_plugin_service.py",
-  "services/rag_pipeline/test_rag_pipeline_service_db.py",
   "services/recommend_app/test_database_retrieval.py",
   "services/test_account_service.py",
   "services/test_advanced_prompt_template_service.py",
   "services/test_agent_service.py",
   "services/test_annotation_service.py",
   "services/test_api_based_extension_service.py",
-  "services/test_api_token_service.py",
   "services/test_app_dsl_service.py",
   "services/test_app_generate_service.py",
   "services/test_app_service.py",
@@ -97,20 +73,15 @@ project-excludes = [
   "services/test_document_service_rename_document.py",
   "services/test_end_user_service.py",
   "services/test_feature_service.py",
-  "services/test_feedback_service.py",
   "services/test_file_service.py",
   "services/test_human_input_delivery_test.py",
-  "services/test_human_input_delivery_test_service.py",
   "services/test_message_export_service.py",
   "services/test_message_service.py",
   "services/test_message_service_execution_extra_content.py",
   "services/test_message_service_extra_contents.py",
   "services/test_messages_clean_service.py",
-  "services/test_metadata_partial_update.py",
-  "services/test_metadata_service.py",
   "services/test_model_load_balancing_service.py",
   "services/test_model_provider_service.py",
-  "services/test_oauth_server_service.py",
   "services/test_ops_service.py",
   "services/test_restore_archived_workflow_run.py",
   "services/test_saved_message_service.py",
@@ -124,7 +95,6 @@ project-excludes = [
   "services/test_workflow_run_service.py",
   "services/test_workflow_service.py",
   "services/test_workspace_service.py",
-  "services/tools/test_api_tools_manage_service.py",
   "services/tools/test_mcp_tools_manage_service.py",
   "services/tools/test_tools_transform_service.py",
   "services/tools/test_workflow_tools_manage_service.py",
@@ -161,7 +131,6 @@ project-excludes = [
   "test_workflow_pause_integration.py",
   "trigger/conftest.py",
   "trigger/test_trigger_e2e.py",
-  "workflow/nodes/code_executor/test_code_executor.py",
   "workflow/nodes/code_executor/test_code_javascript.py",
   "workflow/nodes/code_executor/test_code_jinja2.py",
   "workflow/nodes/code_executor/test_code_python3.py",
@@ -169,6 +138,7 @@ project-excludes = [
 ]
 
 [errors]
+missing-override-decorator = "error"
 redundant-cast = true
 unannotated-return = true
 unnecessary-type-conversion = true

api/tests/test_containers_integration_tests/services/test_app_dsl_service.py

@@ -124,6 +124,7 @@ class TestAppDslService:
             patch("services.app_service.ModelManager.for_tenant") as mock_model_manager,
             patch("services.app_service.FeatureService") as mock_feature_service,
             patch("services.app_service.EnterpriseService") as mock_enterprise_service,
+            patch("services.agent.home_snapshot_service.AgentHomeSnapshotService._client") as mock_home_snapshot_client,
         ):
             mock_workflow_service.return_value.get_draft_workflow.return_value = None
             mock_workflow_service.return_value.sync_draft_workflow.return_value = MagicMock()
@@ -142,6 +143,9 @@ class TestAppDslService:
             mock_feature_service.get_system_features.return_value.webapp_auth.enabled = False
             mock_enterprise_service.WebAppAuth.update_app_access_mode.return_value = None
             mock_enterprise_service.WebAppAuth.cleanup_webapp.return_value = None
+            mock_home_snapshot_client.return_value.__enter__.return_value.initialize_home_snapshot_sync.side_effect = (
+                lambda request: SimpleNamespace(snapshot_ref=f"test:{request.home_snapshot_id}")
+            )
 
             yield {
                 "workflow_service": mock_workflow_service,
@@ -1034,7 +1038,9 @@ class TestAppDslService:
             )
         )
 
-        imported_graph, warnings = AgentDslService(db_session_with_containers).import_workflow_packages(
+        imported_graph, warnings, retirement_candidates = AgentDslService(
+            db_session_with_containers
+        ).import_workflow_packages(
             workflow=workflow,
             portable_graph=graph,
             raw_packages={"agent_1": package.model_dump(mode="json")},
@@ -1043,6 +1049,7 @@ class TestAppDslService:
         db_session_with_containers.commit()
 
         assert warnings == []
+        assert retirement_candidates == set()
         graph_bindings = [node["data"]["agent_binding"] for node in imported_graph["nodes"]]
         assert all(binding["binding_type"] == WorkflowAgentBindingType.INLINE_AGENT.value for binding in graph_bindings)
         assert len({binding["agent_id"] for binding in graph_bindings}) == 2

api/tests/test_containers_integration_tests/services/test_app_service.py

@@ -1,4 +1,5 @@
 from datetime import datetime
+from types import SimpleNamespace
 from unittest.mock import create_autospec, patch
 
 import pytest
@@ -29,6 +30,7 @@ class TestAppService:
             patch("services.app_service.EnterpriseService") as mock_enterprise_service,
             patch("services.app_service.ModelManager.for_tenant") as mock_model_manager,
             patch("services.account_service.FeatureService") as mock_account_feature_service,
+            patch("services.agent.home_snapshot_service.AgentHomeSnapshotService._client") as mock_home_snapshot_client,
         ):
             # Setup default mock returns for app service
             mock_feature_service.get_system_features.return_value.webapp_auth.enabled = False
@@ -42,6 +44,9 @@ class TestAppService:
             mock_model_instance = mock_model_manager.return_value
             mock_model_instance.get_default_model_instance.return_value = None
             mock_model_instance.get_default_provider_model_name.return_value = ("openai", "gpt-3.5-turbo")
+            mock_home_snapshot_client.return_value.__enter__.return_value.initialize_home_snapshot_sync.side_effect = (
+                lambda request: SimpleNamespace(snapshot_ref=f"test:{request.home_snapshot_id}")
+            )
 
             yield {
                 "feature_service": mock_feature_service,

api/tests/test_containers_integration_tests/services/test_conversation_service.py

@@ -6,13 +6,11 @@ from unittest.mock import patch
 from uuid import uuid4
 
 import pytest
-from agenton.compositor import CompositorSessionSnapshot
 from sqlalchemy import select
 from sqlalchemy.orm import Session
 
-from core.app.apps.agent_app.session_store import AgentAppRuntimeSessionStore
 from core.app.entities.app_invoke_entities import InvokeFrom
-from models import AgentRuntimeSession, AgentRuntimeSessionOwnerType, AgentRuntimeSessionStatus, TenantAccountRole
+from models import TenantAccountRole
 from models.account import Account, Tenant, TenantAccountJoin
 from models.enums import ConversationFromSource, EndUserType
 from models.model import App, Conversation, EndUser, Message, MessageAnnotation
@@ -1077,9 +1075,8 @@ class TestConversationServiceExport:
         # Assert
         assert result == conversation
 
-    @patch("services.conversation_service.cleanup_conversation_agent_runtime_session")
     @patch("services.conversation_service.delete_conversation_related_data")
-    def test_delete_conversation(self, mock_delete_task, mock_cleanup_task, db_session_with_containers: Session):
+    def test_delete_conversation(self, mock_delete_task, db_session_with_containers: Session):
         """
         Test conversation deletion with async cleanup.
 
@@ -1098,20 +1095,6 @@ class TestConversationServiceExport:
             user,
         )
         conversation_id = conversation.id
-        runtime_session = AgentRuntimeSession(
-            tenant_id=app_model.tenant_id,
-            app_id=app_model.id,
-            owner_type=AgentRuntimeSessionOwnerType.CONVERSATION,
-            agent_id=str(uuid4()),
-            agent_config_snapshot_id=str(uuid4()),
-            backend_run_id="backend-run-1",
-            session_snapshot=CompositorSessionSnapshot(layers=[]).model_dump_json(),
-            composition_layer_specs='[{"name":"history","type":"pydantic_ai.history","deps":{},"metadata":{},"config":null}]',
-            conversation_id=conversation.id,
-            status=AgentRuntimeSessionStatus.ACTIVE,
-        )
-        db_session_with_containers.add(runtime_session)
-        db_session_with_containers.commit()
 
         # Act - Delete the conversation
         ConversationService.delete(
@@ -1126,27 +1109,11 @@ class TestConversationServiceExport:
         # Step 2: Async cleanup task triggered
         # The Celery task will handle cleanup of messages, annotations, etc.
         mock_delete_task.delay.assert_called_once_with(conversation_id)
-        mock_cleanup_task.delay.assert_called_once()
-        cleanup_payload = mock_cleanup_task.delay.call_args.args[0]
-        assert cleanup_payload["metadata"]["conversation_id"] == conversation_id
-        assert (
-            cleanup_payload["idempotency_key"]
-            == f"{app_model.tenant_id}:{app_model.id}:{conversation_id}:agent-runtime-session-cleanup:"
-            f"{runtime_session.agent_id}:{runtime_session.agent_config_snapshot_id}:{runtime_session.backend_run_id}"
-        )
 
-        runtime_session_row = db_session_with_containers.scalar(
-            select(AgentRuntimeSession).where(AgentRuntimeSession.id == runtime_session.id)
-        )
-        assert runtime_session_row is not None
-        assert runtime_session_row.status == AgentRuntimeSessionStatus.CLEANED
-
-    @patch("services.conversation_service.cleanup_conversation_agent_runtime_session")
     @patch("services.conversation_service.delete_conversation_related_data")
     def test_delete_conversation_not_owned_by_account(
         self,
         mock_delete_task,
-        mock_cleanup_task,
         db_session_with_containers: Session,
     ):
         """
@@ -1178,22 +1145,17 @@ class TestConversationServiceExport:
         not_deleted = db_session_with_containers.scalar(select(Conversation).where(Conversation.id == conversation.id))
         assert not_deleted is not None
         mock_delete_task.delay.assert_not_called()
-        mock_cleanup_task.delay.assert_not_called()
 
-    @patch("services.conversation_service.cleanup_conversation_agent_runtime_session")
     @patch("services.conversation_service.delete_conversation_related_data")
     def test_delete_handles_exception_and_rollback(
         self,
         mock_delete_task,
-        mock_cleanup_task,
         db_session_with_containers: Session,
     ):
         """
         Test that delete propagates exceptions and does not trigger the cleanup task.
 
-        When a DB error occurs during deletion, the conversation row stays in
-        place, but any already-enqueued Agent backend cleanup remains a
-        best-effort terminal lifecycle action.
+        When a DB error occurs during deletion, the conversation row stays in place.
         """
         # Arrange
         app_model, user = ConversationServiceIntegrationTestDataFactory.create_app_and_account(
@@ -1203,20 +1165,6 @@ class TestConversationServiceExport:
             db_session_with_containers, app_model, user
         )
         conversation_id = conversation.id
-        runtime_session = AgentRuntimeSession(
-            tenant_id=app_model.tenant_id,
-            app_id=app_model.id,
-            owner_type=AgentRuntimeSessionOwnerType.CONVERSATION,
-            agent_id=str(uuid4()),
-            agent_config_snapshot_id=str(uuid4()),
-            backend_run_id="backend-run-rollback",
-            session_snapshot=CompositorSessionSnapshot(layers=[]).model_dump_json(),
-            composition_layer_specs='[{"name":"history","type":"pydantic_ai.history","deps":{},"metadata":{},"config":null}]',
-            conversation_id=conversation.id,
-            status=AgentRuntimeSessionStatus.ACTIVE,
-        )
-        db_session_with_containers.add(runtime_session)
-        db_session_with_containers.commit()
 
         # Act — force an error during the delete to exercise the rollback path
         with patch.object(db_session_with_containers, "delete", side_effect=Exception("DB error")):
@@ -1228,111 +1176,9 @@ class TestConversationServiceExport:
                     session=db_session_with_containers,
                 )
 
-        # Assert — related-data deletion is not scheduled, but the backend
-        # cleanup task was already enqueued before the row delete failed.
+        # Assert — related-data deletion is not scheduled.
         mock_delete_task.delay.assert_not_called()
-        mock_cleanup_task.delay.assert_called_once()
-        cleanup_payload = mock_cleanup_task.delay.call_args.args[0]
-        assert (
-            cleanup_payload["idempotency_key"]
-            == f"{app_model.tenant_id}:{app_model.id}:{conversation_id}:agent-runtime-session-cleanup:"
-            f"{runtime_session.agent_id}:{runtime_session.agent_config_snapshot_id}:{runtime_session.backend_run_id}"
-        )
 
         # Conversation is still present because the deletion was never committed
         still_there = db_session_with_containers.scalar(select(Conversation).where(Conversation.id == conversation_id))
         assert still_there is not None
-
-    @patch("services.conversation_service.cleanup_conversation_agent_runtime_session")
-    @patch("services.conversation_service.delete_conversation_related_data")
-    def test_delete_ignores_mark_cleaned_failure(
-        self,
-        mock_delete_task,
-        mock_cleanup_task,
-        db_session_with_containers: Session,
-    ):
-        app_model, user = ConversationServiceIntegrationTestDataFactory.create_app_and_account(
-            db_session_with_containers
-        )
-        conversation = ConversationServiceIntegrationTestDataFactory.create_conversation(
-            db_session_with_containers,
-            app_model,
-            user,
-        )
-        runtime_session = AgentRuntimeSession(
-            tenant_id=app_model.tenant_id,
-            app_id=app_model.id,
-            owner_type=AgentRuntimeSessionOwnerType.CONVERSATION,
-            agent_id=str(uuid4()),
-            agent_config_snapshot_id=str(uuid4()),
-            backend_run_id="backend-run-cleanup-failure",
-            session_snapshot=CompositorSessionSnapshot(layers=[]).model_dump_json(),
-            composition_layer_specs='[{"name":"history","type":"pydantic_ai.history","deps":{},"metadata":{},"config":null}]',
-            conversation_id=conversation.id,
-            status=AgentRuntimeSessionStatus.ACTIVE,
-        )
-        db_session_with_containers.add(runtime_session)
-        db_session_with_containers.commit()
-
-        with patch.object(AgentAppRuntimeSessionStore, "mark_cleaned", side_effect=RuntimeError("cleanup failed")):
-            ConversationService.delete(
-                app_model=app_model,
-                conversation_id=conversation.id,
-                user=user,
-                session=db_session_with_containers,
-            )
-
-        deleted = db_session_with_containers.scalar(select(Conversation).where(Conversation.id == conversation.id))
-        assert deleted is None
-        mock_delete_task.delay.assert_called_once_with(conversation.id)
-        mock_cleanup_task.delay.assert_called_once()
-
-    @patch("services.conversation_service.cleanup_conversation_agent_runtime_session")
-    @patch("services.conversation_service.delete_conversation_related_data")
-    def test_delete_ignores_cleanup_enqueue_failure_and_still_retires_runtime_session(
-        self,
-        mock_delete_task,
-        mock_cleanup_task,
-        db_session_with_containers: Session,
-    ):
-        app_model, user = ConversationServiceIntegrationTestDataFactory.create_app_and_account(
-            db_session_with_containers
-        )
-        conversation = ConversationServiceIntegrationTestDataFactory.create_conversation(
-            db_session_with_containers,
-            app_model,
-            user,
-        )
-        conversation_id = conversation.id
-        runtime_session = AgentRuntimeSession(
-            tenant_id=app_model.tenant_id,
-            app_id=app_model.id,
-            owner_type=AgentRuntimeSessionOwnerType.CONVERSATION,
-            agent_id=str(uuid4()),
-            agent_config_snapshot_id=str(uuid4()),
-            backend_run_id="backend-run-enqueue-failure",
-            session_snapshot=CompositorSessionSnapshot(layers=[]).model_dump_json(),
-            composition_layer_specs='[{"name":"history","type":"pydantic_ai.history","deps":{},"metadata":{},"config":null}]',
-            conversation_id=conversation.id,
-            status=AgentRuntimeSessionStatus.ACTIVE,
-        )
-        db_session_with_containers.add(runtime_session)
-        db_session_with_containers.commit()
-        mock_cleanup_task.delay.side_effect = RuntimeError("queue down")
-
-        ConversationService.delete(
-            app_model=app_model,
-            conversation_id=conversation_id,
-            user=user,
-            session=db_session_with_containers,
-        )
-
-        deleted = db_session_with_containers.scalar(select(Conversation).where(Conversation.id == conversation_id))
-        assert deleted is None
-        mock_delete_task.delay.assert_called_once_with(conversation_id)
-        mock_cleanup_task.delay.assert_called_once()
-        runtime_session_row = db_session_with_containers.scalar(
-            select(AgentRuntimeSession).where(AgentRuntimeSession.id == runtime_session.id)
-        )
-        assert runtime_session_row is not None
-        assert runtime_session_row.status == AgentRuntimeSessionStatus.CLEANED

api/tests/test_containers_integration_tests/services/test_workflow_service.py

@@ -870,12 +870,13 @@ class TestWorkflowService:
         from unittest.mock import patch
 
         with patch("flask_login.utils._get_user", return_value=account, autospec=True):
-            result = workflow_service.publish_workflow(
+            result, retirement_candidates = workflow_service.publish_workflow(
                 session=db_session_with_containers, app_model=app, account=account
             )
 
         # Assert
         assert result is not None
+        assert retirement_candidates == set()
         assert result.version != Workflow.VERSION_DRAFT
         # Version should be a timestamp format like '2025-08-22 00:10:24.722051'
         assert isinstance(result.version, str)

api/tests/unit_tests/.ruff.toml

@@ -0,0 +1,430 @@
+extend = "../../.ruff.toml"
+src = ["../.."]
+
+[lint]
+extend-select = ["ANN401", "ARG"]
+
+# Existing strict-mode debt. Remove a file entry when bringing it under strict checking.
+[lint.per-file-ignores]
+"clients/agent_backend/test_request_builder.py" = ["TID251"]
+"commands/test_archive_workflow_runs.py" = ["ARG005"]
+"commands/test_data_migration_wizard.py" = ["ARG005"]
+"commands/test_legacy_model_type_migration.py" = ["ARG001", "ARG002", "ARG005"]
+"controllers/common/test_agent_app_parameters.py" = ["ARG005", "TID251"]
+"controllers/common/test_app_access.py" = ["ARG005"]
+"controllers/console/agent/test_agent_controllers.py" = ["ARG001", "ARG002", "ARG003", "ARG005", "TID251"]
+"controllers/console/app/test_agent_app_sandbox.py" = ["ARG002", "ARG005"]
+"controllers/console/app/test_agent_config_inspector.py" = ["ARG005"]
+"controllers/console/app/test_agent_drive_inspector.py" = ["ARG005"]
+"controllers/console/app/test_agent_manage_guard.py" = ["ARG001"]
+"controllers/console/app/test_agent_skills.py" = ["ARG005"]
+"controllers/console/app/test_annotation_security.py" = ["ARG002"]
+"controllers/console/app/test_app_apis.py" = ["ARG001", "ARG002"]
+"controllers/console/app/test_app_import_api.py" = ["ARG001", "ARG002", "ARG005"]
+"controllers/console/app/test_app_response_models.py" = ["ARG002", "ARG004", "ARG005"]
+"controllers/console/app/test_conversation_api.py" = ["ARG001"]
+"controllers/console/app/test_generator_api.py" = ["ARG001"]
+"controllers/console/app/test_mcp_server_response.py" = ["ARG002"]
+"controllers/console/app/test_message_api.py" = ["ARG001"]
+"controllers/console/app/test_statistic_api.py" = ["ANN401", "ARG001", "TID251"]
+"controllers/console/app/test_workflow.py" = ["ARG001", "ARG005"]
+"controllers/console/app/test_workflow_convert_api.py" = ["ARG005"]
+"controllers/console/app/test_workflow_node_output_inspector.py" = ["ANN401", "ARG001", "TID251"]
+"controllers/console/app/test_workflow_run_api.py" = ["ANN401", "TID251"]
+"controllers/console/app/workflow_draft_variables_test.py" = ["TID251"]
+"controllers/console/auth/test_account_activation.py" = ["ARG002"]
+"controllers/console/auth/test_authentication_security.py" = ["ARG002"]
+"controllers/console/auth/test_email_verification.py" = ["ARG002"]
+"controllers/console/auth/test_login_logout.py" = ["ARG002"]
+"controllers/console/auth/test_oauth.py" = ["ARG002"]
+"controllers/console/auth/test_oauth_timezone.py" = ["ARG001"]
+"controllers/console/auth/test_password_reset.py" = ["ARG002"]
+"controllers/console/auth/test_token_refresh.py" = ["ARG002"]
+"controllers/console/billing/test_billing.py" = ["ARG002"]
+"controllers/console/datasets/test_datasets.py" = ["ARG005"]
+"controllers/console/datasets/test_datasets_document.py" = ["ARG002", "ARG005"]
+"controllers/console/datasets/test_datasets_document_download.py" = ["ARG005"]
+"controllers/console/datasets/test_datasets_segments.py" = ["TID251"]
+"controllers/console/datasets/test_external.py" = ["TID251"]
+"controllers/console/datasets/test_wraps.py" = ["ARG001"]
+"controllers/console/explore/test_trial.py" = ["ANN401", "ARG001", "ARG005", "TID251"]
+"controllers/console/explore/test_wraps.py" = ["ARG001"]
+"controllers/console/snippets/test_snippet_workflow.py" = ["ARG001"]
+"controllers/console/tag/test_tags.py" = ["ARG002"]
+"controllers/console/test_files.py" = ["ARG001", "ARG002"]
+"controllers/console/test_human_input_form.py" = ["ARG001", "ARG005"]
+"controllers/console/test_init_validate.py" = ["ARG005"]
+"controllers/console/test_workspace_account.py" = ["ARG002"]
+"controllers/console/test_workspace_members.py" = ["ARG002", "ARG005"]
+"controllers/console/test_wraps.py" = ["ARG001", "ARG002"]
+"controllers/console/workspace/test_load_balancing_config.py" = ["ARG001"]
+"controllers/console/workspace/test_plugin.py" = ["ARG002", "TID251"]
+"controllers/console/workspace/test_snippets.py" = ["ARG002"]
+"controllers/console/workspace/test_tool_providers.py" = ["ARG001", "ARG005"]
+"controllers/console/workspace/test_trigger_providers.py" = ["ARG001"]
+"controllers/console/workspace/test_workspace.py" = ["ARG005"]
+"controllers/files/test_image_preview.py" = ["ARG005"]
+"controllers/files/test_tool_files.py" = ["ARG002", "ARG005"]
+"controllers/files/test_upload.py" = ["ARG002", "ARG005"]
+"controllers/inner_api/plugin/test_plugin.py" = ["ARG002"]
+"controllers/inner_api/plugin/test_plugin_wraps.py" = ["ARG001", "ARG002", "ARG003", "TID251"]
+"controllers/inner_api/test_runtime_credentials.py" = ["ARG001"]
+"controllers/mcp/test_mcp.py" = ["ARG002"]
+"controllers/openapi/auth/test_conditions.py" = ["ARG005"]
+"controllers/openapi/auth/test_flow.py" = ["ARG005"]
+"controllers/openapi/auth/test_pipeline.py" = ["ARG001"]
+"controllers/openapi/conftest.py" = ["ARG001"]
+"controllers/openapi/test_account.py" = ["ARG005"]
+"controllers/openapi/test_app_describe_builder.py" = ["ARG001"]
+"controllers/openapi/test_app_run_streaming.py" = ["ARG001"]
+"controllers/openapi/test_contract.py" = ["ARG001", "TID251"]
+"controllers/openapi/test_error_contract.py" = ["ARG002"]
+"controllers/openapi/test_human_input_form.py" = ["ARG002"]
+"controllers/openapi/test_oauth_sso_claims.py" = ["ARG002"]
+"controllers/openapi/test_workflow_events_openapi.py" = ["ARG002", "ARG005"]
+"controllers/openapi/test_workspaces_members.py" = ["ARG001"]
+"controllers/service_api/app/test_app.py" = ["ARG002"]
+"controllers/service_api/app/test_completion.py" = ["ARG002"]
+"controllers/service_api/app/test_file.py" = ["ARG002"]
+"controllers/service_api/app/test_hitl_service_api.py" = ["ARG002", "ARG005"]
+"controllers/service_api/app/test_workflow_events.py" = ["ARG005"]
+"controllers/service_api/dataset/rag_pipeline/test_rag_pipeline_workflow.py" = ["ARG002"]
+"controllers/service_api/dataset/test_dataset_segment.py" = ["ARG002"]
+"controllers/service_api/dataset/test_document.py" = ["ARG001", "ARG002"]
+"controllers/service_api/dataset/test_metadata.py" = ["ARG002"]
+"controllers/service_api/test_trace_session_id_parsing.py" = ["ARG001"]
+"controllers/service_api/test_wraps.py" = ["ARG001", "ARG002"]
+"controllers/trigger/test_trigger.py" = ["ARG002"]
+"controllers/trigger/test_webhook.py" = ["ARG002"]
+"controllers/web/conftest.py" = ["ANN401", "TID251"]
+"controllers/web/test_app.py" = ["ARG002", "ARG005"]
+"controllers/web/test_audio.py" = ["ARG002"]
+"controllers/web/test_completion.py" = ["ARG002"]
+"controllers/web/test_feature.py" = ["ARG002"]
+"controllers/web/test_human_input_form.py" = ["ARG001", "ARG002", "ARG005"]
+"controllers/web/test_message_endpoints.py" = ["ARG002"]
+"controllers/web/test_remote_files.py" = ["ARG002"]
+"controllers/web/test_saved_message.py" = ["ARG002"]
+"controllers/web/test_web_login.py" = ["ARG002"]
+"controllers/web/test_web_passport.py" = ["ARG002"]
+"controllers/web/test_workflow.py" = ["ARG002"]
+"core/agent/test_base_agent_runner.py" = ["ARG002"]
+"core/agent/test_cot_agent_runner.py" = ["ARG001"]
+"core/agent/test_cot_chat_agent_runner.py" = ["ARG002"]
+"core/agent/test_fc_agent_runner.py" = ["TID251"]
+"core/app/app_config/common/test_parameters_mapping.py" = ["ARG002"]
+"core/app/app_config/easy_ui_based_app/test_dataset_manager.py" = ["ARG001", "ARG002"]
+"core/app/app_config/easy_ui_based_app/test_model_config_converter.py" = ["ARG002"]
+"core/app/app_config/easy_ui_based_app/test_variables_manager.py" = ["ARG002"]
+"core/app/apps/advanced_chat/test_app_generator.py" = ["ARG001", "ARG002", "ARG005"]
+"core/app/apps/advanced_chat/test_app_runner_input_moderation.py" = ["ARG001", "ARG005"]
+"core/app/apps/advanced_chat/test_generate_task_pipeline.py" = ["ARG005"]
+"core/app/apps/advanced_chat/test_generate_task_pipeline_core.py" = ["ARG001", "ARG002", "ARG005"]
+"core/app/apps/agent_app/test_app_generator.py" = ["ARG001", "ARG005"]
+"core/app/apps/agent_app/test_app_runner.py" = ["ANN401", "ARG002", "ARG005", "TID251"]
+"core/app/apps/agent_app/test_input_guards.py" = ["ANN401", "ARG002", "TID251"]
+"core/app/apps/agent_app/test_resolve_agent.py" = ["ANN401", "TID251"]
+"core/app/apps/agent_app/test_runtime_request_builder.py" = ["ARG002", "TID251"]
+"core/app/apps/agent_chat/test_agent_chat_app_config_manager.py" = ["ARG005"]
+"core/app/apps/agent_chat/test_agent_chat_app_generator.py" = ["ARG001"]
+"core/app/apps/chat/test_app_config_manager.py" = ["ARG001"]
+"core/app/apps/chat/test_app_generator_and_runner.py" = ["ARG001", "ARG002", "ARG005"]
+"core/app/apps/common/test_workflow_response_converter_truncation.py" = ["TID251"]
+"core/app/apps/completion/test_app_runner.py" = ["ARG001", "ARG002", "ARG005"]
+"core/app/apps/pipeline/test_pipeline_generator.py" = ["ARG001", "ARG002", "ARG005"]
+"core/app/apps/pipeline/test_pipeline_runner.py" = ["ARG001", "ARG002"]
+"core/app/apps/test_advanced_chat_app_generator.py" = ["ARG001"]
+"core/app/apps/test_base_app_generator.py" = ["ARG005"]
+"core/app/apps/test_base_app_runner.py" = ["ARG001", "ARG002", "ARG005"]
+"core/app/apps/test_pause_resume.py" = ["ANN401", "TID251"]
+"core/app/apps/test_streaming_utils.py" = ["ARG001"]
+"core/app/apps/test_workflow_app_generator.py" = ["ARG005"]
+"core/app/apps/test_workflow_app_runner_core.py" = ["ARG001", "ARG002", "ARG004", "ARG005"]
+"core/app/apps/test_workflow_app_runner_single_node.py" = ["ANN401", "TID251"]
+"core/app/apps/test_workflow_pause_events.py" = ["ARG005"]
+"core/app/apps/workflow/test_app_generator_extra.py" = ["ARG005"]
+"core/app/apps/workflow/test_generate_task_pipeline_core.py" = ["ARG002", "ARG005"]
+"core/app/features/rate_limiting/test_rate_limit.py" = ["ARG001"]
+"core/app/task_pipeline/test_easy_ui_based_generate_task_pipeline.py" = ["ARG002"]
+"core/app/task_pipeline/test_easy_ui_based_generate_task_pipeline_core.py" = ["ARG001", "ARG002", "ARG005"]
+"core/app/task_pipeline/test_message_cycle_manager_optimization.py" = ["ARG002"]
+"core/app/test_easy_ui_model_config_manager.py" = ["ARG005"]
+"core/app/workflow/layers/test_persistence_inspector_publish.py" = ["ANN401", "ARG005", "TID251"]
+"core/app/workflow/test_file_runtime.py" = ["ARG001", "ARG005"]
+"core/app/workflow/test_observability_layer_extra.py" = ["ARG005"]
+"core/app/workflow/test_persistence_layer.py" = ["ARG001"]
+"core/base/test_app_generator_tts_publisher.py" = ["ARG002"]
+"core/callback_handler/test_agent_tool_callback_handler.py" = ["ARG002"]
+"core/callback_handler/test_workflow_tool_callback_handler.py" = ["ARG002"]
+"core/datasource/__base/test_datasource_provider.py" = ["ARG002"]
+"core/datasource/test_datasource_file_manager.py" = ["ARG001", "ARG002"]
+"core/datasource/test_notion_provider.py" = ["ARG002", "TID251"]
+"core/datasource/test_website_crawl.py" = ["ARG002"]
+"core/datasource/utils/test_message_transformer.py" = ["ARG002"]
+"core/entities/test_entities_mcp_provider.py" = ["ARG001"]
+"core/entities/test_entities_provider_configuration.py" = ["ANN401", "ARG001", "ARG005", "TID251"]
+"core/extension/test_extensible.py" = ["ARG002", "ARG005"]
+"core/external_data_tool/api/test_api.py" = ["ARG001"]
+"core/external_data_tool/test_base.py" = ["TID251"]
+"core/external_data_tool/test_external_data_fetch.py" = ["ARG001"]
+"core/helper/code_executor/test_code_executor.py" = ["TID251"]
+"core/helper/code_executor/test_template_transformer.py" = ["ANN401", "TID251"]
+"core/llm_generator/test_llm_generator.py" = ["ARG002"]
+"core/mcp/auth/test_auth_flow.py" = ["ARG002"]
+"core/mcp/client/test_session.py" = ["ARG001", "TID251"]
+"core/mcp/client/test_sse.py" = ["ARG001", "TID251"]
+"core/mcp/client/test_streamable_http.py" = ["ARG001", "ARG005", "S110", "TID251"]
+"core/mcp/session/test_base_session.py" = ["S110"]
+"core/mcp/session/test_client_session.py" = ["ARG005"]
+"core/mcp/test_mcp_client.py" = ["ARG002"]
+"core/memory/test_token_buffer_memory.py" = ["ARG002"]
+"core/moderation/test_content_moderation.py" = ["TID251"]
+"core/moderation/test_output_moderation.py" = ["ARG001", "ARG002"]
+"core/ops/test_base_trace_instance.py" = ["ARG001"]
+"core/ops/test_lookup_helpers.py" = ["ARG002"]
+"core/ops/test_ops_trace_manager.py" = ["ARG001", "ARG002", "ARG005"]
+"core/ops/test_trace_queue_manager.py" = ["ARG004"]
+"core/ops/test_trace_session_metadata.py" = ["ARG001", "ARG005"]
+"core/plugin/impl/test_agent_client.py" = ["ARG001"]
+"core/plugin/impl/test_datasource_manager.py" = ["ARG001"]
+"core/plugin/impl/test_oauth_handler.py" = ["ARG001"]
+"core/plugin/impl/test_tool_manager.py" = ["ARG001"]
+"core/plugin/impl/test_trigger_client.py" = ["ARG001"]
+"core/plugin/test_endpoint_client.py" = ["ARG002"]
+"core/plugin/test_model_runtime_adapter.py" = ["ARG002"]
+"core/plugin/test_plugin_runtime.py" = ["ARG001", "ARG002", "TID251"]
+"core/prompt/test_advanced_prompt_transform.py" = ["ARG005"]
+"core/prompt/test_prompt_transform.py" = ["ARG005"]
+"core/rag/datasource/keyword/jieba/test_jieba.py" = ["ARG001", "TID251"]
+"core/rag/datasource/keyword/jieba/test_jieba_keyword_table_handler.py" = ["ARG004"]
+"core/rag/datasource/keyword/test_keyword_factory.py" = ["ARG005"]
+"core/rag/datasource/test_datasource_retrieval.py" = ["ARG001", "ARG002", "ARG005", "TID251"]
+"core/rag/datasource/test_retrieval_attachment_access.py" = ["ARG005"]
+"core/rag/datasource/vdb/test_vector_factory.py" = ["ARG002"]
+"core/rag/embedding/test_embedding_base.py" = ["TID251"]
+"core/rag/embedding/test_embedding_service.py" = ["ARG001"]
+"core/rag/extractor/firecrawl/test_firecrawl.py" = ["TID251"]
+"core/rag/extractor/test_csv_extractor.py" = ["ARG001", "ARG005"]
+"core/rag/extractor/test_excel_extractor.py" = ["ARG002", "ARG005"]
+"core/rag/extractor/test_extract_processor.py" = ["ARG001", "ARG005"]
+"core/rag/extractor/test_helpers.py" = ["ARG002"]
+"core/rag/extractor/test_markdown_extractor.py" = ["ARG001"]
+"core/rag/extractor/test_notion_extractor.py" = ["ARG002", "ARG005"]
+"core/rag/extractor/test_pdf_extractor.py" = ["ARG001", "ARG005"]
+"core/rag/extractor/test_text_extractor.py" = ["ARG001"]
+"core/rag/extractor/test_word_extractor.py" = ["ARG001", "ARG002", "ARG005"]
+"core/rag/extractor/unstructured/test_unstructured_extractors.py" = ["ARG001", "ARG005"]
+"core/rag/extractor/watercrawl/test_watercrawl.py" = ["ARG001", "ARG005", "TID251"]
+"core/rag/indexing/processor/conftest.py" = ["ANN401", "ARG002", "TID251"]
+"core/rag/indexing/processor/test_paragraph_index_processor.py" = ["ARG002", "TID251"]
+"core/rag/indexing/processor/test_qa_index_processor.py" = ["ARG001", "TID251"]
+"core/rag/indexing/test_index_processor.py" = ["ARG005"]
+"core/rag/indexing/test_indexing_runner.py" = ["ARG005", "TID251"]
+"core/rag/pipeline/test_queue.py" = ["ARG002"]
+"core/rag/retrieval/test_dataset_retrieval.py" = ["ARG001", "ARG002", "ARG005", "TID251"]
+"core/rag/splitter/test_text_splitter.py" = ["ARG002", "ARG005"]
+"core/repositories/test_celery_workflow_execution_repository.py" = ["ARG002"]
+"core/repositories/test_celery_workflow_node_execution_repository.py" = ["ARG002"]
+"core/repositories/test_human_input_form_repository_impl.py" = ["ARG001", "ARG005"]
+"core/repositories/test_human_input_repository.py" = ["ANN401", "ARG001", "ARG005", "TID251"]
+"core/repositories/test_sqlalchemy_workflow_node_execution_repository.py" = ["ANN401", "ARG002", "ARG005", "TID251"]
+"core/repositories/test_workflow_node_execution_truncation.py" = ["TID251"]
+"core/schemas/test_resolver.py" = ["ARG005", "T201"]
+"core/telemetry/test_facade.py" = ["ARG002", "ARG004"]
+"core/telemetry/test_gateway_integration.py" = ["ARG002"]
+"core/test_model_manager.py" = ["ARG001"]
+"core/test_trigger_debug_event_selectors.py" = ["ARG002"]
+"core/tools/test_base_tool.py" = ["ANN401", "ARG002", "TID251"]
+"core/tools/test_builtin_tool_base.py" = ["ARG001", "ARG002", "TID251"]
+"core/tools/test_builtin_tool_provider.py" = ["ARG001", "ARG005", "TID251"]
+"core/tools/test_builtin_tools_extra.py" = ["ARG005"]
+"core/tools/test_custom_tool.py" = ["ARG001", "ARG005", "TID251"]
+"core/tools/test_dataset_retriever_tool.py" = ["ARG005"]
+"core/tools/test_mcp_tool.py" = ["S110"]
+"core/tools/test_tool_engine.py" = ["ANN401", "ARG002", "ARG005", "TID251"]
+"core/tools/test_tool_file_manager.py" = ["ARG001"]
+"core/tools/test_tool_label_manager.py" = ["TID251"]
+"core/tools/test_tool_manager.py" = ["ANN401", "ARG001", "ARG005", "TID251"]
+"core/tools/test_tool_provider_controller.py" = ["TID251"]
+"core/tools/utils/test_configuration.py" = ["ARG001", "ARG002", "TID251"]
+"core/tools/utils/test_encryption.py" = ["ANN401", "TID251"]
+"core/tools/utils/test_message_transformer.py" = ["TID251"]
+"core/tools/utils/test_misc_utils_extra.py" = ["ARG002"]
+"core/tools/utils/test_model_invocation_utils.py" = ["ARG005", "TID251"]
+"core/tools/utils/test_parser.py" = ["TID251"]
+"core/tools/utils/test_web_reader_tool.py" = ["ARG001", "ARG002", "ARG005"]
+"core/tools/workflow_as_tool/test_provider.py" = ["TID251"]
+"core/tools/workflow_as_tool/test_tool.py" = ["ANN401", "ARG001", "ARG005", "TID251"]
+"core/trigger/conftest.py" = ["ANN401", "TID251"]
+"core/trigger/debug/test_debug_event_selectors.py" = ["ARG002", "TID251"]
+"core/variables/test_segment_type_validation.py" = ["TID251"]
+"core/workflow/context/test_execution_context.py" = ["ANN401", "ARG002", "S110", "TID251"]
+"core/workflow/context/test_flask_app_context.py" = ["ARG002"]
+"core/workflow/generator/test_runner.py" = ["ARG001", "ARG002", "TID251"]
+"core/workflow/generator/test_runner_missing.py" = ["ARG003"]
+"core/workflow/generator/test_tool_catalogue.py" = ["ARG002"]
+"core/workflow/graph_engine/layers/test_observability.py" = ["ARG002"]
+"core/workflow/graph_engine/test_mock_config.py" = ["TID251"]
+"core/workflow/graph_engine/test_mock_factory.py" = ["TID251"]
+"core/workflow/graph_engine/test_mock_nodes.py" = ["ANN401", "S110", "TID251"]
+"core/workflow/graph_engine/test_parallel_human_input_join_resume.py" = ["ARG002", "TID251"]
+"core/workflow/graph_engine/test_table_runner.py" = ["ARG001", "TID251"]
+"core/workflow/nodes/agent_v2/test_agent_node.py" = ["ARG001", "ARG002", "ARG005"]
+"core/workflow/nodes/agent_v2/test_ask_human_hitl.py" = ["ANN401", "TID251"]
+"core/workflow/nodes/agent_v2/test_dify_tools_builder.py" = ["ANN401", "ARG001", "ARG002", "ARG005", "TID251"]
+"core/workflow/nodes/agent_v2/test_output_adapter.py" = ["ARG005"]
+"core/workflow/nodes/agent_v2/test_runtime_request_builder.py" = ["ARG002"]
+"core/workflow/nodes/agent_v2/test_validators.py" = ["ARG001"]
+"core/workflow/nodes/http_request/test_http_request_node.py" = ["ANN401", "ARG002", "TID251"]
+"core/workflow/nodes/human_input/test_entities.py" = ["TID251"]
+"core/workflow/nodes/human_input/test_human_input_form_filled_event.py" = ["TID251"]
+"core/workflow/nodes/iteration/test_iteration_child_engine_errors.py" = ["ARG002", "TID251"]
+"core/workflow/nodes/knowledge_index/test_knowledge_index_node.py" = ["ARG002"]
+"core/workflow/nodes/knowledge_retrieval/test_knowledge_retrieval_node.py" = ["ARG002"]
+"core/workflow/nodes/llm/test_node.py" = ["ARG002"]
+"core/workflow/nodes/parameter_extractor/test_parameter_extractor_node.py" = ["TID251"]
+"core/workflow/nodes/test_document_extractor_node.py" = ["ARG001"]
+"core/workflow/nodes/tool/test_tool_node.py" = ["ANN401", "ARG002", "TID251"]
+"core/workflow/nodes/webhook/test_webhook_file_conversion.py" = ["TID251"]
+"core/workflow/nodes/webhook/test_webhook_node.py" = ["TID251"]
+"core/workflow/test_form_input_serialization_compat.py" = ["ANN401", "TID251"]
+"core/workflow/test_human_input_adapter.py" = ["ARG005"]
+"core/workflow/test_node_factory.py" = ["ARG002"]
+"core/workflow/test_workflow_entry.py" = ["ARG001"]
+"enterprise/telemetry/test_enterprise_trace.py" = ["ARG002", "TID251"]
+"enterprise/telemetry/test_exporter.py" = ["ARG001"]
+"enterprise/telemetry/test_gateway.py" = ["ARG002"]
+"events/event_handlers/test_delete_tool_parameters_cache_when_sync_draft_workflow.py" = ["ARG005"]
+"extensions/logstore/test_sql_escape.py" = ["ARG001", "ARG002"]
+"extensions/otel/decorators/handlers/test_generate_handler.py" = ["ARG001", "ARG002"]
+"extensions/otel/decorators/handlers/test_workflow_app_runner_handler.py" = ["ARG001"]
+"extensions/otel/decorators/test_base.py" = ["ARG002"]
+"extensions/otel/decorators/test_handler.py" = ["ARG002"]
+"extensions/otel/test_retrieval_tracing.py" = ["ARG001"]
+"extensions/test_ext_request_logging.py" = ["ARG002"]
+"extensions/test_redis.py" = ["ARG001"]
+"factories/test_build_from_mapping.py" = ["ARG001"]
+"factories/test_file_factory.py" = ["ARG001"]
+"factories/test_variable_factory.py" = ["TID251"]
+"fields/test_file_fields.py" = ["ARG005"]
+"libs/_human_input/support.py" = ["TID251"]
+"libs/broadcast_channel/redis/test_channel_unit_tests.py" = ["ARG002", "ARG005"]
+"libs/broadcast_channel/redis/test_streams_channel_unit_tests.py" = ["ARG001", "ARG002", "TID251"]
+"libs/test_cron_compatibility.py" = ["S110"]
+"libs/test_email_i18n.py" = ["ANN401", "TID251"]
+"libs/test_oauth_bearer_rate_limit_ordering.py" = ["ARG001"]
+"libs/test_pyrefly_type_coverage.py" = ["TID251"]
+"libs/test_schedule_utils_enhanced.py" = ["S110"]
+"libs/test_sendgrid_client.py" = ["ARG001", "TID251"]
+"libs/test_smtp_client.py" = ["TID251"]
+"models/test_dataset_models.py" = ["ARG005"]
+"models/test_plugin_entities.py" = ["TID251"]
+"models/test_snippet.py" = ["ARG001"]
+"oss/__mock/aliyun_oss.py" = ["ARG002"]
+"oss/__mock/baidu_obs.py" = ["ARG002"]
+"oss/__mock/base.py" = ["ARG002"]
+"oss/__mock/tencent_cos.py" = ["ARG002"]
+"oss/__mock/volcengine_tos.py" = ["ARG002"]
+"oss/aliyun_oss/aliyun_oss/test_aliyun_oss.py" = ["ARG002"]
+"oss/baidu_obs/test_baidu_obs.py" = ["ARG002"]
+"oss/opendal/test_opendal.py" = ["ARG002"]
+"oss/tencent_cos/test_tencent_cos.py" = ["ARG002"]
+"oss/volcengine_tos/test_volcengine_tos.py" = ["ARG002"]
+"services/agent/test_agent_observability_service.py" = ["ARG002", "ARG005"]
+"services/agent/test_agent_services.py" = ["ARG001", "ARG002", "ARG003", "ARG005"]
+"services/agent/test_composer_candidates.py" = ["ARG005"]
+"services/agent/test_prompt_mentions.py" = ["ARG005"]
+"services/agent/test_skill_tool_inference_service.py" = ["ARG001", "ARG005"]
+"services/auth/test_jina_auth_standalone_module.py" = ["TID251"]
+"services/controller_api.py" = ["ARG002"]
+"services/data_migration/test_import_service.py" = ["ARG002", "ARG005"]
+"services/dataset_service_test_helpers.py" = ["TID251"]
+"services/enterprise/test_account_deletion_sync.py" = ["ARG001"]
+"services/enterprise/test_rbac_service.py" = ["ARG002"]
+"services/enterprise/test_traceparent_propagation.py" = ["ARG002"]
+"services/hit_service.py" = ["TID251"]
+"services/plugin/test_plugin_parameter_service.py" = ["ARG002"]
+"services/rag_pipeline/pipeline_template/test_built_in_retrieval.py" = ["ARG001"]
+"services/rag_pipeline/test_rag_pipeline_dsl_service.py" = ["ARG001", "ARG005", "T201", "TID251"]
+"services/rag_pipeline/test_rag_pipeline_service.py" = ["ARG001", "ARG005"]
+"services/rag_pipeline/test_rag_pipeline_task_proxy.py" = ["ARG001", "ARG005"]
+"services/rag_pipeline/test_rag_pipeline_transform_service.py" = ["ARG001"]
+"services/recommend_app/test_remote_retrieval.py" = ["ARG002"]
+"services/retention/workflow_run/test_archive_download_preparation.py" = ["ARG002"]
+"services/retention/workflow_run/test_archive_log_service.py" = ["ARG001", "ARG002"]
+"services/retention/workflow_run/test_bundle_archive_maintenance.py" = ["TID251"]
+"services/retention/workflow_run/test_restore_archived_workflow_run.py" = ["ARG002"]
+"services/test_account_service.py" = ["ARG001", "ARG002"]
+"services/test_annotation_service.py" = ["ANN401", "TID251"]
+"services/test_api_token_service.py" = ["ARG002"]
+"services/test_app_generate_service.py" = ["ARG001", "ARG002", "ARG004"]
+"services/test_app_generate_service_streaming_integration.py" = ["ARG002", "TID251"]
+"services/test_archive_workflow_run_logs.py" = ["ARG002"]
+"services/test_audio_service.py" = ["ARG002", "TID251"]
+"services/test_batch_indexing_base.py" = ["ANN401", "TID251"]
+"services/test_billing_service.py" = ["ARG001", "ARG002"]
+"services/test_clear_free_plan_expired_workflow_run_logs.py" = ["ANN401", "ARG002", "ARG005", "TID251"]
+"services/test_clear_free_plan_tenant_expired_logs.py" = ["ARG002", "ARG003"]
+"services/test_dataset_service_document.py" = ["ARG002"]
+"services/test_dataset_service_lock_not_owned.py" = ["ARG001", "ARG005"]
+"services/test_dataset_service_segment.py" = ["ARG002"]
+"services/test_datasource_provider_service.py" = ["ARG002"]
+"services/test_external_dataset_service.py" = ["ARG002", "TID251"]
+"services/test_feature_service_human_input_email_delivery.py" = ["ARG005"]
+"services/test_feedback_service.py" = ["ARG002"]
+"services/test_human_input_delivery_test_service.py" = ["ARG005"]
+"services/test_knowledge_service.py" = ["TID251"]
+"services/test_message_service.py" = ["ARG002"]
+"services/test_messages_clean_service.py" = ["TID251"]
+"services/test_model_load_balancing_service.py" = ["ANN401", "ARG001", "ARG005", "TID251"]
+"services/test_model_provider_service.py" = ["ANN401", "TID251"]
+"services/test_model_provider_service_sanitization.py" = ["ARG002", "ARG005"]
+"services/test_oauth_server_service.py" = ["ARG002"]
+"services/test_operation_service.py" = ["TID251"]
+"services/test_rag_pipeline_task_proxy.py" = ["ARG002"]
+"services/test_recommended_app_service.py" = ["ARG001"]
+"services/test_schedule_service.py" = ["ANN401", "TID251"]
+"services/test_snippet_service.py" = ["ARG001", "ARG002"]
+"services/test_summary_index_service.py" = ["ARG001"]
+"services/test_telemetry_service.py" = ["ARG001", "ARG005"]
+"services/test_variable_truncator.py" = ["ARG002", "TID251"]
+"services/test_variable_truncator_additional.py" = ["ANN401", "TID251"]
+"services/test_vector_service.py" = ["ARG001", "TID251"]
+"services/test_webhook_service_additional.py" = ["ANN401", "ARG002", "TID251"]
+"services/test_website_service.py" = ["TID251"]
+"services/test_workflow_comment_service.py" = ["ARG001", "ARG002"]
+"services/test_workflow_run_service.py" = ["ANN401", "ARG002", "TID251"]
+"services/test_workflow_service.py" = ["ANN401", "ARG002", "TID251"]
+"services/tools/test_builtin_tools_manage_service.py" = ["ARG001", "ARG002"]
+"services/tools/test_tools_manage_service.py" = ["ARG002"]
+"services/workflow/test_inspector_events.py" = ["ANN401", "TID251"]
+"services/workflow/test_node_output_inspector_service.py" = ["TID251"]
+"services/workflow/test_workflow_converter_additional.py" = ["ANN401", "ARG001", "ARG005", "TID251"]
+"services/workflow/test_workflow_event_snapshot_service.py" = ["ANN401", "ARG002", "ARG005", "TID251"]
+"services/workflow/test_workflow_event_snapshot_service_additional.py" = ["ANN401", "ARG002", "ARG005", "TID251"]
+"tasks/test_agent_backend_session_cleanup_task.py" = ["ARG005"]
+"tasks/test_clean_dataset_task.py" = ["ARG002"]
+"tasks/test_clean_document_task.py" = ["ARG002"]
+"tasks/test_dataset_indexing_task.py" = ["ARG001", "ARG002"]
+"tasks/test_document_indexing_sync_task.py" = ["ARG002"]
+"tasks/test_duplicate_document_indexing_task.py" = ["ARG002"]
+"tasks/test_human_input_timeout_tasks.py" = ["ARG001", "ARG002", "ARG005", "TID251"]
+"tasks/test_initialize_created_app_rbac_access_task.py" = ["ARG005"]
+"tasks/test_mail_send_task.py" = ["ARG002"]
+"tasks/test_ops_trace_task.py" = ["ARG004"]
+"tasks/test_process_tenant_plugin_autoupgrade_check_task.py" = ["ARG001"]
+"tasks/test_remove_app_and_related_data_task.py" = ["ARG002"]
+"tasks/test_trigger_processing_tasks.py" = ["ARG002"]
+"tasks/test_workflow_execute_task.py" = ["ARG005"]
+"test_app_factory.py" = ["ARG001"]
+"test_pytest_dify.py" = ["ARG001"]
+"tools/test_mcp_tool.py" = ["TID251"]
+
+[lint.flake8-tidy-imports.banned-api."flask_restx.reqparse"]
+msg = "Use Pydantic payload/query models instead of reqparse."
+
+[lint.flake8-tidy-imports.banned-api."flask_restx.reqparse.RequestParser"]
+msg = "Use Pydantic payload/query models instead of reqparse."
+
+[lint.flake8-tidy-imports.banned-api."typing.Any"]
+msg = "Use object, Protocol, TypedDict, TypeVar, ParamSpec, or a localized cast instead."

api/tests/unit_tests/clients/agent_backend/test_cleanup_composition_compositor_integration.py

@@ -1,134 +0,0 @@
-"""Integration test for the cleanup request against the real agenton compositor.
-
-The bug fixed by A+D was invisible to unit tests that use ``FakeAgentBackendRunClient``
-because the fake client never runs agenton's ``_validate_session_snapshot``. This
-test plugs a cleanup request through the real ``Compositor`` (with the same
-providers the agent backend wires in production) so that the snapshot-vs-
-composition name-order check would fail loudly if the cleanup builder ever
-regressed back to the empty-composition shape.
-"""
-
-from __future__ import annotations
-
-from typing import cast
-
-import pytest
-from agenton.compositor import Compositor, CompositorSessionSnapshot, LayerProvider
-from agenton.compositor.schemas import LayerSessionSnapshot
-from agenton.layers.base import LifecycleState
-from agenton_collections.layers.plain import PLAIN_PROMPT_LAYER_TYPE_ID
-from agenton_collections.layers.plain.basic import PromptLayer
-from agenton_collections.layers.pydantic_ai import PYDANTIC_AI_HISTORY_LAYER_TYPE_ID, PydanticAIHistoryLayer
-
-from clients.agent_backend import AgentBackendRunRequestBuilder, RuntimeLayerSpec
-
-
-def test_cleanup_request_passes_agenton_snapshot_validation():
-    """The cleanup request's composition layer names must match the (filtered)
-    snapshot's layer names exactly — agenton's compositor enforces this and
-    the agent backend rejects mismatches as ``run_failed`` asynchronously,
-    which is the trap A/D fixed."""
-    # Persisted (non-plugin) layer specs — these are what cleanup will replay.
-    # We exclude the dify.execution_context layer from this integration check
-    # because its real provider needs a plugin-daemon HTTP client; the cleanup
-    # validation we are exercising is the snapshot-vs-composition name check,
-    # which is purely structural and does not depend on which non-plugin layer
-    # types appear.
-    persisted_specs = [
-        RuntimeLayerSpec(
-            name="workflow_node_job_prompt",
-            type=PLAIN_PROMPT_LAYER_TYPE_ID,
-            config={"prefix": "Do the cleanup."},
-        ),
-        RuntimeLayerSpec(name="history", type=PYDANTIC_AI_HISTORY_LAYER_TYPE_ID),
-    ]
-    # Saved snapshot still carries the LLM layer entry — cleanup's
-    # ``_filter_snapshot_to_specs`` must drop it so names match.
-    full_snapshot = CompositorSessionSnapshot(
-        layers=[
-            LayerSessionSnapshot(
-                name="workflow_node_job_prompt",
-                lifecycle_state=LifecycleState.SUSPENDED,
-                runtime_state={},
-            ),
-            LayerSessionSnapshot(
-                name="history",
-                lifecycle_state=LifecycleState.SUSPENDED,
-                runtime_state={"messages": []},
-            ),
-            LayerSessionSnapshot(
-                name="llm",
-                lifecycle_state=LifecycleState.SUSPENDED,
-                runtime_state={},
-            ),
-        ]
-    )
-
-    cleanup_request = AgentBackendRunRequestBuilder().build_cleanup_request(
-        session_snapshot=full_snapshot,
-        runtime_layer_specs=persisted_specs,
-    )
-
-    # Drive the real agenton compositor through ``from_config`` + ``_create_run``
-    # the same way the agent backend's RunScheduler does. ``_create_run`` is the
-    # private path that calls ``_validate_session_snapshot``; we use it directly
-    # to keep the test synchronous (no async ``enter()`` lifecycle needed —
-    # validation is the only thing under test).
-    config = {
-        "schema_version": 1,
-        "layers": [
-            {"name": layer.name, "type": layer.type, "deps": dict(layer.deps), "metadata": dict(layer.metadata)}
-            for layer in cleanup_request.composition.layers
-        ],
-    }
-    compositor = Compositor.from_config(
-        config,
-        providers=[
-            LayerProvider.from_layer_type(PromptLayer),
-            LayerProvider.from_layer_type(PydanticAIHistoryLayer),
-        ],
-    )
-
-    layer_configs = {layer.name: layer.config for layer in cleanup_request.composition.layers}
-    # This is the call that would raise ``ValueError`` if the cleanup snapshot
-    # and composition disagreed on layer names — the exact failure mode the
-    # original ``layers=[]`` cleanup hit.
-    run = compositor._create_run(  # type: ignore[reportPrivateUsage]
-        configs=cast(dict[str, object], layer_configs),
-        session_snapshot=cleanup_request.session_snapshot,
-    )
-    assert list(run.slots.keys()) == ["workflow_node_job_prompt", "history"]
-
-
-def test_cleanup_request_with_mismatched_specs_would_be_rejected_by_agenton():
-    """Regression sentinel: if a future refactor stops filtering the snapshot,
-    agenton would reject the request — and that rejection is what the runtime
-    fix is preventing. We confirm the validator does fail when given the
-    pre-fix shape so the previous test's success is not a coincidence."""
-    snapshot_with_extra = CompositorSessionSnapshot(
-        layers=[
-            LayerSessionSnapshot(
-                name="history",
-                lifecycle_state=LifecycleState.SUSPENDED,
-                runtime_state={},
-            ),
-            LayerSessionSnapshot(
-                name="llm",  # extra layer not in composition
-                lifecycle_state=LifecycleState.SUSPENDED,
-                runtime_state={},
-            ),
-        ]
-    )
-    compositor = Compositor.from_config(
-        {
-            "schema_version": 1,
-            "layers": [{"name": "history", "type": PYDANTIC_AI_HISTORY_LAYER_TYPE_ID, "deps": {}, "metadata": {}}],
-        },
-        providers=[LayerProvider.from_layer_type(PydanticAIHistoryLayer)],
-    )
-
-    with pytest.raises(ValueError, match="layer names must match"):
-        compositor._create_run(  # type: ignore[reportPrivateUsage]
-            configs={},
-            session_snapshot=snapshot_with_extra,
-        )

api/tests/unit_tests/clients/agent_backend/test_client.py

@@ -40,6 +40,7 @@ def _request():
                 agent_mode="workflow_run",
                 invoke_from="debugger",
             ),
+            backend_binding_ref="binding-ref-1",
             workflow_node_job_prompt="Do the task.",
             user_prompt="hello",
         )

api/tests/unit_tests/clients/agent_backend/test_fake_client.py

@@ -23,6 +23,7 @@ def _request():
                 agent_mode="workflow_run",
                 invoke_from="debugger",
             ),
+            backend_binding_ref="binding-ref-1",
             workflow_node_job_prompt="Do the task.",
             user_prompt="hello",
         )

api/tests/unit_tests/clients/agent_backend/test_request_builder.py

@@ -1,10 +1,7 @@
 from typing import Any, cast
 
 import pytest
-from agenton.compositor import CompositorSessionSnapshot
-from agenton.compositor.schemas import LayerSessionSnapshot
 from agenton.layers import ExitIntent
-from agenton.layers.base import LifecycleState
 from agenton_collections.layers.plain import PLAIN_PROMPT_LAYER_TYPE_ID, PromptLayerConfig
 from agenton_collections.layers.pydantic_ai import PYDANTIC_AI_HISTORY_LAYER_TYPE_ID
 from dify_agent.layers.dify_core_tools import (
@@ -45,8 +42,6 @@ from clients.agent_backend import (
     AgentBackendOutputConfig,
     AgentBackendRunRequestBuilder,
     AgentBackendWorkflowNodeRunInput,
-    RuntimeLayerSpec,
-    extract_runtime_layer_specs,
     redact_for_agent_backend_log,
 )
 from clients.agent_backend.request_builder import DIFY_DRIVE_LAYER_ID, DIFY_SHELL_LAYER_ID
@@ -71,6 +66,7 @@ def _run_input() -> AgentBackendWorkflowNodeRunInput:
             agent_mode="workflow_run",
             invoke_from="debugger",
         ),
+        backend_binding_ref="binding-ref-1",
         idempotency_key="workflow-run-1:node-execution-1",
         agent_soul_prompt="You are a careful reviewer.",
         workflow_node_job_prompt="Review the previous node output.",
@@ -271,89 +267,6 @@ def test_request_builder_adds_knowledge_layer_when_configured():
     assert knowledge_config.sets[0].dataset_ids == ["dataset-1"]
 
 
-def test_request_builder_can_delete_on_exit_for_cleanup_paths():
-    run_input = _run_input()
-    run_input.suspend_on_exit = False
-
-    request = AgentBackendRunRequestBuilder().build_for_workflow_node(run_input)
-
-    assert request.on_exit.default is ExitIntent.DELETE
-
-
-def test_request_builder_builds_cleanup_request_replays_persisted_layer_specs():
-    """The cleanup request must replay the persisted (non-plugin) layer specs
-    and filter the snapshot to match so the agenton compositor's
-    snapshot-vs-composition name-order validator passes."""
-    session_snapshot = CompositorSessionSnapshot(
-        layers=[
-            LayerSessionSnapshot(name="history", lifecycle_state=LifecycleState.SUSPENDED, runtime_state={"k": 1}),
-            LayerSessionSnapshot(name="llm", lifecycle_state=LifecycleState.SUSPENDED, runtime_state={}),
-        ]
-    )
-    specs = [RuntimeLayerSpec(name="history", type="pydantic_ai.history")]
-
-    request = AgentBackendRunRequestBuilder().build_cleanup_request(
-        session_snapshot=session_snapshot,
-        runtime_layer_specs=specs,
-        idempotency_key="run-1:node-1:binding-1:agent-session-cleanup",
-        metadata={"workflow_run_id": "run-1"},
-    )
-
-    assert [layer.name for layer in request.composition.layers] == ["history"]
-    assert request.session_snapshot is not None
-    assert [layer.name for layer in request.session_snapshot.layers] == ["history"]
-    assert request.on_exit.default is ExitIntent.DELETE
-    assert request.idempotency_key == "run-1:node-1:binding-1:agent-session-cleanup"
-    assert request.metadata["agent_backend_lifecycle"] == "session_cleanup"
-    assert "purpose" not in request.model_dump(mode="json")
-
-
-def test_request_builder_rejects_empty_runtime_layer_specs():
-    """Empty specs would put us back in the original ``layers=[]`` trap that
-    fails on agenton's snapshot-vs-composition validation."""
-    with pytest.raises(ValueError, match="runtime_layer_specs"):
-        AgentBackendRunRequestBuilder().build_cleanup_request(
-            session_snapshot=CompositorSessionSnapshot(layers=[]),
-            runtime_layer_specs=[],
-        )
-
-
-def test_extract_runtime_layer_specs_drops_plugin_layers_keeps_configs():
-    from dify_agent.protocol import RunComposition, RunLayerSpec
-
-    composition = RunComposition(
-        layers=[
-            RunLayerSpec(
-                name="agent_soul_prompt",
-                type="plain.prompt",
-                config=PromptLayerConfig(prefix="hello"),
-            ),
-            RunLayerSpec(
-                name="llm",
-                type="dify.plugin.llm",
-                config=None,  # protocol allows None; the redacted config is what matters
-            ),
-            RunLayerSpec(
-                name="tools",
-                type="dify.plugin.tools",
-            ),
-            RunLayerSpec(
-                name="history",
-                type="pydantic_ai.history",
-            ),
-        ]
-    )
-
-    specs = extract_runtime_layer_specs(composition)
-
-    assert [spec.name for spec in specs] == ["agent_soul_prompt", "history"]
-    # Non-plugin configs are dumped as JSON-compatible dicts so the persisted
-    # row can be replayed without holding live pydantic instances.
-    soul_config = specs[0].config
-    assert isinstance(soul_config, dict)
-    assert soul_config.get("prefix") == "hello"
-
-
 def test_request_builder_rejects_blank_prompts():
     with pytest.raises(ValidationError):
         AgentBackendWorkflowNodeRunInput(
@@ -397,6 +310,7 @@ def _agent_app_input(*, include_shell: bool = False) -> AgentBackendAgentAppRunI
             agent_mode="agent_app",
             invoke_from="web-app",
         ),
+        backend_binding_ref="binding-ref-1",
         agent_soul_prompt="You are Iris.",
         user_prompt="List files.",
         include_shell=include_shell,
@@ -422,7 +336,7 @@ def test_workflow_request_builder_adds_shell_layer_when_include_shell():
     assert shell.type == DIFY_SHELL_LAYER_TYPE_ID
     # The shell layer depends on execution_context so the agent server can mint
     # per-command Agent Stub env for sandbox CLI forwarding.
-    assert shell.deps == {"execution_context": DIFY_EXECUTION_CONTEXT_LAYER_ID}
+    assert shell.deps == {"execution_context": DIFY_EXECUTION_CONTEXT_LAYER_ID, "runtime": "runtime"}
     shell_config = cast(DifyShellLayerConfig, shell.config)
     assert shell_config.env[0].name == "PROJECT_NAME"
 
@@ -436,7 +350,10 @@ def test_workflow_request_builder_binds_drive_to_shell_when_configured():
     layers = {layer.name: layer for layer in request.composition.layers}
     layer_names = [layer.name for layer in request.composition.layers]
 
-    assert layers[DIFY_SHELL_LAYER_ID].deps == {"execution_context": DIFY_EXECUTION_CONTEXT_LAYER_ID}
+    assert layers[DIFY_SHELL_LAYER_ID].deps == {
+        "execution_context": DIFY_EXECUTION_CONTEXT_LAYER_ID,
+        "runtime": "runtime",
+    }
     shell_config = cast(DifyShellLayerConfig, layers[DIFY_SHELL_LAYER_ID].config)
     assert shell_config.agent_stub_drive_ref == "agent-agent-1"
     assert layers[DIFY_DRIVE_LAYER_ID].deps == {"shell": DIFY_SHELL_LAYER_ID}
@@ -470,7 +387,10 @@ def test_agent_app_request_builder_adds_shell_layer_when_include_shell():
 
     assert DIFY_SHELL_LAYER_ID in layers
     assert layers[DIFY_SHELL_LAYER_ID].type == DIFY_SHELL_LAYER_TYPE_ID
-    assert layers[DIFY_SHELL_LAYER_ID].deps == {"execution_context": DIFY_EXECUTION_CONTEXT_LAYER_ID}
+    assert layers[DIFY_SHELL_LAYER_ID].deps == {
+        "execution_context": DIFY_EXECUTION_CONTEXT_LAYER_ID,
+        "runtime": "runtime",
+    }
     shell_config = cast(DifyShellLayerConfig, layers[DIFY_SHELL_LAYER_ID].config)
     assert shell_config.env[0].name == "APP_ENV"
 
@@ -483,7 +403,10 @@ def test_agent_app_request_builder_binds_drive_to_shell_when_configured():
     layers = {layer.name: layer for layer in request.composition.layers}
     layer_names = [layer.name for layer in request.composition.layers]
 
-    assert layers[DIFY_SHELL_LAYER_ID].deps == {"execution_context": DIFY_EXECUTION_CONTEXT_LAYER_ID}
+    assert layers[DIFY_SHELL_LAYER_ID].deps == {
+        "execution_context": DIFY_EXECUTION_CONTEXT_LAYER_ID,
+        "runtime": "runtime",
+    }
     shell_config = cast(DifyShellLayerConfig, layers[DIFY_SHELL_LAYER_ID].config)
     assert shell_config.agent_stub_drive_ref == "agent-agent-1"
     assert layers[DIFY_DRIVE_LAYER_ID].deps == {"shell": DIFY_SHELL_LAYER_ID}

api/tests/unit_tests/clients/agent_backend/test_session_cleanup.py

@@ -1,124 +0,0 @@
-from datetime import UTC, datetime
-
-from agenton.compositor import CompositorSessionSnapshot
-from agenton.compositor.schemas import LayerSessionSnapshot
-from agenton.layers.base import LifecycleState
-from dify_agent.protocol import RunStatusResponse
-
-from clients.agent_backend import (
-    AgentBackendError,
-    AgentBackendSessionCleanupPayload,
-    FakeAgentBackendRunClient,
-    RuntimeLayerSpec,
-    cleanup_agent_backend_session,
-)
-
-
-def _payload() -> AgentBackendSessionCleanupPayload:
-    return AgentBackendSessionCleanupPayload(
-        session_snapshot=CompositorSessionSnapshot(
-            layers=[
-                LayerSessionSnapshot(
-                    name="history",
-                    lifecycle_state=LifecycleState.SUSPENDED,
-                    runtime_state={},
-                )
-            ]
-        ),
-        runtime_layer_specs=[RuntimeLayerSpec(name="history", type="pydantic_ai.history")],
-        idempotency_key="cleanup-1",
-        metadata={"tenant_id": "tenant-1"},
-        timeout_seconds=15.0,
-    )
-
-
-def test_cleanup_agent_backend_session_runs_create_and_wait_until_success():
-    client = FakeAgentBackendRunClient(run_id="cleanup-run-1")
-
-    result = cleanup_agent_backend_session(payload=_payload(), client=client)
-
-    assert result.status == "succeeded"
-    assert result.cleanup_run_id == "cleanup-run-1"
-    assert client.request is not None
-    assert [layer.name for layer in client.request.composition.layers] == ["history"]
-
-
-def test_cleanup_agent_backend_session_skips_when_client_is_missing():
-    result = cleanup_agent_backend_session(
-        payload=_payload(),
-        client=None,
-    )
-
-    assert result.status == "skipped"
-    assert result.reason == "no_agent_backend_client"
-
-
-def test_cleanup_agent_backend_session_skips_when_session_snapshot_is_missing():
-    payload = _payload().model_copy(update={"session_snapshot": None})
-
-    result = cleanup_agent_backend_session(payload=payload, client=FakeAgentBackendRunClient())
-
-    assert result.status == "skipped"
-    assert result.reason == "missing_session_snapshot"
-
-
-def test_cleanup_agent_backend_session_skips_when_runtime_layer_specs_are_missing():
-    payload = _payload().model_copy(update={"runtime_layer_specs": []})
-
-    result = cleanup_agent_backend_session(payload=payload, client=FakeAgentBackendRunClient())
-
-    assert result.status == "skipped"
-    assert result.reason == "missing_runtime_layer_specs"
-
-
-class _FailedStatusClient(FakeAgentBackendRunClient):
-    def wait_run(self, run_id: str, *, timeout_seconds: float | None = None) -> RunStatusResponse:
-        del timeout_seconds
-        return RunStatusResponse(
-            run_id=run_id,
-            status="failed",
-            created_at=datetime(2026, 1, 1, tzinfo=UTC),
-            updated_at=datetime(2026, 1, 1, tzinfo=UTC),
-            error="snapshot mismatch",
-        )
-
-
-def test_cleanup_agent_backend_session_reports_failed_terminal_status():
-    client = _FailedStatusClient(run_id="cleanup-run-2")
-
-    result = cleanup_agent_backend_session(payload=_payload(), client=client)
-
-    assert result.status == "failed"
-    assert result.reason == "snapshot mismatch"
-    assert result.cleanup_run_id == "cleanup-run-2"
-
-
-class _CreateRunFailureClient(FakeAgentBackendRunClient):
-    def create_run(self, request):  # type: ignore[override]
-        del request
-        raise AgentBackendError("create run failed")
-
-
-class _WaitRunFailureClient(FakeAgentBackendRunClient):
-    def wait_run(self, run_id: str, *, timeout_seconds: float | None = None) -> RunStatusResponse:
-        del run_id, timeout_seconds
-        raise AgentBackendError("wait run failed")
-
-
-def test_cleanup_agent_backend_session_returns_failed_when_create_run_raises():
-    result = cleanup_agent_backend_session(payload=_payload(), client=_CreateRunFailureClient())
-
-    assert result.status == "failed"
-    assert result.reason == "create run failed"
-    assert result.cleanup_run_id is None
-
-
-def test_cleanup_agent_backend_session_returns_failed_with_cleanup_run_id_when_wait_run_raises():
-    result = cleanup_agent_backend_session(
-        payload=_payload(),
-        client=_WaitRunFailureClient(run_id="cleanup-run-3"),
-    )
-
-    assert result.status == "failed"
-    assert result.reason == "wait run failed"
-    assert result.cleanup_run_id == "cleanup-run-3"

api/tests/unit_tests/commands/test_check_no_new_getattr.py

@@ -85,6 +85,42 @@ def main_branch_rev(repo: Path) -> str:
     return git(repo, "rev-parse", "main")
 
 
+@pytest.mark.parametrize(
+    ("source_line", "rule_id"),
+    [
+        (
+            "value = getattr(module, name)  # guard-ignore: no-new-getattr -- lazy export proxy",
+            "no-new-getattr",
+        ),
+        (
+            "session.rollback()  # guard-ignore: no-new-controller-sqlalchemy -- decorator owns rollback",
+            "no-new-controller-sqlalchemy",
+        ),
+    ],
+)
+def test_has_reasoned_guard_ignore_accepts_custom_rules(source_line: str, rule_id: str) -> None:
+    module = load_guard_module()
+
+    assert module.has_reasoned_guard_ignore(source_line, rule_id)
+
+
+@pytest.mark.parametrize(
+    ("source_line", "rule_id"),
+    [
+        ("value = getattr(module, name)  # noqa: no-new-getattr legacy marker", "no-new-getattr"),
+        ("value = getattr(module, name)  # guard-ignore: no-new-getattr", "no-new-getattr"),
+        (
+            "value = getattr(module, name)  # guard-ignore: another-rule -- wrong rule",
+            "no-new-getattr",
+        ),
+    ],
+)
+def test_has_reasoned_guard_ignore_rejects_invalid_markers(source_line: str, rule_id: str) -> None:
+    module = load_guard_module()
+
+    assert not module.has_reasoned_guard_ignore(source_line, rule_id)
+
+
 def test_resolve_ast_grep_command_prefers_ast_grep(monkeypatch: pytest.MonkeyPatch) -> None:
     module = load_guard_module()
     monkeypatch.setattr(
@@ -776,7 +812,7 @@ def test_modified_hunk_with_increased_getattr_count_fails(tmp_path: Path) -> Non
     assert "net-new getattr" in result.stderr
 
 
-def test_inline_noqa_suppression_with_explanatory_text_skips_added_getattr(tmp_path: Path) -> None:
+def test_inline_guard_ignore_with_explanatory_text_skips_added_getattr(tmp_path: Path) -> None:
     init_repo(tmp_path)
     write_repo_file(
         tmp_path,
@@ -795,20 +831,20 @@ def test_inline_noqa_suppression_with_explanatory_text_skips_added_getattr(tmp_p
         "pkg/existing.py",
         """
         def read_value(obj):
-            return getattr(obj, "dynamic_name", None)  # noqa: no-new-getattr needed for plugin-defined attributes
+            return getattr(obj, "dynamic_name", None)  # guard-ignore: no-new-getattr -- plugin-defined attributes
         """,
     )
     commit_all(tmp_path, "add suppressed getattr")
 
     result = run_script(tmp_path, "--base-rev", base_rev)
 
-    assert "no-new-getattr needed for plugin-defined attributes" in (tmp_path / "pkg/existing.py").read_text(
+    assert "guard-ignore: no-new-getattr -- plugin-defined attributes" in (tmp_path / "pkg/existing.py").read_text(
         encoding="utf-8"
     )
     assert result.returncode == 0, stderr_lines(result)
 
 
-def test_inline_noqa_without_explanatory_text_is_not_sufficient(tmp_path: Path) -> None:
+def test_inline_guard_ignore_without_explanatory_text_is_not_sufficient(tmp_path: Path) -> None:
     init_repo(tmp_path)
     write_repo_file(
         tmp_path,
@@ -827,10 +863,10 @@ def test_inline_noqa_without_explanatory_text_is_not_sufficient(tmp_path: Path)
         "pkg/existing.py",
         """
         def read_value(obj):
-            return getattr(obj, "dynamic_name", None)  # noqa: no-new-getattr
+            return getattr(obj, "dynamic_name", None)  # guard-ignore: no-new-getattr
         """,
     )
-    commit_all(tmp_path, "add bare noqa getattr")
+    commit_all(tmp_path, "add bare guard ignore getattr")
 
     result = run_script(tmp_path, "--base-rev", base_rev)
 

api/tests/unit_tests/conftest.py

@@ -43,7 +43,7 @@ from models.account import Account, Tenant, TenantAccountJoin, TenantAccountRole
 from models.base import TypeBase
 
 
-def _patch_redis_clients_on_loaded_modules():
+def _patch_redis_clients_on_loaded_modules() -> None:
     """Ensure any module-level redis_client references point to the shared redis_mock."""
 
     import sys
@@ -51,10 +51,9 @@ def _patch_redis_clients_on_loaded_modules():
     for module in list(sys.modules.values()):
         if module is None:
             continue
-        if hasattr(module, "redis_client"):
-            module.redis_client = redis_mock
-        if hasattr(module, "_pubsub_redis_client"):
-            module.pubsub_redis_client = redis_mock
+        for client_attribute in ("redis_client", "_pubsub_redis_client"):
+            if hasattr(module, client_attribute):
+                setattr(module, client_attribute, redis_mock)
 
 
 @pytest.fixture
@@ -63,13 +62,13 @@ def app() -> Flask:
 
 
 @pytest.fixture(autouse=True)
-def _provide_app_context(app: Flask):
+def _provide_app_context(app: Flask) -> Iterator[None]:
     with app.app_context():
         yield
 
 
 @pytest.fixture(autouse=True)
-def _patch_redis_clients():
+def _patch_redis_clients() -> Iterator[None]:
     """Patch redis_client to MagicMock only for unit test executions."""
 
     with (
@@ -81,7 +80,7 @@ def _patch_redis_clients():
 
 
 @pytest.fixture(autouse=True)
-def reset_redis_mock():
+def reset_redis_mock() -> None:
     """reset the Redis mock before each test"""
     redis_mock.reset_mock()
     redis_mock.get.return_value = None
@@ -91,7 +90,7 @@ def reset_redis_mock():
     redis_mock.exists.return_value = False
     redis_mock.set.return_value = None
     redis_mock.expire.return_value = None
-    redis_mock.hgetall.return_value = {}
+    redis_mock.hgetall.return_value = dict[bytes, bytes]()
     redis_mock.hdel.return_value = None
     redis_mock.incr.return_value = 1
 
@@ -100,7 +99,7 @@ def reset_redis_mock():
 
 
 @pytest.fixture(autouse=True)
-def reset_secret_key():
+def reset_secret_key() -> Iterator[None]:
     """Ensure SECRET_KEY-dependent logic sees an empty config value by default."""
 
     from configs import dify_config
@@ -153,6 +152,18 @@ def _sqlite_session_factory(
     return factory
 
 
+@pytest.fixture
+def _unbound_session_factory(
+    _sqlite_session_factory: sessionmaker[Session],
+    monkeypatch: pytest.MonkeyPatch,
+) -> sessionmaker[Session]:
+    """Create one unbound factory and install it as the global test factory."""
+
+    factory = sessionmaker()
+    monkeypatch.setattr(session_factory_module, "_session_maker", factory)
+    return factory
+
+
 @pytest.fixture
 def sqlite_engine(_sqlite_engine: Engine) -> Engine:
     """Expose the pristine full-schema SQLite engine to tests."""
@@ -179,6 +190,25 @@ def sqlite_session(_sqlite_session_factory: sessionmaker[Session]) -> Iterator[S
         yield session
 
 
+@pytest.fixture
+def unbound_session_factory(_unbound_session_factory: sessionmaker[Session]) -> sessionmaker[Session]:
+    """Expose an unbound factory for paths that must not require persistence."""
+
+    return _unbound_session_factory
+
+
+@pytest.fixture
+def unbound_session(_unbound_session_factory: sessionmaker[Session]) -> Iterator[Session]:
+    """Yield an unbound Session for paths that must not require persistence.
+
+    Bind-requiring database access fails, while bind-free Session operations can
+    still succeed.
+    """
+
+    with _unbound_session_factory() as session:
+        yield session
+
+
 def persist_service_api_tenant_owner(session: Session, tenant: Tenant, owner: Account) -> TenantAccountJoin:
     """Persist the owner identity resolved by service-API app authentication.
 

api/tests/unit_tests/controllers/common/test_agent_app_parameters.py

@@ -61,6 +61,7 @@ def _persist_snapshot(
         tenant_id=tenant_id,
         agent_id=agent_id,
         version=1,
+        home_snapshot_id=_stable_uuid(f"home-snapshot:{snapshot_id}"),
         config_snapshot=config_snapshot,
     )
     session.add(snapshot)

api/tests/unit_tests/controllers/console/agent/test_agent_controllers.py

@@ -279,12 +279,7 @@ def test_agent_app_list_and_create_use_agent_route(
     )
     monkeypatch.setattr(
         roster_controller.AgentRosterService,
-        "get_or_create_agent_app_debug_conversation_id",
-        lambda _self, **kwargs: "debug-conversation-detail",
-    )
-    monkeypatch.setattr(
-        roster_controller.AgentRosterService,
-        "get_or_create_agent_app_debug_conversation_id",
+        "get_or_create_build_conversation",
         lambda _self, **kwargs: "debug-conversation-detail",
     )
     monkeypatch.setattr(
@@ -314,7 +309,7 @@ def test_agent_app_list_and_create_use_agent_route(
     )
     monkeypatch.setattr(
         roster_controller.AgentRosterService,
-        "load_or_create_agent_app_debug_conversation_ids_by_agent_id",
+        "load_or_create_build_conversation_ids_by_agent_id",
         lambda _self, **kwargs: {"agent-list": "debug-conversation-list"},
     )
     monkeypatch.setattr(
@@ -327,7 +322,7 @@ def test_agent_app_list_and_create_use_agent_route(
 
     monkeypatch.setattr(
         roster_controller.AgentRosterService,
-        "get_or_create_agent_app_debug_conversation_id",
+        "get_or_create_build_conversation",
         get_or_create_debug_conversation,
     )
     monkeypatch.setattr(
@@ -387,7 +382,6 @@ def test_agent_app_list_and_create_use_agent_route(
         "tenant_id": "tenant-1",
         "agent_id": "agent-created",
         "account_id": account_id,
-        "draft_type": AgentConfigDraftType.DEBUG_BUILD,
         "commit": False,
     }
 
@@ -454,7 +448,7 @@ def test_agent_app_detail_update_delete_resolve_app_from_agent_id(
     monkeypatch.setattr(roster_controller.AgentRosterService, "get_app_backing_agent", lambda _self, **kwargs: agent)
     monkeypatch.setattr(
         roster_controller.AgentRosterService,
-        "get_or_create_agent_app_debug_conversation_id",
+        "get_or_create_build_conversation",
         lambda _self, **kwargs: "debug-conversation-detail",
     )
     monkeypatch.setattr(
@@ -561,25 +555,16 @@ def test_agent_app_copy_uses_agent_id_and_returns_agent_detail(
     }
 
 
-@pytest.mark.parametrize(
-    ("payload", "expected_draft_type"),
-    [
-        (None, AgentConfigDraftType.DEBUG_BUILD),
-        ({"draft_type": "draft"}, AgentConfigDraftType.DRAFT),
-    ],
-)
-def test_agent_debug_conversation_refresh_uses_current_user_and_draft_type(
+def test_agent_debug_conversation_refresh_resets_build_for_current_user(
     app: Flask,
     monkeypatch: pytest.MonkeyPatch,
     account_id: str,
-    payload: dict[str, str] | None,
-    expected_draft_type: AgentConfigDraftType,
 ) -> None:
     agent_id = "00000000-0000-0000-0000-000000000001"
     captured: dict[str, object] = {}
 
     class FakeRosterService:
-        def refresh_agent_app_debug_conversation_id(self, **kwargs: object) -> str:
+        def reset_build_conversation(self, **kwargs: object) -> str:
             captured.update(kwargs)
             return "new-debug-conversation-id"
 
@@ -587,7 +572,6 @@ def test_agent_debug_conversation_refresh_uses_current_user_and_draft_type(
     with app.test_request_context(
         "/console/api/agent/00000000-0000-0000-0000-000000000001/debug-conversation/refresh",
         method="POST",
-        json=payload,
     ):
         response = unwrap(AgentDebugConversationRefreshApi.post)(
             AgentDebugConversationRefreshApi(), MagicMock(), "tenant-1", SimpleNamespace(id=account_id), agent_id
@@ -601,7 +585,6 @@ def test_agent_debug_conversation_refresh_uses_current_user_and_draft_type(
         "tenant_id": "tenant-1",
         "agent_id": agent_id,
         "account_id": account_id,
-        "draft_type": expected_draft_type,
     }
 
 
@@ -840,7 +823,7 @@ def test_agent_app_update_allows_empty_role(app: Flask, monkeypatch: pytest.Monk
     )
     monkeypatch.setattr(
         roster_controller.AgentRosterService,
-        "get_or_create_agent_app_debug_conversation_id",
+        "get_or_create_build_conversation",
         lambda _self, **kwargs: "debug-conversation-detail",
     )
     monkeypatch.setattr(
@@ -1203,6 +1186,10 @@ def test_workflow_composer_get_uses_write_transaction() -> None:
     assert "@with_session\n    @get_app_model" in getsource(WorkflowAgentComposerApi)
 
 
+def test_build_draft_apply_leaves_transaction_ownership_to_service() -> None:
+    assert "@with_session(write=False)\n    def post" in getsource(AgentBuildDraftApplyApi)
+
+
 def test_workflow_composer_copy_from_roster(app: Flask, monkeypatch: pytest.MonkeyPatch, account_id: str) -> None:
     app_model = SimpleNamespace(id="app-1")
     captured: dict[str, object] = {}
@@ -1500,7 +1487,7 @@ def test_build_chat_finalization_helper_forces_debug_build_and_push_prompt(
     assert args["conversation_id"] == "debug-conversation-1"
     assert args["inputs"] == {}
     assert args["auto_generate_name"] is False
-    assert args[completion_controller.AGENT_RUNTIME_EXIT_INTENT_ARG] == "delete"
+    assert "_agent_runtime_exit_intent" not in args
     assert args["external_trace_id"] == "trace-1"
 
 
@@ -1643,7 +1630,7 @@ def test_agent_chat_helper_ignores_private_exit_intent_payload_key(
             "inputs": {},
             "query": "hello",
             "response_mode": "streaming",
-            completion_controller.AGENT_RUNTIME_EXIT_INTENT_ARG: "delete",
+            "_agent_runtime_exit_intent": "delete",
         }
     ):
         result = completion_controller._create_chat_message(
@@ -1657,7 +1644,7 @@ def test_agent_chat_helper_ignores_private_exit_intent_payload_key(
     args = cast(dict[str, object], captured["args"])
     assert args["response_mode"] == "streaming"
     assert args["conversation_id"] == "debug-conversation-1"
-    assert completion_controller.AGENT_RUNTIME_EXIT_INTENT_ARG not in args
+    assert "_agent_runtime_exit_intent" not in args
 
 
 @pytest.mark.parametrize(
@@ -1717,14 +1704,18 @@ def test_resolve_current_user_agent_debug_conversation_uses_agent_or_backing_app
         def __init__(self, session: object) -> None:
             calls.append({"session": session})
 
-        def get_or_create_agent_app_debug_conversation_id(self, **kwargs: object) -> str:
-            calls.append({"get_or_create": kwargs})
+        def get_or_create_build_conversation(self, **kwargs: object) -> str:
+            calls.append({"get_build": kwargs})
             return f"debug-{kwargs['agent_id']}"
 
-        def refresh_agent_app_debug_conversation_id(self, **kwargs: object) -> str:
-            calls.append({"refresh": kwargs})
+        def rotate_preview_conversation(self, **kwargs: object) -> str:
+            calls.append({"rotate_preview": kwargs})
             return f"new-{kwargs['agent_id']}"
 
+        def get_current_preview_conversation(self, **kwargs: object) -> str:
+            calls.append({"get_preview": kwargs})
+            return f"preview-{kwargs['agent_id']}"
+
         def get_app_backing_agent(self, **kwargs: object) -> object:
             calls.append({"get_app_backing_agent": kwargs})
             return SimpleNamespace(id="backing-agent")
@@ -1756,33 +1747,46 @@ def test_resolve_current_user_agent_debug_conversation_uses_agent_or_backing_app
         draft_type=AgentConfigDraftType.DRAFT,
         start_new=True,
     )
+    current_preview_id = completion_controller._resolve_current_user_agent_debug_conversation_id(
+        session="session-1",  # type: ignore[arg-type]
+        current_tenant_id="tenant-1",
+        current_user=SimpleNamespace(id="account-1"),
+        app_model=SimpleNamespace(id="app-1"),
+        agent_id="agent-1",
+        draft_type=AgentConfigDraftType.DRAFT,
+    )
     assert explicit_id == "new-agent-1"
     assert fallback_id == "debug-backing-agent"
     assert fallback_preview_id == "new-backing-agent"
+    assert current_preview_id == "preview-agent-1"
     assert calls[1] == {
-        "refresh": {
+        "rotate_preview": {
             "tenant_id": "tenant-1",
             "agent_id": "agent-1",
             "account_id": "account-1",
-            "draft_type": AgentConfigDraftType.DRAFT,
         }
     }
     assert calls[3] == {"get_app_backing_agent": {"tenant_id": "tenant-1", "app_id": "app-1"}}
     assert calls[4] == {
-        "get_or_create": {
+        "get_build": {
             "tenant_id": "tenant-1",
             "agent_id": "backing-agent",
             "account_id": "account-1",
-            "draft_type": AgentConfigDraftType.DEBUG_BUILD,
         }
     }
     assert calls[6] == {"get_app_backing_agent": {"tenant_id": "tenant-1", "app_id": "app-1"}}
     assert calls[7] == {
-        "refresh": {
+        "rotate_preview": {
             "tenant_id": "tenant-1",
             "agent_id": "backing-agent",
             "account_id": "account-1",
-            "draft_type": AgentConfigDraftType.DRAFT,
+        }
+    }
+    assert calls[9] == {
+        "get_preview": {
+            "tenant_id": "tenant-1",
+            "agent_id": "agent-1",
+            "account_id": "account-1",
         }
     }
 

api/tests/unit_tests/controllers/console/app/test_agent_app_sandbox.py

@@ -6,7 +6,7 @@ from unittest.mock import MagicMock
 
 import pytest
 from dify_agent.client import DifyAgentClientError, DifyAgentHTTPError, DifyAgentTimeoutError
-from dify_agent.protocol import SandboxListResponse, SandboxReadResponse
+from dify_agent.protocol import WorkspaceListResponse, WorkspaceReadResponse
 
 from controllers.console import agent_app_sandbox as module
 from models.model import App, AppMode, IconType
@@ -15,30 +15,67 @@ from services.agent_app_sandbox_service import AgentSandboxInfo, AgentSandboxIns
 
 class _AgentAppService:
     def __init__(self) -> None:
-        self.calls: list[tuple[str, str, str, str, str]] = []
+        self.calls: list[tuple[str, str, str, str, str, str, str, str]] = []
 
-    def get_info(self, *, tenant_id: str, app_id: str, conversation_id: str) -> AgentSandboxInfo:
-        self.calls.append(("info", tenant_id, app_id, conversation_id, ""))
-        return AgentSandboxInfo(session_id="abc1234", workspace_cwd="~/workspace/abc1234")
+    def get_info(
+        self,
+        *,
+        tenant_id: str,
+        app_id: str,
+        agent_id: str,
+        caller_type: str,
+        caller_id: str,
+        account_id: str,
+    ) -> AgentSandboxInfo:
+        self.calls.append(("info", tenant_id, app_id, agent_id, caller_type, caller_id, account_id, ""))
+        return AgentSandboxInfo(workspace_cwd=".")
 
-    def list_files(self, *, tenant_id: str, app_id: str, conversation_id: str, path: str) -> SandboxListResponse:
-        self.calls.append(("list", tenant_id, app_id, conversation_id, path))
-        return SandboxListResponse(path=path, entries=[], truncated=False)
+    def list_files(
+        self,
+        *,
+        tenant_id: str,
+        app_id: str,
+        agent_id: str,
+        caller_type: str,
+        caller_id: str,
+        account_id: str,
+        path: str,
+    ) -> WorkspaceListResponse:
+        self.calls.append(("list", tenant_id, app_id, agent_id, caller_type, caller_id, account_id, path))
+        return WorkspaceListResponse(path=path, entries=[], truncated=False)
 
-    def read_file(self, *, tenant_id: str, app_id: str, conversation_id: str, path: str) -> SandboxReadResponse:
-        self.calls.append(("read", tenant_id, app_id, conversation_id, path))
-        return SandboxReadResponse(path=path, size=5, truncated=False, binary=False, text="hello")
+    def read_file(
+        self,
+        *,
+        tenant_id: str,
+        app_id: str,
+        agent_id: str,
+        caller_type: str,
+        caller_id: str,
+        account_id: str,
+        path: str,
+    ) -> WorkspaceReadResponse:
+        self.calls.append(("read", tenant_id, app_id, agent_id, caller_type, caller_id, account_id, path))
+        return WorkspaceReadResponse(path=path, size=5, truncated=False, binary=False, text="hello")
 
     def upload_file(
-        self, *, tenant_id: str, app_id: str, conversation_id: str, path: str
+        self,
+        *,
+        tenant_id: str,
+        app_id: str,
+        agent_id: str,
+        caller_type: str,
+        caller_id: str,
+        account_id: str,
+        path: str,
     ) -> AgentSandboxUploadDownload:
-        self.calls.append(("upload", tenant_id, app_id, conversation_id, path))
+        self.calls.append(("upload", tenant_id, app_id, agent_id, caller_type, caller_id, account_id, path))
         return AgentSandboxUploadDownload(url="https://files.example/report.txt")
 
 
 class _WorkflowService:
     def __init__(self) -> None:
-        self.calls: list[tuple[str, str, str, str, str, str | None, str]] = []
+        self.calls: list[tuple[str, str, str, str, str, str]] = []
 
     def list_files(
         self,
@@ -47,12 +84,12 @@ class _WorkflowService:
         app_id: str,
         workflow_run_id: str,
         node_id: str,
-        node_execution_id: str | None,
+        node_execution_id: str,
         path: str,
         session,
-    ) -> SandboxListResponse:
-        self.calls.append(("list", tenant_id, app_id, workflow_run_id, node_id, node_execution_id, path))
-        return SandboxListResponse(path=path, entries=[], truncated=False)
+    ) -> WorkspaceListResponse:
+        self.calls.append(("list", tenant_id, app_id, workflow_run_id, node_id, path))
+        return WorkspaceListResponse(path=path, entries=[], truncated=False)
 
     def read_file(
         self,
@@ -61,12 +98,12 @@ class _WorkflowService:
         app_id: str,
         workflow_run_id: str,
         node_id: str,
-        node_execution_id: str | None,
+        node_execution_id: str,
         path: str,
         session,
-    ) -> SandboxReadResponse:
-        self.calls.append(("read", tenant_id, app_id, workflow_run_id, node_id, node_execution_id, path))
-        return SandboxReadResponse(path=path, size=5, truncated=False, binary=False, text="hello")
+    ) -> WorkspaceReadResponse:
+        self.calls.append(("read", tenant_id, app_id, workflow_run_id, node_id, path))
+        return WorkspaceReadResponse(path=path, size=5, truncated=False, binary=False, text="hello")
 
     def upload_file(
         self,
@@ -75,11 +112,11 @@ class _WorkflowService:
         app_id: str,
         workflow_run_id: str,
         node_id: str,
-        node_execution_id: str | None,
+        node_execution_id: str,
         path: str,
         session,
     ) -> AgentSandboxUploadDownload:
-        self.calls.append(("upload", tenant_id, app_id, workflow_run_id, node_id, node_execution_id, path))
+        self.calls.append(("upload", tenant_id, app_id, workflow_run_id, node_id, path))
         return AgentSandboxUploadDownload(url="https://files.example/upload.txt")
 
 
@@ -125,34 +162,41 @@ def test_handle_maps_sandbox_and_agent_backend_errors() -> None:
 def test_agent_app_sandbox_resources_proxy_service(monkeypatch: pytest.MonkeyPatch) -> None:
     service = _AgentAppService()
     session = MagicMock()
+    account = SimpleNamespace(id="account-1")
     resolver = MagicMock(return_value=_app_model())
     monkeypatch.setattr(module, "AgentAppSandboxService", lambda: service)
     monkeypatch.setattr(module, "resolve_agent_runtime_app_model", resolver)
     monkeypatch.setattr(
         module,
         "query_params_from_request",
-        lambda model: SimpleNamespace(conversation_id="conv-1", path="sub/report.txt"),
+        lambda model: SimpleNamespace(caller_type="build_draft", caller_id="build-1", path="sub/report.txt"),
     )
     monkeypatch.setattr(
         module,
         "request",
-        SimpleNamespace(get_json=lambda silent=True: {"conversation_id": "conv-1", "path": "report.txt"}),
+        SimpleNamespace(
+            get_json=lambda silent=True: {
+                "caller_type": "build_draft",
+                "caller_id": "build-1",
+                "path": "report.txt",
+            }
+        ),
     )
 
-    info = unwrap(module.AgentAppSandboxInfoResource.get)(object(), session, "tenant-1", "agent-1")
-    listing = unwrap(module.AgentAppSandboxListResource.get)(object(), session, "tenant-1", "agent-1")
-    preview = unwrap(module.AgentAppSandboxReadResource.get)(object(), session, "tenant-1", "agent-1")
-    upload = unwrap(module.AgentAppSandboxUploadResource.post)(object(), session, "tenant-1", "agent-1")
+    info = unwrap(module.AgentAppSandboxInfoResource.get)(object(), session, account, "tenant-1", "agent-1")
+    listing = unwrap(module.AgentAppSandboxListResource.get)(object(), session, account, "tenant-1", "agent-1")
+    preview = unwrap(module.AgentAppSandboxReadResource.get)(object(), session, account, "tenant-1", "agent-1")
+    upload = unwrap(module.AgentAppSandboxUploadResource.post)(object(), session, account, "tenant-1", "agent-1")
 
-    assert info == {"session_id": "abc1234", "workspace_cwd": "~/workspace/abc1234"}
+    assert info == {"workspace_cwd": "."}
     assert listing["path"] == "sub/report.txt"
     assert preview["text"] == "hello"
     assert upload == {"url": "https://files.example/report.txt"}
     assert service.calls == [
-        ("info", "tenant-1", "app-1", "conv-1", ""),
-        ("list", "tenant-1", "app-1", "conv-1", "sub/report.txt"),
-        ("read", "tenant-1", "app-1", "conv-1", "sub/report.txt"),
-        ("upload", "tenant-1", "app-1", "conv-1", "report.txt"),
+        ("info", "tenant-1", "app-1", "agent-1", "build_draft", "build-1", "account-1", ""),
+        ("list", "tenant-1", "app-1", "agent-1", "build_draft", "build-1", "account-1", "sub/report.txt"),
+        ("read", "tenant-1", "app-1", "agent-1", "build_draft", "build-1", "account-1", "sub/report.txt"),
+        ("upload", "tenant-1", "app-1", "agent-1", "build_draft", "build-1", "account-1", "report.txt"),
     ]
     assert all(call.kwargs["session"] is session for call in resolver.call_args_list)
 
@@ -160,24 +204,27 @@ def test_agent_app_sandbox_resources_proxy_service(monkeypatch: pytest.MonkeyPat
 def test_agent_app_sandbox_resource_returns_normalized_errors(monkeypatch: pytest.MonkeyPatch) -> None:
     class FailingService:
         def get_info(self, **kwargs):
-            raise AgentSandboxInspectorError("no_active_session", "no active session", status_code=404)
+            raise AgentSandboxInspectorError("no_active_binding", "no active binding", status_code=404)
 
         def list_files(self, **kwargs):
-            raise AgentSandboxInspectorError("no_active_session", "no active session", status_code=404)
+            raise AgentSandboxInspectorError("no_active_binding", "no active binding", status_code=404)
 
     monkeypatch.setattr(module, "AgentAppSandboxService", FailingService)
     session = MagicMock()
+    account = SimpleNamespace(id="account-1")
     monkeypatch.setattr(module, "resolve_agent_runtime_app_model", MagicMock(return_value=_app_model()))
     monkeypatch.setattr(
-        module, "query_params_from_request", lambda model: SimpleNamespace(conversation_id="conv-1", path=".")
+        module,
+        "query_params_from_request",
+        lambda model: SimpleNamespace(caller_type="conversation", caller_id="conv-1", path="."),
     )
 
-    assert unwrap(module.AgentAppSandboxInfoResource.get)(object(), session, "tenant-1", "agent-1") == (
-        {"code": "no_active_session", "message": "no active session"},
+    assert unwrap(module.AgentAppSandboxInfoResource.get)(object(), session, account, "tenant-1", "agent-1") == (
+        {"code": "no_active_binding", "message": "no active binding"},
         404,
     )
-    assert unwrap(module.AgentAppSandboxListResource.get)(object(), session, "tenant-1", "agent-1") == (
-        {"code": "no_active_session", "message": "no active session"},
+    assert unwrap(module.AgentAppSandboxListResource.get)(object(), session, account, "tenant-1", "agent-1") == (
+        {"code": "no_active_binding", "message": "no active binding"},
         404,
     )
 
@@ -188,12 +235,12 @@ def test_workflow_agent_sandbox_resources_proxy_service(monkeypatch: pytest.Monk
     monkeypatch.setattr(
         module,
         "query_params_from_request",
-        lambda model: SimpleNamespace(node_execution_id="exec-1", path="out.txt"),
+        lambda model: SimpleNamespace(node_execution_id="execution-1", path="out.txt"),
     )
     monkeypatch.setattr(
         module,
         "request",
-        SimpleNamespace(get_json=lambda silent=True: {"node_execution_id": "exec-1", "path": "upload.txt"}),
+        SimpleNamespace(get_json=lambda silent=True: {"node_execution_id": "execution-1", "path": "upload.txt"}),
     )
     app_model = _app_model()
 
@@ -211,7 +258,7 @@ def test_workflow_agent_sandbox_resources_proxy_service(monkeypatch: pytest.Monk
     assert preview["text"] == "hello"
     assert upload == {"url": "https://files.example/upload.txt"}
     assert service.calls == [
-        ("list", "tenant-1", "app-1", "run-1", "agent-node", "exec-1", "out.txt"),
-        ("read", "tenant-1", "app-1", "run-1", "agent-node", "exec-1", "out.txt"),
-        ("upload", "tenant-1", "app-1", "run-1", "agent-node", "exec-1", "upload.txt"),
+        ("list", "tenant-1", "app-1", "run-1", "agent-node", "out.txt"),
+        ("read", "tenant-1", "app-1", "run-1", "agent-node", "out.txt"),
+        ("upload", "tenant-1", "app-1", "run-1", "agent-node", "upload.txt"),
     ]

api/tests/unit_tests/controllers/test_swagger.py

@@ -574,7 +574,7 @@ def test_console_account_avatar_query_param_renders_as_query(monkeypatch: pytest
     assert params["avatar"]["required"] is True
 
 
-def test_console_agent_debug_conversation_refresh_body_is_optional(monkeypatch: pytest.MonkeyPatch):
+def test_console_agent_debug_conversation_refresh_has_no_body(monkeypatch: pytest.MonkeyPatch):
     from configs import dify_config
     from controllers.console import bp as console_bp
 
@@ -586,13 +586,9 @@ def test_console_agent_debug_conversation_refresh_body_is_optional(monkeypatch:
 
     payload = app.test_client().get("/console/api/openapi.json").get_json()
     operation = payload["paths"]["/agent/{agent_id}/debug-conversation/refresh"]["post"]
-    request_body = operation["requestBody"]
 
-    assert request_body["required"] is False
-    assert request_body["content"]["application/json"]["schema"] == {
-        "$ref": "#/components/schemas/AgentDebugConversationRefreshPayload"
-    }
-    assert "AgentDebugConversationRefreshPayload" in payload["components"]["schemas"]
+    assert "requestBody" not in operation
+    assert "AgentDebugConversationRefreshPayload" not in payload["components"]["schemas"]
 
 
 def test_console_member_invite_documents_bad_request_response(monkeypatch: pytest.MonkeyPatch):

api/tests/unit_tests/core/agent/test_publish_visibility.py

@@ -71,6 +71,7 @@ def _add_agent(
             tenant_id="tenant-1",
             agent_id=agent_id,
             version=1,
+            home_snapshot_id=f"home-{agent_id}",
             config_snapshot=_agent_soul(),
         )
     )
@@ -245,6 +246,7 @@ def test_workflow_callable_filter_distinguishes_never_published_from_dirty_draft
             tenant_id="tenant-1",
             agent_id=stale_published_snapshot.id,
             version=2,
+            home_snapshot_id="home-stale-published-old",
             config_snapshot=_agent_soul(),
         )
     )

api/tests/unit_tests/core/app/apps/agent_app/test_app_generator.py

@@ -22,11 +22,10 @@ from core.app.apps.agent_app.app_generator import (
     AgentAppGeneratorError,
 )
 from core.app.apps.exc import GenerateTaskStoppedError
-from core.app.entities.app_invoke_entities import AGENT_RUNTIME_EXIT_INTENT_ARG, InvokeFrom, UserFrom
+from core.app.entities.app_invoke_entities import InvokeFrom, UserFrom
 from core.app.entities.queue_entities import QueueAnnotationReplyEvent
 from core.workflow.file_reference import build_file_reference
 from models import Account, AppModelConfig
-from models.agent import AgentConfigDraftType
 
 MODULE = "core.app.apps.agent_app.app_generator"
 
@@ -79,14 +78,18 @@ class TestGenerateGuards:
 
 
 class TestGenerateSuccess:
-    def test_runtime_session_snapshot_id_preserves_snapshot_for_debugger_and_web_app(self):
+    def test_session_scope_config_version_id_preserves_draft_or_snapshot_id(self):
         assert (
-            AgentAppGenerator._runtime_session_snapshot_id(invoke_from=InvokeFrom.DEBUGGER, snapshot_id="snap-1")
-            == "snap-1"
+            AgentAppGenerator._session_scope_config_version_id(
+                invoke_from=InvokeFrom.DEBUGGER, config_version_id="draft-1"
+            )
+            == "draft-1"
         )
         assert (
-            AgentAppGenerator._runtime_session_snapshot_id(invoke_from=InvokeFrom.WEB_APP, snapshot_id="snap-1")
-            == "snap-1"
+            AgentAppGenerator._session_scope_config_version_id(
+                invoke_from=InvokeFrom.WEB_APP, config_version_id="snapshot-1"
+            )
+            == "snapshot-1"
         )
 
     def test_generate_orchestrates_and_starts_worker(self, generator, mocker: MockerFixture):
@@ -145,84 +148,11 @@ class TestGenerateSuccess:
             draft_type=None,
             user=user,
             session=session,
+            conversation=None,
         )
         session.get.assert_called_once_with(AppModelConfig, "config-1")
         assert generate_entity.call_args.kwargs["prompt_file_mappings"] == file_mappings
-        assert generate_entity.call_args.kwargs["agent_runtime_exit_intent"] == "suspend"
-
-    def test_generate_uses_delete_exit_intent_from_internal_arg(self, generator, mocker: MockerFixture):
-        app_model = mocker.MagicMock(id="app1", tenant_id="tenant", mode="agent")
-        user = DummyAccount("user")
-
-        generator._resolve_agent = mocker.MagicMock(
-            return_value=(mocker.MagicMock(id="agent1"), "snap1", "snapshot", mocker.MagicMock())
-        )
-        generator._prepare_user_inputs = mocker.MagicMock(return_value={})
-        generator._init_generate_records = mocker.MagicMock(
-            return_value=(mocker.MagicMock(id="conv", mode="agent"), mocker.MagicMock(id="msg"))
-        )
-        generator._handle_response = mocker.MagicMock(return_value="raw-response")
-
-        mocker.patch(
-            f"{MODULE}.AgentAppConfigManager.get_app_config",
-            return_value=mocker.MagicMock(variables=[], tenant_id="tenant", app_id="app1"),
-        )
-        mocker.patch(f"{MODULE}.ModelConfigConverter.convert", return_value=mocker.MagicMock(model="gpt-4o-mini"))
-        mocker.patch(f"{MODULE}.TraceQueueManager", return_value=mocker.MagicMock())
-        generate_entity = mocker.patch(
-            f"{MODULE}.AgentAppGenerateEntity", return_value=mocker.MagicMock(task_id="t", user_id="user")
-        )
-        mocker.patch(f"{MODULE}.MessageBasedAppQueueManager", return_value=mocker.MagicMock())
-        mocker.patch(f"{MODULE}.threading.Thread", return_value=mocker.MagicMock())
-        mocker.patch(f"{MODULE}.AgentAppGenerateResponseConverter.convert", return_value={"result": "ok"})
-
-        generator.generate(
-            app_model=app_model,
-            user=user,
-            args={"query": "hello", "inputs": {}, AGENT_RUNTIME_EXIT_INTENT_ARG: "delete"},
-            invoke_from=InvokeFrom.DEBUGGER,
-            session=mocker.MagicMock(),
-            streaming=True,
-        )
-
-        assert generate_entity.call_args.kwargs["agent_runtime_exit_intent"] == "delete"
-
-    def test_generate_falls_back_to_suspend_for_invalid_internal_exit_intent(self, generator, mocker: MockerFixture):
-        app_model = mocker.MagicMock(id="app1", tenant_id="tenant", mode="agent")
-        user = DummyAccount("user")
-
-        generator._resolve_agent = mocker.MagicMock(
-            return_value=(mocker.MagicMock(id="agent1"), "snap1", "snapshot", mocker.MagicMock())
-        )
-        generator._prepare_user_inputs = mocker.MagicMock(return_value={})
-        generator._init_generate_records = mocker.MagicMock(
-            return_value=(mocker.MagicMock(id="conv", mode="agent"), mocker.MagicMock(id="msg"))
-        )
-        generator._handle_response = mocker.MagicMock(return_value="raw-response")
-
-        mocker.patch(
-            f"{MODULE}.AgentAppConfigManager.get_app_config",
-            return_value=mocker.MagicMock(variables=[], tenant_id="tenant", app_id="app1"),
-        )
-        mocker.patch(f"{MODULE}.ModelConfigConverter.convert", return_value=mocker.MagicMock(model="gpt-4o-mini"))
-        mocker.patch(f"{MODULE}.TraceQueueManager", return_value=mocker.MagicMock())
-        generate_entity = mocker.patch(
-            f"{MODULE}.AgentAppGenerateEntity", return_value=mocker.MagicMock(task_id="t", user_id="user")
-        )
-        mocker.patch(f"{MODULE}.MessageBasedAppQueueManager", return_value=mocker.MagicMock())
-        mocker.patch(f"{MODULE}.threading.Thread", return_value=mocker.MagicMock())
-        mocker.patch(f"{MODULE}.AgentAppGenerateResponseConverter.convert", return_value={"result": "ok"})
-
-        generator.generate(
-            app_model=app_model,
-            user=user,
-            args={"query": "hello", "inputs": {}, AGENT_RUNTIME_EXIT_INTENT_ARG: "bogus"},
-            invoke_from=InvokeFrom.DEBUGGER,
-            session=mocker.MagicMock(),
-            streaming=True,
-        )
-
-        assert generate_entity.call_args.kwargs["agent_runtime_exit_intent"] == "suspend"
+        assert "agent_runtime_exit_intent" not in generate_entity.call_args.kwargs
 
     def test_generate_loads_existing_conversation(self, generator: AgentAppGenerator, mocker: MockerFixture):
         app_model = mocker.MagicMock(id="app1", tenant_id="tenant", mode="agent")
@@ -263,6 +193,7 @@ class TestGenerateSuccess:
             user=user,
             session=session,
         )
+        assert generator._resolve_agent.call_args.kwargs["conversation"].id == "conv"
         assert generator._init_generate_records.call_args.kwargs["session"] is session
 
     def test_generate_does_not_include_trace_session_id_in_extras(
@@ -326,8 +257,10 @@ class TestGenerateWorker:
         generator._get_conversation = mocker.MagicMock(return_value=mocker.MagicMock(id="conv"))
         generator._get_message = mocker.MagicMock(return_value=mocker.MagicMock(id="msg"))
         generator._run_input_guards = mocker.MagicMock(return_value=(handled, guard_query, None))
+        resolved_agent = mocker.MagicMock(id="a")
+        resolved_config = mocker.MagicMock(id="s", home_snapshot_id="home-1")
         generator._resolve_agent_by_id = mocker.MagicMock(
-            return_value=(mocker.MagicMock(), mocker.MagicMock(), mocker.MagicMock())
+            return_value=(resolved_agent, resolved_config, mocker.MagicMock())
         )
         session = mocker.MagicMock()
         session.get.return_value = mocker.MagicMock(id="app1")
@@ -345,7 +278,7 @@ class TestGenerateWorker:
         mocker.patch(f"{MODULE}.AgentAppRuntimeRequestBuilder", return_value=mocker.MagicMock())
         mocker.patch(f"{MODULE}.create_agent_backend_run_client", return_value=mocker.MagicMock())
         mocker.patch(f"{MODULE}.AgentBackendRunEventAdapter", return_value=mocker.MagicMock())
-        mocker.patch(f"{MODULE}.AgentAppRuntimeSessionStore", return_value=mocker.MagicMock())
+        mocker.patch(f"{MODULE}.AgentAppWorkspaceStore", return_value=mocker.MagicMock())
         runner = mocker.MagicMock()
         if run_side_effect is not None:
             runner.run.side_effect = run_side_effect
@@ -360,9 +293,8 @@ class TestGenerateWorker:
         *,
         is_resume=False,
         query="query",
-        runtime_session_snapshot_id="s",
+        session_scope_config_version_id="s",
         prompt_file_mappings=(),
-        agent_runtime_exit_intent="suspend",
     ):
         generator._generate_worker(
             flask_app=mocker.MagicMock(),
@@ -370,8 +302,7 @@ class TestGenerateWorker:
             application_generate_entity=mocker.MagicMock(
                 agent_id="a",
                 agent_config_snapshot_id="s",
-                agent_runtime_session_snapshot_id=runtime_session_snapshot_id,
-                agent_runtime_exit_intent=agent_runtime_exit_intent,
+                agent_session_scope_config_version_id=session_scope_config_version_id,
                 model_conf=mocker.MagicMock(model="m"),
                 query=query,
                 prompt_file_mappings=prompt_file_mappings,
@@ -389,25 +320,19 @@ class TestGenerateWorker:
         self._call(generator, mocker, queue_manager)
         runner.run.assert_called_once()
         assert generator._resolve_agent_by_id.call_args.kwargs["session"] is resolver_session
+        assert runner.run.call_args.kwargs["home_snapshot_id"] == "home-1"
+        assert "home_snapshot_ref" not in runner.run.call_args.kwargs
         queue_manager.publish_error.assert_not_called()
 
-    def test_worker_passes_runtime_session_scope_to_runner(self, generator, mocker: MockerFixture):
+    def test_worker_passes_session_scope_config_version_to_runner(self, generator, mocker: MockerFixture):
         runner, _ = self._wire(generator, mocker)
         queue_manager = mocker.MagicMock()
 
-        self._call(generator, mocker, queue_manager, runtime_session_snapshot_id=None)
+        self._call(generator, mocker, queue_manager, session_scope_config_version_id=None)
 
         assert runner.run.call_args.kwargs["agent_config_snapshot_id"] == "s"
         assert runner.run.call_args.kwargs["session_scope_snapshot_id"] is None
 
-    def test_worker_forwards_runtime_exit_intent_to_runner(self, generator, mocker: MockerFixture):
-        runner, _ = self._wire(generator, mocker)
-        queue_manager = mocker.MagicMock()
-
-        self._call(generator, mocker, queue_manager, agent_runtime_exit_intent="delete")
-
-        assert runner.run.call_args.kwargs["agent_runtime_exit_intent"] == "delete"
-
     def test_worker_appends_prompt_files_to_backend_query(self, generator, mocker: MockerFixture):
         runner, _ = self._wire(generator, mocker, guard_query="你看得见这张图片吗")
         queue_manager = mocker.MagicMock()
@@ -526,7 +451,7 @@ class TestResumeAfterFormSubmission:
         mocker.patch(f"{MODULE}.TraceQueueManager", return_value=mocker.MagicMock())
         mocker.patch(f"{MODULE}.MessageBasedAppQueueManager", return_value=mocker.MagicMock())
         mocker.patch(f"{MODULE}.threading.Thread", return_value=mocker.MagicMock())
-        mocker.patch(f"{MODULE}.AgentAppRuntimeSessionStore")
+        generator._resolve_resume_draft = mocker.MagicMock(return_value=(None, None))
         return (
             mocker.patch(
                 f"{MODULE}.AgentAppGenerateEntity", return_value=mocker.MagicMock(task_id="t", user_id="user")
@@ -547,6 +472,7 @@ class TestResumeAfterFormSubmission:
             app_model=app_model,
             user=user,
             conversation_id="conv",
+            form_id="form-1",
             invoke_from=InvokeFrom.WEB_APP,
             session=session,
         )
@@ -573,6 +499,7 @@ class TestResumeAfterFormSubmission:
             app_model=mocker.MagicMock(id="app1", tenant_id="tenant", mode="agent"),
             user=DummyAccount("user"),
             conversation_id="conv",
+            form_id="form-1",
             invoke_from=InvokeFrom.WEB_APP,
             session=session,
         )
@@ -584,25 +511,24 @@ class TestResumeAfterFormSubmission:
         self._wire(generator, mocker)
         conversation = mocker.MagicMock(id="conv", invoke_from=InvokeFrom.DEBUGGER)
         mocker.patch(f"{MODULE}.ConversationService.get_conversation", return_value=conversation)
-        session_store = mocker.patch(f"{MODULE}.AgentAppRuntimeSessionStore")
-        session_store.return_value.load_active_session_for_conversation.return_value = mocker.MagicMock(
-            scope=mocker.MagicMock(agent_config_snapshot_id="draft-build-1")
-        )
-        draft_row = mocker.MagicMock(draft_type=AgentConfigDraftType.DEBUG_BUILD, account_id="user")
+        generator._resolve_resume_draft.return_value = ("debug_build", "draft-build-1")
         account_user = mocker.MagicMock(spec=Account)
         account_user.id = "user"
         app_model = mocker.MagicMock(id="app1", tenant_id="tenant", mode="agent")
         app_model.app_model_config_id = "config-1"
         session = mocker.MagicMock()
-        session.scalar.side_effect = [draft_row, mocker.MagicMock(query="original question")]
+        session.scalar.return_value = mocker.MagicMock(query="original question")
 
         generator.resume_after_form_submission(
             app_model=app_model,
             user=account_user,
             conversation_id="conv",
+            form_id="form-1",
             invoke_from=InvokeFrom.DEBUGGER,
             session=session,
         )
 
         assert generator._resolve_agent.call_args.kwargs["draft_type"] == "debug_build"
+        assert generator._resolve_agent.call_args.kwargs["draft_id"] == "draft-build-1"
         assert generator._resolve_agent.call_args.kwargs["session"] is session
+        assert generator._resolve_agent.call_args.kwargs["conversation"] is conversation

api/tests/unit_tests/core/app/apps/agent_app/test_app_runner.py

@@ -23,7 +23,6 @@ from dify_agent.protocol import (
     RunStartedEvent,
     RunSucceededEvent,
     RunSucceededEventData,
-    RuntimeLayerSpec,
 )
 from pydantic_ai.messages import (
     FunctionToolCallEvent,
@@ -36,7 +35,6 @@ from pydantic_ai.messages import (
 )
 
 from clients.agent_backend import (
-    AgentBackendError,
     AgentBackendRunEventAdapter,
     AgentBackendRunFailedError,
     AgentBackendRunFailedInternalEvent,
@@ -155,43 +153,6 @@ class _StreamingFakeAgentBackendRunClient(FakeAgentBackendRunClient):
         )
 
 
-class _StreamingRecordingFakeAgentBackendRunClient(_RecordingFakeAgentBackendRunClient):
-    @override
-    def stream_events(
-        self,
-        run_id: str,
-        *,
-        after: str | None = None,
-        should_stop: Callable[[], bool] | None = None,
-    ) -> Iterator[RunEvent]:
-        del after, should_stop
-        created_at = datetime(2026, 1, 1, tzinfo=UTC)
-        yield RunStartedEvent(id="1-0", run_id=run_id, created_at=created_at)
-        yield PydanticAIStreamRunEvent(
-            id="2-0",
-            run_id=run_id,
-            created_at=created_at,
-            data=PartDeltaEvent(index=0, delta=TextPartDelta(content_delta="hello ")),
-            agent_message_delta="hello ",
-        )
-        yield PydanticAIStreamRunEvent(
-            id="3-0",
-            run_id=run_id,
-            created_at=created_at,
-            data=PartDeltaEvent(index=0, delta=TextPartDelta(content_delta="agent")),
-            agent_message_delta="agent",
-        )
-        yield RunSucceededEvent(
-            id="4-0",
-            run_id=run_id,
-            created_at=created_at,
-            data=RunSucceededEventData(
-                output={"text": "hello agent"},
-                session_snapshot=CompositorSessionSnapshot(layers=[]),
-            ),
-        )
-
-
 class _StreamingStopAfterFirstDeltaFakeAgentBackendRunClient(_RecordingFakeAgentBackendRunClient):
     def __init__(self, *, queue_manager: _FakeQueueManager, **kwargs: Any) -> None:
         super().__init__(**kwargs)
@@ -419,60 +380,48 @@ class _FakeSessionStore:
         self,
         loaded: CompositorSessionSnapshot | None = None,
         loaded_session: StoredAgentAppSession | None = None,
-        listed_sessions: list[StoredAgentAppSession] | None = None,
+        binding_id: str = "binding-1",
+        workspace_id: str = "workspace-1",
+        backend_binding_ref: str = "backend-binding-1",
     ) -> None:
         self.loaded = loaded
         self._loaded_session = loaded_session
-        self._listed_sessions = list(listed_sessions or [])
-        self.loaded_scopes: list[AgentAppSessionScope] = []
+        self.binding_id = binding_id
+        self.workspace_id = workspace_id
+        self.backend_binding_ref = backend_binding_ref
+        self.resolved_scopes: list[AgentAppSessionScope] = []
         self.saved: list[
             tuple[
                 AgentAppSessionScope,
                 str,
                 CompositorSessionSnapshot | None,
-                list[RuntimeLayerSpec],
                 str | None,
                 str | None,
             ]
         ] = []
-        self.cleaned: list[tuple[AgentAppSessionScope, str | None]] = []
 
-    def load_active_snapshot(self, scope: AgentAppSessionScope) -> CompositorSessionSnapshot | None:
-        self.loaded_scopes.append(scope)
-        return self.loaded
-
-    def load_active_session(self, scope: AgentAppSessionScope) -> StoredAgentAppSession | None:
-        self.loaded_scopes.append(scope)
+    def load_or_create(self, scope: AgentAppSessionScope) -> StoredAgentAppSession:
+        self.resolved_scopes.append(scope)
         if self._loaded_session is not None:
             return self._loaded_session
-        if self.loaded is None:
-            return None
-        return StoredAgentAppSession(scope=scope, session_snapshot=self.loaded, backend_run_id=None)
-
-    def list_active_sessions_for_conversation(
-        self, *, tenant_id: str, app_id: str, conversation_id: str
-    ) -> list[StoredAgentAppSession]:
-        assert tenant_id == "tenant-1"
-        assert app_id == "app-1"
-        assert conversation_id == "conv-1"
-        return list(self._listed_sessions)
+        return StoredAgentAppSession(
+            scope=scope,
+            binding_id=self.binding_id,
+            workspace_id=self.workspace_id,
+            backend_binding_ref=self.backend_binding_ref,
+            session_snapshot=self.loaded,
+        )
 
     def save_active_snapshot(
         self,
         *,
-        scope,
-        backend_run_id,
-        snapshot,
-        runtime_layer_specs,
-        pending_form_id=None,
-        pending_tool_call_id=None,
+        scope: AgentAppSessionScope,
+        binding_id: str,
+        snapshot: CompositorSessionSnapshot | None,
+        pending_form_id: str | None = None,
+        pending_tool_call_id: str | None = None,
     ) -> None:
-        self.saved.append(
-            (scope, backend_run_id, snapshot, list(runtime_layer_specs), pending_form_id, pending_tool_call_id)
-        )
-
-    def mark_cleaned(self, *, scope: AgentAppSessionScope, backend_run_id: str | None = None) -> None:
-        self.cleaned.append((scope, backend_run_id))
+        self.saved.append((scope, binding_id, snapshot, pending_form_id, pending_tool_call_id))
 
 
 class _MonotonicClock:
@@ -529,18 +478,18 @@ def _runner(
     )
 
 
-def _run(runner: AgentAppRunner, qm: _FakeQueueManager, *, agent_runtime_exit_intent: str = "suspend") -> None:
+def _run(runner: AgentAppRunner, qm: _FakeQueueManager) -> None:
     runner.run(
         dify_context=_dify_ctx(),
         agent_id="agent-1",
         agent_config_snapshot_id="snap-1",
         agent_soul=_soul(),
+        home_snapshot_id="home-1",
         conversation_id="conv-1",
         query="hello",
         message_id="msg-1",
         model_name="gpt-4o-mini",
         queue_manager=qm,  # type: ignore[arg-type]
-        agent_runtime_exit_intent=agent_runtime_exit_intent,  # type: ignore[arg-type]
     )
 
 
@@ -578,134 +527,27 @@ def test_successful_turn_publishes_chunk_and_message_end_and_saves_session():
     assert _saved_user_query(qm) == "hello"
     # The conversation session snapshot is persisted for multi-turn continuity.
     assert store.saved
-    saved_scope, saved_run_id, saved_snapshot, saved_specs, pending_form_id, pending_tool_call_id = store.saved[0]
+    saved_scope, saved_binding_id, saved_snapshot, pending_form_id, pending_tool_call_id = store.saved[0]
     assert saved_scope.conversation_id == "conv-1"
     assert saved_scope.agent_config_snapshot_id == "snap-1"
-    assert saved_run_id == "fake-run-1"
+    assert saved_binding_id == "binding-1"
     assert saved_snapshot is not None
-    assert saved_specs
     # A successful turn carries no ask_human pause correlation.
     assert pending_form_id is None
     assert pending_tool_call_id is None
-    assert store.cleaned == []
 
 
-def test_successful_turn_enqueues_cleanup_for_superseded_sessions_after_saving_snapshot(monkeypatch):
-    superseded = StoredAgentAppSession(
-        scope=AgentAppSessionScope(
-            tenant_id="tenant-1",
-            app_id="app-1",
-            conversation_id="conv-1",
-            agent_id="agent-2",
-            agent_config_snapshot_id="snap-2",
-        ),
-        session_snapshot=CompositorSessionSnapshot(layers=[]),
-        backend_run_id="run-old",
-        runtime_layer_specs=[RuntimeLayerSpec(name="history", type="pydantic_ai.history")],
-    )
-    current_scope_session = StoredAgentAppSession(
-        scope=AgentAppSessionScope(
-            tenant_id="tenant-1",
-            app_id="app-1",
-            conversation_id="conv-1",
-            agent_id="agent-1",
-            agent_config_snapshot_id="snap-1",
-        ),
-        session_snapshot=CompositorSessionSnapshot(layers=[]),
-        backend_run_id="run-current",
-        runtime_layer_specs=[RuntimeLayerSpec(name="history", type="pydantic_ai.history")],
-    )
-    store = _FakeSessionStore(listed_sessions=[current_scope_session, superseded])
+def test_turn_uses_resolved_backend_binding_before_backend_invocation() -> None:
     client = FakeAgentBackendRunClient()
-    qm = _FakeQueueManager()
-    cleanup_delay = MagicMock()
-    monkeypatch.setattr(app_runner_module.cleanup_conversation_agent_runtime_session, "delay", cleanup_delay)
+    store = _FakeSessionStore(binding_id="binding-2", backend_binding_ref="backend-binding-2")
 
-    _run(_runner(client, store), qm)
-
-    assert store.saved
-    cleanup_delay.assert_called_once()
-    payload = cleanup_delay.call_args.args[0]
-    assert payload["metadata"]["conversation_id"] == "conv-1"
-    assert payload["metadata"]["agent_id"] == "agent-2"
-    assert payload["metadata"]["previous_agent_backend_run_id"] == "run-old"
-    assert payload["idempotency_key"] == "tenant-1:app-1:conv-1:agent-2:snap-2:superseded-session-cleanup:run-old"
-
-
-def test_superseded_session_cleanup_enqueue_failure_does_not_fail_turn(monkeypatch):
-    superseded = StoredAgentAppSession(
-        scope=AgentAppSessionScope(
-            tenant_id="tenant-1",
-            app_id="app-1",
-            conversation_id="conv-1",
-            agent_id="agent-2",
-            agent_config_snapshot_id="snap-2",
-        ),
-        session_snapshot=CompositorSessionSnapshot(layers=[]),
-        backend_run_id="run-old",
-        runtime_layer_specs=[RuntimeLayerSpec(name="history", type="pydantic_ai.history")],
-    )
-    store = _FakeSessionStore(listed_sessions=[superseded])
-    client = FakeAgentBackendRunClient()
-    qm = _FakeQueueManager()
-    cleanup_delay = MagicMock(side_effect=RuntimeError("queue down"))
-    monkeypatch.setattr(app_runner_module.cleanup_conversation_agent_runtime_session, "delay", cleanup_delay)
-
-    _run(_runner(client, store), qm)
-
-    cleanup_delay.assert_called_once()
-    end_events = [e for e in qm.events if isinstance(e, QueueMessageEndEvent)]
-    assert len(end_events) == 1
-    assert end_events[0].llm_result.message.content == "hello agent"
-
-
-def test_delete_on_exit_turn_marks_session_cleaned_without_saving_snapshot():
-    client = _StreamingRecordingFakeAgentBackendRunClient()
-    store = _FakeSessionStore()
-    qm = _FakeQueueManager()
-
-    _run(_runner(client, store), qm, agent_runtime_exit_intent="delete")
+    _run(_runner(client, store), _FakeQueueManager())
 
     assert client.request is not None
-    assert client.request.on_exit.default.value == "delete"
-    assert store.saved == []
-    assert len(store.cleaned) == 1
-    cleaned_scope, cleaned_run_id = store.cleaned[0]
-    assert cleaned_scope.conversation_id == "conv-1"
-    assert cleaned_scope.agent_config_snapshot_id == "snap-1"
-    assert cleaned_run_id == "fake-run-1"
-    end_events = [e for e in qm.events if isinstance(e, QueueMessageEndEvent)]
-    assert len(end_events) == 1
-    assert end_events[0].llm_result.message.content == "hello agent"
-
-
-def test_delete_on_exit_turn_swallows_cleanup_failure_after_success():
-    client = _StreamingRecordingFakeAgentBackendRunClient()
-    store = _FakeSessionStore()
-    store.mark_cleaned = MagicMock(side_effect=RuntimeError("cleanup failed"))  # type: ignore[method-assign]
-    qm = _FakeQueueManager()
-
-    _run(_runner(client, store), qm, agent_runtime_exit_intent="delete")
-
-    assert store.saved == []
-    store.mark_cleaned.assert_called_once()
-    end_events = [e for e in qm.events if isinstance(e, QueueMessageEndEvent)]
-    assert len(end_events) == 1
-
-
-def test_delete_on_exit_turn_marks_session_cleaned_when_publish_fails():
-    client = _StreamingRecordingFakeAgentBackendRunClient()
-    store = _FakeSessionStore()
-    store.mark_cleaned = MagicMock(side_effect=RuntimeError("cleanup failed"))  # type: ignore[method-assign]
-    qm = _FakeQueueManager()
-    runner = _runner(client, store)
-    runner._publish_terminal_answer = MagicMock(side_effect=RuntimeError("publish failed"))
-
-    with pytest.raises(RuntimeError, match="publish failed"):
-        _run(runner, qm, agent_runtime_exit_intent="delete")
-
-    assert store.saved == []
-    store.mark_cleaned.assert_called_once()
+    layers = {layer["name"]: layer for layer in client.request.model_dump(mode="json")["composition"]["layers"]}
+    assert layers["runtime"]["config"]["backend_binding_ref"] == "backend-binding-2"
+    assert store.saved[0][1] == "binding-2"
+    assert len(store.resolved_scopes) == 1
 
 
 def test_successful_turn_routes_stream_text_to_agent_message_and_uses_terminal_output(monkeypatch):
@@ -1254,6 +1096,7 @@ def test_debug_session_scope_can_reuse_conversation_across_config_snapshots():
         agent_id="agent-1",
         agent_config_snapshot_id="snap-new",
         agent_soul=_soul(),
+        home_snapshot_id="home-1",
         conversation_id="conv-1",
         query="hello",
         message_id="msg-1",
@@ -1264,8 +1107,8 @@ def test_debug_session_scope_can_reuse_conversation_across_config_snapshots():
 
     assert client.request is not None
     assert client.request.session_snapshot is prior
-    assert store.loaded_scopes[0].agent_config_snapshot_id is None
-    assert store.saved[0][0].agent_config_snapshot_id is None
+    assert store.resolved_scopes[0].agent_config_snapshot_id == "snap-new"
+    assert store.saved[0][0].agent_config_snapshot_id == "snap-new"
 
 
 def test_failed_run_raises_agent_backend_error():
@@ -1358,24 +1201,8 @@ def test_ask_human_pauses_turn_creates_form_and_persists_correlation():
     assert _saved_user_query(qm) == "hello"
     # The pause correlation is persisted so a form submission can resume the run.
     assert store.saved
-    assert store.saved[0][4] == "form-1"
-    assert store.saved[0][5] == "fake-ask-human-1"
-
-
-def test_delete_on_exit_deferred_tool_marks_session_cleaned_and_raises_error():
-    client = FakeAgentBackendRunClient(scenario=FakeAgentBackendScenario.PAUSED)
-    store = _FakeSessionStore()
-    store.mark_cleaned = MagicMock(side_effect=RuntimeError("cleanup failed"))  # type: ignore[method-assign]
-    qm = _FakeQueueManager()
-    runner = _runner(client, store)
-    runner._pause_for_ask_human = MagicMock()
-
-    with pytest.raises(AgentBackendError, match="finalization cannot pause for human input"):
-        _run(runner, qm, agent_runtime_exit_intent="delete")
-
-    runner._pause_for_ask_human.assert_not_called()
-    assert store.saved == []
-    store.mark_cleaned.assert_called_once()
+    assert store.saved[0][3] == "form-1"
+    assert store.saved[0][4] == "fake-ask-human-1"
 
 
 def test_submitted_form_resumes_turn_with_deferred_tool_results(monkeypatch):
@@ -1389,9 +1216,12 @@ def test_submitted_form_resumes_turn_with_deferred_tool_results(monkeypatch):
             conversation_id="conv-1",
             agent_id="agent-1",
             agent_config_snapshot_id="snap-1",
+            home_snapshot_id="home-1",
         ),
+        binding_id="binding-1",
+        workspace_id="workspace-1",
+        backend_binding_ref="backend-binding-1",
         session_snapshot=snapshot,
-        backend_run_id="run-0",
         pending_form_id="form-1",
         pending_tool_call_id="call-1",
     )

api/tests/unit_tests/core/app/apps/agent_app/test_resolve_agent.py

@@ -9,13 +9,18 @@ from __future__ import annotations
 
 from types import SimpleNamespace
 from typing import Any
+from unittest.mock import MagicMock
 
 import pytest
 
 from core.app.apps.agent_app.app_generator import AgentAppGenerator, AgentAppGeneratorError, AgentAppNotPublishedError
 from core.app.entities.app_invoke_entities import InvokeFrom
-from models.agent import AgentConfigDraft, AgentConfigDraftType, AgentScope, AgentSource
+from models.agent import AgentConfigDraft, AgentConfigDraftType, AgentConfigVersionKind, AgentScope, AgentSource
 from models.agent_config_entities import AgentSoulConfig
+from services.agent.workspace_service import (
+    AgentWorkspaceBindingGenerationMismatchError,
+    AgentWorkspaceService,
+)
 
 _SOUL_DICT = {
     "model": {
@@ -34,8 +39,10 @@ class _FakeScalarSession:
         self._values = list(values)
         self.added: list[Any] = []
         self.flush_count = 0
+        self.scalar_statements: list[Any] = []
 
-    def scalar(self, _stmt: Any) -> Any:
+    def scalar(self, stmt: Any) -> Any:
+        self.scalar_statements.append(stmt)
         return self._values.pop(0) if self._values else None
 
     def add(self, value: Any) -> None:
@@ -46,7 +53,7 @@ class _FakeScalarSession:
 
 
 def _snapshot() -> SimpleNamespace:
-    return SimpleNamespace(id="snap-1", config_snapshot_dict=_SOUL_DICT)
+    return SimpleNamespace(id="snap-1", home_snapshot_id="home-1", config_snapshot_dict=_SOUL_DICT)
 
 
 class TestResolveAgentById:
@@ -102,6 +109,7 @@ class TestResolveDebugDraft:
             tenant_id="t1",
             agent=agent,
             draft_type=None,
+            draft_id=None,
             account_id=None,
             session=session,
         )
@@ -127,10 +135,12 @@ class TestResolveDebugDraft:
             account_id=None,
             draft_owner_key="",
             base_snapshot_id="snap-1",
+            home_snapshot_id="home-1",
             config_snapshot=AgentSoulConfig.model_validate({"prompt": {"system_prompt": "old"}}),
         )
         active_snapshot = SimpleNamespace(
             id="snap-2",
+            home_snapshot_id="home-2",
             config_snapshot_dict={"prompt": {"system_prompt": "new"}},
         )
         session = _FakeScalarSession([draft, active_snapshot])
@@ -146,6 +156,7 @@ class TestResolveDebugDraft:
         assert resolved is draft
         assert resolved.id == "draft-1"
         assert resolved.base_snapshot_id == "snap-2"
+        assert resolved.home_snapshot_id == "home-2"
         assert resolved.config_snapshot_dict["prompt"]["system_prompt"] == "new"
         assert session.flush_count == 1
 
@@ -165,6 +176,7 @@ class TestResolveDebugDraft:
             account_id="account-1",
             draft_owner_key="account-1",
             base_snapshot_id="snap-1",
+            home_snapshot_id="home-build",
             config_snapshot=AgentSoulConfig.model_validate({"prompt": {"system_prompt": "build edit"}}),
         )
         session = _FakeScalarSession([draft])
@@ -182,6 +194,48 @@ class TestResolveDebugDraft:
         assert resolved.config_snapshot_dict["prompt"]["system_prompt"] == "build edit"
         assert session.flush_count == 0
 
+    def test_build_draft_uses_exact_draft_id(self):
+        agent = SimpleNamespace(
+            id="agent-1",
+            scope=AgentScope.WORKFLOW_ONLY,
+            active_config_snapshot_id="snap-2",
+            created_by="creator-1",
+            updated_by="updater-1",
+        )
+        draft = AgentConfigDraft(
+            id="exact-build-draft",
+            tenant_id="t1",
+            agent_id="agent-1",
+            draft_type=AgentConfigDraftType.DEBUG_BUILD,
+            account_id="account-1",
+            draft_owner_key="account-1",
+            base_snapshot_id="snap-1",
+            home_snapshot_id="home-build",
+            config_snapshot=AgentSoulConfig(),
+        )
+        session = _FakeScalarSession([draft])
+        statements: list[Any] = []
+        scalar = session.scalar
+
+        def capture_scalar(statement: Any) -> Any:
+            statements.append(statement)
+            return scalar(statement)
+
+        session.scalar = capture_scalar  # type: ignore[method-assign]
+
+        resolved = AgentAppGenerator._resolve_debug_draft(
+            tenant_id="t1",
+            agent=agent,
+            draft_type=AgentConfigDraftType.DEBUG_BUILD.value,
+            draft_id="exact-build-draft",
+            account_id="account-1",
+            session=session,
+        )
+
+        assert resolved is draft
+        assert "agent_config_drafts.id =" in str(statements[0])
+        assert "exact-build-draft" in statements[0].compile().params.values()
+
 
 class TestResolveAgent:
     def test_success_chains_to_resolve_by_id(self):
@@ -235,6 +289,126 @@ class TestResolveAgent:
         assert config_version_kind == "snapshot"
         assert soul.prompt.system_prompt == "You are Iris."
 
+    def test_existing_conversation_resolves_binding_snapshot_instead_of_latest_active_snapshot(
+        self, monkeypatch: pytest.MonkeyPatch
+    ):
+        bound_agent = SimpleNamespace(
+            id="agent-1",
+            source=AgentSource.AGENT_APP,
+            active_config_snapshot_id="snap-2",
+            active_config_is_published=True,
+        )
+        inner_agent = SimpleNamespace(id="agent-1")
+        pinned_snapshot = SimpleNamespace(
+            id="snap-1",
+            home_snapshot_id="home-1",
+            config_snapshot_dict=_SOUL_DICT,
+        )
+        conversation = SimpleNamespace(id="conversation-1", agent_workspace_binding_id="binding-1")
+        binding = SimpleNamespace(
+            id="binding-1",
+            agent_id="agent-1",
+            base_home_snapshot_id="home-1",
+            agent_config_version_id="snap-1",
+            agent_config_version_kind=AgentConfigVersionKind.SNAPSHOT,
+        )
+        get_active_binding = MagicMock(return_value=binding)
+        validate_generation = MagicMock()
+        monkeypatch.setattr(AgentWorkspaceService, "get_active_binding", get_active_binding)
+        monkeypatch.setattr(AgentWorkspaceService, "validate_binding_generation", validate_generation)
+        session = _FakeScalarSession([bound_agent, inner_agent, pinned_snapshot])
+        app_model = SimpleNamespace(id="app-1", tenant_id="t1")
+
+        _, config_id, config_version_kind, soul = AgentAppGenerator()._resolve_agent(
+            app_model,
+            invoke_from=InvokeFrom.WEB_APP,
+            draft_type=None,
+            user=SimpleNamespace(id="user-1"),
+            session=session,
+            conversation=conversation,
+        )  # type: ignore[arg-type]
+
+        assert config_id == "snap-1"
+        assert config_version_kind == "snapshot"
+        assert soul.prompt.system_prompt == "You are Iris."
+        assert get_active_binding.call_args.kwargs["binding_id"] == "binding-1"
+        validate_generation.assert_called_once_with(
+            binding,
+            base_home_snapshot_id="home-1",
+            agent_config_version_id="snap-1",
+            agent_config_version_kind=AgentConfigVersionKind.SNAPSHOT,
+        )
+
+    def test_existing_conversation_rejects_unavailable_binding(self, monkeypatch: pytest.MonkeyPatch):
+        bound_agent = SimpleNamespace(
+            id="agent-1",
+            source=AgentSource.AGENT_APP,
+            active_config_snapshot_id="snap-active",
+            active_config_is_published=True,
+        )
+        conversation = SimpleNamespace(id="conversation-1", agent_workspace_binding_id="binding-missing")
+        monkeypatch.setattr(AgentWorkspaceService, "get_active_binding", MagicMock(return_value=None))
+
+        with pytest.raises(AgentAppGeneratorError, match="Conversation participant Binding is unavailable"):
+            AgentAppGenerator()._resolve_agent(
+                SimpleNamespace(id="app-1", tenant_id="t1"),
+                invoke_from=InvokeFrom.WEB_APP,
+                draft_type=None,
+                user=SimpleNamespace(id="user-1"),
+                session=_FakeScalarSession([bound_agent]),
+                conversation=conversation,
+            )  # type: ignore[arg-type]
+
+    @pytest.mark.parametrize(
+        ("binding_home_id", "binding_version_kind", "snapshot_home_id"),
+        [
+            ("home-binding", AgentConfigVersionKind.SNAPSHOT, "home-other"),
+            ("home-pinned", AgentConfigVersionKind.DRAFT, "home-pinned"),
+        ],
+    )
+    def test_existing_conversation_generation_mismatch_does_not_fallback_to_active_snapshot(
+        self,
+        monkeypatch: pytest.MonkeyPatch,
+        binding_home_id: str,
+        binding_version_kind: AgentConfigVersionKind,
+        snapshot_home_id: str,
+    ):
+        bound_agent = SimpleNamespace(
+            id="agent-1",
+            source=AgentSource.AGENT_APP,
+            active_config_snapshot_id="snap-active",
+            active_config_is_published=True,
+        )
+        conversation = SimpleNamespace(id="conversation-1", agent_workspace_binding_id="binding-1")
+        binding = SimpleNamespace(
+            id="binding-1",
+            agent_id="agent-1",
+            base_home_snapshot_id=binding_home_id,
+            agent_config_version_id="snap-pinned",
+            agent_config_version_kind=binding_version_kind,
+        )
+        pinned_snapshot = SimpleNamespace(
+            id="snap-pinned",
+            home_snapshot_id=snapshot_home_id,
+            config_snapshot_dict=_SOUL_DICT,
+        )
+        monkeypatch.setattr(AgentWorkspaceService, "get_active_binding", MagicMock(return_value=binding))
+        session = _FakeScalarSession([bound_agent, SimpleNamespace(id="agent-1"), pinned_snapshot])
+
+        with pytest.raises(AgentWorkspaceBindingGenerationMismatchError):
+            AgentAppGenerator()._resolve_agent(
+                SimpleNamespace(id="app-1", tenant_id="t1"),
+                invoke_from=InvokeFrom.WEB_APP,
+                draft_type=None,
+                user=SimpleNamespace(id="user-1"),
+                session=session,
+                conversation=conversation,
+            )  # type: ignore[arg-type]
+
+        snapshot_query_params = session.scalar_statements[-1].compile().params.values()
+        assert "snap-pinned" in snapshot_query_params
+        assert "snap-active" not in snapshot_query_params
+
     def test_unpublished_imported_agent_is_not_available_to_public_runtime(self):
         bound_agent = SimpleNamespace(
             id="agent-1",

api/tests/unit_tests/core/app/apps/agent_app/test_runtime_request_builder.py

@@ -44,6 +44,7 @@ class TestBuildForAgentApp:
             AgentBackendAgentAppRunInput(
                 model=AgentBackendModelConfig(plugin_id="langgenius/openai", model_provider="openai", model="gpt-test"),
                 execution_context=_exec_ctx(),
+                backend_binding_ref="binding-ref-1",
                 user_prompt="hello",
                 agent_soul_prompt="You are Iris.",
             )
@@ -65,6 +66,7 @@ class TestBuildForAgentApp:
             AgentBackendAgentAppRunInput(
                 model=AgentBackendModelConfig(plugin_id="p/q", model_provider="openai", model="m"),
                 execution_context=_exec_ctx(),
+                backend_binding_ref="binding-ref-1",
                 user_prompt="   ",
             )
 
@@ -73,6 +75,7 @@ class TestBuildForAgentApp:
             AgentBackendAgentAppRunInput(
                 model=AgentBackendModelConfig(plugin_id="langgenius/openai", model_provider="openai", model="gpt-test"),
                 execution_context=_exec_ctx(),
+                backend_binding_ref="binding-ref-1",
                 user_prompt="hi",
             )
         )
@@ -142,7 +145,6 @@ def _ctx(
     *,
     query: str = "hello",
     agent_config_version_kind: str = "snapshot",
-    suspend_on_exit: bool = True,
 ) -> AgentAppRuntimeBuildContext:
     dify_context = SimpleNamespace(
         tenant_id="tenant-1",
@@ -159,8 +161,9 @@ def _ctx(
         conversation_id="conv-1",
         user_query=query,
         idempotency_key="msg-1",
+        binding_id="binding-1",
+        backend_binding_ref="binding-ref-1",
         agent_config_version_kind=agent_config_version_kind,  # type: ignore[arg-type]
-        suspend_on_exit=suspend_on_exit,
     )
 
 
@@ -191,6 +194,7 @@ class TestAgentAppRuntimeRequestBuilder:
             "agent_soul_prompt",
             "agent_app_user_prompt",
             "execution_context",
+            "runtime",
             DIFY_SHELL_LAYER_ID,
             DIFY_CONFIG_LAYER_ID,
             "history",
@@ -243,16 +247,6 @@ class TestAgentAppRuntimeRequestBuilder:
         assert execution_context.config.agent_config_version_kind == "draft"
         assert config_layer.config.config_version.kind == "draft"
 
-    def test_build_uses_delete_on_exit_when_requested(self):
-        builder = AgentAppRuntimeRequestBuilder(
-            credentials_provider=_FakeCredentialsProvider(),
-            dify_tools_builder=_NoToolsBuilder(),  # type: ignore[arg-type]
-        )
-
-        result = builder.build(_ctx(_soul_with_model(), suspend_on_exit=False))
-
-        assert result.request.on_exit.default.value == "delete"
-
     def test_build_includes_plugin_tools_layer_returned_by_injected_builder_for_draft(self):
         soul = _soul_with_model()
         soul.tools.dify_tools = [
@@ -410,7 +404,7 @@ class TestAgentAppRuntimeRequestBuilder:
         ]
         assert shell_config["cli_tools"][0]["install_commands"] == ["apt-get install -y ripgrep"]
         assert shell_config["env"][0] == {"name": "PROJECT_NAME", "value": "demo"}
-        assert shell_config["sandbox"] == {"provider": "independent", "config": {"cpu": 2}}
+        assert "sandbox" not in shell_config
         assert result.metadata["agent_tools"] == {
             "dify_tool_count": 0,
             "dify_tool_names": [],
@@ -461,7 +455,7 @@ class TestAgentAppConfigLayer:
         assert config.config.mentioned_file_names == []
         # shell enters first; config uses that shell to materialize mentioned targets.
         names = [layer.name for layer in result.request.composition.layers]
-        assert names.index(DIFY_SHELL_LAYER_ID) == names.index("execution_context") + 1
+        assert names.index(DIFY_SHELL_LAYER_ID) == names.index("execution_context") + 2
         assert names.index(DIFY_CONFIG_LAYER_ID) == names.index(DIFY_SHELL_LAYER_ID) + 1
 
     def test_config_layer_present_when_agent_soul_has_no_config_assets(self, monkeypatch: pytest.MonkeyPatch):
@@ -484,7 +478,10 @@ class TestAgentAppConfigLayer:
             "mentioned_skill_names": [],
             "mentioned_file_names": [],
         }
-        assert layers[DIFY_SHELL_LAYER_ID].deps == {"execution_context": "execution_context"}
+        assert layers[DIFY_SHELL_LAYER_ID].deps == {
+            "execution_context": "execution_context",
+            "runtime": "runtime",
+        }
         assert layers[DIFY_SHELL_LAYER_ID].config.agent_stub_drive_ref is None
 
     def test_config_layer_for_build_draft_marks_config_writable(self):

api/tests/unit_tests/core/app/apps/agent_app/test_session_store.py

@@ -1,348 +1,123 @@
-"""Unit tests for the conversation-keyed Agent App session store.
-
-Exercises the real ORM round-trip against the project's in-memory SQLite engine
-(per-test create/drop of the unified ``agent_runtime_sessions`` table), so the
-conversation owner path is verified without Postgres.
-"""
-
-from __future__ import annotations
-
-from collections.abc import Generator
+from types import SimpleNamespace
+from unittest.mock import MagicMock
 
 import pytest
 from agenton.compositor import CompositorSessionSnapshot
-from agenton.compositor.schemas import LayerSessionSnapshot
-from agenton.layers.base import LifecycleState
-from dify_agent.protocol import RuntimeLayerSpec
-from sqlalchemy import delete
 
-from core.app.apps.agent_app.session_store import AgentAppRuntimeSessionStore, AgentAppSessionScope
-from core.db.session_factory import session_factory
-from models.agent import AgentRuntimeSession, AgentRuntimeSessionOwnerType, AgentRuntimeSessionStatus
+from core.app.apps.agent_app.session_store import AgentAppSessionScope, AgentAppWorkspaceStore
+from models.agent import (
+    AgentConfigVersionKind,
+    AgentWorkspaceOwnerType,
+)
+from services.agent.workspace_service import AgentWorkspaceNotFoundError, AgentWorkspaceService
 
 
 def _scope(
-    conversation_id: str = "conv-1", agent_id: str = "agent-1", agent_config_snapshot_id: str | None = "snap-1"
+    *,
+    kind: AgentConfigVersionKind = AgentConfigVersionKind.SNAPSHOT,
+    build_draft_id: str | None = None,
 ) -> AgentAppSessionScope:
     return AgentAppSessionScope(
         tenant_id="tenant-1",
         app_id="app-1",
-        conversation_id=conversation_id,
-        agent_id=agent_id,
-        agent_config_snapshot_id=agent_config_snapshot_id,
+        conversation_id="conversation-1",
+        agent_id="agent-1",
+        agent_config_snapshot_id="config-1",
+        home_snapshot_id="home-1",
+        agent_config_version_kind=kind,
+        build_draft_id=build_draft_id,
     )
 
 
-def _snapshot(messages: int = 1) -> CompositorSessionSnapshot:
-    return CompositorSessionSnapshot(
-        layers=[
-            LayerSessionSnapshot(
-                name="history",
-                lifecycle_state=LifecycleState.SUSPENDED,
-                runtime_state={"messages": [{"role": "user", "content": f"m{i}"} for i in range(messages)]},
-            )
-        ]
+def _binding() -> SimpleNamespace:
+    return SimpleNamespace(
+        id="binding-1",
+        workspace_id="workspace-1",
+        backend_binding_ref="backend-binding-1",
+        agent_id="agent-1",
+        agent_config_version_id="config-1",
+        agent_config_version_kind=AgentConfigVersionKind.SNAPSHOT,
+        base_home_snapshot_id="home-1",
+        session_snapshot=None,
+        pending_form_id=None,
+        pending_tool_call_id=None,
     )
 
 
-def _runtime_layer_specs() -> list[RuntimeLayerSpec]:
-    return [
-        RuntimeLayerSpec(name="execution_context", type="dify.execution_context", config={"tenant_id": "tenant-1"}),
-        RuntimeLayerSpec(name="history", type="pydantic_ai.history"),
-    ]
+def test_scope_selects_conversation_or_build_draft_workspace_owner() -> None:
+    assert _scope().workspace_owner.owner_type is AgentWorkspaceOwnerType.CONVERSATION
+    build_owner = _scope(
+        kind=AgentConfigVersionKind.BUILD_DRAFT,
+        build_draft_id="build-draft-1",
+    ).workspace_owner
+    assert build_owner.owner_type is AgentWorkspaceOwnerType.BUILD_DRAFT
+    assert build_owner.owner_id == "build-draft-1"
 
 
-@pytest.fixture(autouse=True)
-def _create_table() -> Generator[None, None, None]:
-    engine = session_factory.get_session_maker().kw["bind"]
-    AgentRuntimeSession.__table__.create(bind=engine, checkfirst=True)
-    yield
-    with session_factory.create_session() as session:
-        session.execute(delete(AgentRuntimeSession))
-        session.commit()
-    AgentRuntimeSession.__table__.drop(bind=engine, checkfirst=True)
+def test_load_or_create_persists_new_binding_on_caller(monkeypatch) -> None:
+    caller = SimpleNamespace(agent_workspace_binding_id=None)
+    context = MagicMock()
+    session = context.__enter__.return_value
+    create = MagicMock(return_value=_binding())
+    store = AgentAppWorkspaceStore()
+    monkeypatch.setattr("core.app.apps.agent_app.session_store.session_factory.create_session", lambda: context)
+    monkeypatch.setattr(store, "_load_caller", MagicMock(return_value=caller))
+    monkeypatch.setattr(AgentWorkspaceService, "create_binding", create)
+
+    stored = store.load_or_create(_scope())
+
+    assert stored.binding_id == "binding-1"
+    assert stored.workspace_id == "workspace-1"
+    assert stored.backend_binding_ref == "backend-binding-1"
+    assert caller.agent_workspace_binding_id == "binding-1"
+    assert create.call_args.kwargs["session"] is session
+    session.commit.assert_called_once_with()
 
 
-def test_load_returns_none_when_no_row():
-    assert AgentAppRuntimeSessionStore().load_active_snapshot(_scope()) is None
+def test_load_or_create_uses_exact_caller_binding(monkeypatch) -> None:
+    caller = SimpleNamespace(agent_workspace_binding_id="binding-1")
+    context = MagicMock()
+    context.__enter__.return_value = MagicMock()
+    get_binding = MagicMock(return_value=_binding())
+    create = MagicMock()
+    store = AgentAppWorkspaceStore()
+    monkeypatch.setattr("core.app.apps.agent_app.session_store.session_factory.create_session", lambda: context)
+    monkeypatch.setattr(store, "_load_caller", MagicMock(return_value=caller))
+    monkeypatch.setattr(AgentWorkspaceService, "get_active_binding", get_binding)
+    monkeypatch.setattr(AgentWorkspaceService, "validate_binding_generation", MagicMock())
+    monkeypatch.setattr(AgentWorkspaceService, "create_binding", create)
+
+    stored = store.load_or_create(_scope())
+
+    assert stored.binding_id == "binding-1"
+    assert get_binding.call_args.kwargs["binding_id"] == "binding-1"
+    create.assert_not_called()
 
 
-def test_save_creates_conversation_owned_row_and_round_trips():
-    store = AgentAppRuntimeSessionStore()
-    store.save_active_snapshot(
-        scope=_scope(),
-        backend_run_id="run-1",
-        snapshot=_snapshot(messages=2),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
+def test_normal_conversation_pointer_does_not_create_replacement_binding(monkeypatch) -> None:
+    caller = SimpleNamespace(agent_workspace_binding_id="unavailable-binding")
+    context = MagicMock()
+    get_binding = MagicMock(return_value=None)
+    create = MagicMock()
+    store = AgentAppWorkspaceStore()
+    monkeypatch.setattr("core.app.apps.agent_app.session_store.session_factory.create_session", lambda: context)
+    monkeypatch.setattr(store, "_load_caller", MagicMock(return_value=caller))
+    monkeypatch.setattr(AgentWorkspaceService, "get_active_binding", get_binding)
+    monkeypatch.setattr(AgentWorkspaceService, "create_binding", create)
 
-    loaded = store.load_active_snapshot(_scope())
-    assert loaded is not None
-    assert loaded.layers[0].runtime_state["messages"] == [
-        {"role": "user", "content": "m0"},
-        {"role": "user", "content": "m1"},
-    ]
-    with session_factory.create_session() as session:
-        row = session.query(AgentRuntimeSession).one()
-        assert row.owner_type == AgentRuntimeSessionOwnerType.CONVERSATION
-        assert row.conversation_id == "conv-1"
-        assert row.agent_config_snapshot_id == "snap-1"
-        assert row.workflow_run_id is None  # conversation owner leaves workflow cols NULL
-        assert row.backend_run_id == "run-1"
-        assert "execution_context" in row.composition_layer_specs
-        assert "history" in row.composition_layer_specs
+    with pytest.raises(AgentWorkspaceNotFoundError, match="Caller participant Binding is unavailable"):
+        store.load_or_create(_scope())
+
+    assert get_binding.call_args.kwargs["binding_id"] == "unavailable-binding"
+    create.assert_not_called()
 
 
-def test_save_is_noop_when_snapshot_missing():
-    store = AgentAppRuntimeSessionStore()
-    store.save_active_snapshot(
-        scope=_scope(),
-        backend_run_id="run-x",
-        snapshot=None,
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-    with session_factory.create_session() as session:
-        assert session.query(AgentRuntimeSession).count() == 0
+def test_save_snapshot_targets_binding(monkeypatch) -> None:
+    save = MagicMock()
+    monkeypatch.setattr(AgentWorkspaceService, "save_binding_session_snapshot", save)
+    snapshot = CompositorSessionSnapshot(layers=[])
 
+    AgentAppWorkspaceStore().save_active_snapshot(scope=_scope(), binding_id="binding-1", snapshot=snapshot)
 
-def test_second_turn_updates_same_conversation_row():
-    store = AgentAppRuntimeSessionStore()
-    store.save_active_snapshot(
-        scope=_scope(),
-        backend_run_id="run-1",
-        snapshot=_snapshot(messages=1),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-    store.save_active_snapshot(
-        scope=_scope(),
-        backend_run_id="run-2",
-        snapshot=_snapshot(messages=3),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-    with session_factory.create_session() as session:
-        rows = session.query(AgentRuntimeSession).all()
-        assert len(rows) == 1
-        assert rows[0].backend_run_id == "run-2"
-
-
-def test_debug_scope_with_null_snapshot_id_updates_same_conversation_row():
-    store = AgentAppRuntimeSessionStore()
-    scope = _scope(agent_config_snapshot_id=None)
-    store.save_active_snapshot(
-        scope=scope,
-        backend_run_id="run-1",
-        snapshot=_snapshot(messages=1),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-    store.save_active_snapshot(
-        scope=scope,
-        backend_run_id="run-2",
-        snapshot=_snapshot(messages=3),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-
-    loaded = store.load_active_snapshot(scope)
-
-    assert loaded is not None
-    assert loaded.layers[0].runtime_state["messages"] == [
-        {"role": "user", "content": "m0"},
-        {"role": "user", "content": "m1"},
-        {"role": "user", "content": "m2"},
-    ]
-    with session_factory.create_session() as session:
-        row = session.query(AgentRuntimeSession).one()
-        assert row.agent_config_snapshot_id is None
-        assert row.backend_run_id == "run-2"
-
-
-def test_mark_cleaned_then_load_returns_none_and_save_resurrects():
-    store = AgentAppRuntimeSessionStore()
-    store.save_active_snapshot(
-        scope=_scope(),
-        backend_run_id="run-1",
-        snapshot=_snapshot(),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-    store.mark_cleaned(scope=_scope(), backend_run_id="cleanup-1")
-    assert store.load_active_snapshot(_scope()) is None
-    # Re-entry revives the row.
-    store.save_active_snapshot(
-        scope=_scope(),
-        backend_run_id="run-2",
-        snapshot=_snapshot(messages=2),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-    with session_factory.create_session() as session:
-        row = session.query(AgentRuntimeSession).one()
-        assert row.status == AgentRuntimeSessionStatus.ACTIVE
-        assert row.cleaned_at is None
-        assert row.backend_run_id == "run-2"
-
-
-def test_distinct_conversations_do_not_collide():
-    store = AgentAppRuntimeSessionStore()
-    store.save_active_snapshot(
-        scope=_scope(conversation_id="conv-A"),
-        backend_run_id="a",
-        snapshot=_snapshot(),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-    store.save_active_snapshot(
-        scope=_scope(conversation_id="conv-B"),
-        backend_run_id="b",
-        snapshot=_snapshot(),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-    assert store.load_active_snapshot(_scope(conversation_id="conv-A")) is not None
-    assert store.load_active_snapshot(_scope(conversation_id="conv-B")) is not None
-    with session_factory.create_session() as session:
-        assert session.query(AgentRuntimeSession).count() == 2
-
-
-def test_distinct_agent_config_snapshots_keep_only_latest_active_session():
-    store = AgentAppRuntimeSessionStore()
-    store.save_active_snapshot(
-        scope=_scope(agent_config_snapshot_id="snap-1"),
-        backend_run_id="a",
-        snapshot=_snapshot(),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-    store.save_active_snapshot(
-        scope=_scope(agent_config_snapshot_id="snap-2"),
-        backend_run_id="b",
-        snapshot=_snapshot(messages=2),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-
-    assert store.load_active_snapshot(_scope(agent_config_snapshot_id="snap-1")) is None
-    assert store.load_active_snapshot(_scope(agent_config_snapshot_id="snap-2")) is not None
-    with session_factory.create_session() as session:
-        rows = session.query(AgentRuntimeSession).order_by(AgentRuntimeSession.backend_run_id).all()
-        assert len(rows) == 2
-        assert [row.agent_config_snapshot_id for row in rows] == ["snap-1", "snap-2"]
-        assert [row.status for row in rows] == [AgentRuntimeSessionStatus.CLEANED, AgentRuntimeSessionStatus.ACTIVE]
-
-
-def test_load_active_session_for_conversation_resolves_without_agent_or_config_scope():
-    store = AgentAppRuntimeSessionStore()
-    store.save_active_snapshot(
-        scope=_scope(),
-        backend_run_id="run-1",
-        snapshot=_snapshot(messages=2),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-
-    loaded = store.load_active_session_for_conversation(tenant_id="tenant-1", app_id="app-1", conversation_id="conv-1")
-    assert loaded is not None
-    assert loaded.session_snapshot.layers[0].runtime_state["messages"] == [
-        {"role": "user", "content": "m0"},
-        {"role": "user", "content": "m1"},
-    ]
-    assert [spec.name for spec in loaded.runtime_layer_specs] == ["execution_context", "history"]
-
-
-def test_load_active_session_for_conversation_uses_latest_active_snapshot_after_config_change():
-    store = AgentAppRuntimeSessionStore()
-    store.save_active_snapshot(
-        scope=_scope(agent_config_snapshot_id="snap-1"),
-        backend_run_id="a",
-        snapshot=_snapshot(),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-    store.save_active_snapshot(
-        scope=_scope(agent_config_snapshot_id="snap-2"),
-        backend_run_id="b",
-        snapshot=_snapshot(messages=3),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-
-    loaded = store.load_active_session_for_conversation(tenant_id="tenant-1", app_id="app-1", conversation_id="conv-1")
-
-    assert loaded is not None
-    assert loaded.session_snapshot.layers[0].runtime_state["messages"] == [
-        {"role": "user", "content": "m0"},
-        {"role": "user", "content": "m1"},
-        {"role": "user", "content": "m2"},
-    ]
-
-
-def test_load_active_session_for_conversation_returns_none_when_cleaned_or_absent():
-    store = AgentAppRuntimeSessionStore()
-    assert (
-        store.load_active_session_for_conversation(tenant_id="tenant-1", app_id="app-1", conversation_id="conv-1")
-        is None
-    )
-
-    store.save_active_snapshot(
-        scope=_scope(),
-        backend_run_id="run-1",
-        snapshot=_snapshot(),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-    store.mark_cleaned(scope=_scope(), backend_run_id="cleanup-1")
-    assert (
-        store.load_active_session_for_conversation(tenant_id="tenant-1", app_id="app-1", conversation_id="conv-1")
-        is None
-    )
-
-
-def test_load_active_session_for_conversation_isolates_other_conversations():
-    store = AgentAppRuntimeSessionStore()
-    store.save_active_snapshot(
-        scope=_scope(conversation_id="conv-A"),
-        backend_run_id="a",
-        snapshot=_snapshot(),
-        runtime_layer_specs=_runtime_layer_specs(),
-    )
-
-    assert (
-        store.load_active_session_for_conversation(tenant_id="tenant-1", app_id="app-1", conversation_id="conv-B")
-        is None
-    )
-    assert (
-        store.load_active_session_for_conversation(tenant_id="tenant-1", app_id="app-1", conversation_id="conv-A")
-        is not None
-    )
-
-
-def test_list_active_sessions_for_conversation_returns_all_active_rows():
-    store = AgentAppRuntimeSessionStore()
-    with session_factory.create_session() as session:
-        session.add(
-            AgentRuntimeSession(
-                tenant_id="tenant-1",
-                app_id="app-1",
-                owner_type=AgentRuntimeSessionOwnerType.CONVERSATION,
-                agent_id="agent-1",
-                agent_config_snapshot_id="snap-1",
-                conversation_id="conv-1",
-                backend_run_id="run-1",
-                session_snapshot=_snapshot(messages=1).model_dump_json(),
-                composition_layer_specs='[{"name":"execution_context","type":"dify.execution_context","deps":{},"metadata":{},"config":{"tenant_id":"tenant-1"}},{"name":"history","type":"pydantic_ai.history","deps":{},"metadata":{},"config":null}]',
-                status=AgentRuntimeSessionStatus.ACTIVE,
-            )
-        )
-        session.add(
-            AgentRuntimeSession(
-                tenant_id="tenant-1",
-                app_id="app-1",
-                owner_type=AgentRuntimeSessionOwnerType.CONVERSATION,
-                agent_id="agent-2",
-                agent_config_snapshot_id="snap-2",
-                conversation_id="conv-1",
-                backend_run_id="run-2",
-                session_snapshot=_snapshot(messages=2).model_dump_json(),
-                composition_layer_specs='[{"name":"execution_context","type":"dify.execution_context","deps":{},"metadata":{},"config":{"tenant_id":"tenant-1"}},{"name":"history","type":"pydantic_ai.history","deps":{},"metadata":{},"config":null}]',
-                status=AgentRuntimeSessionStatus.ACTIVE,
-            )
-        )
-        session.commit()
-
-    loaded = store.list_active_sessions_for_conversation(
-        tenant_id="tenant-1",
-        app_id="app-1",
-        conversation_id="conv-1",
-    )
-
-    assert [session.scope.agent_id for session in loaded] == ["agent-2", "agent-1"]
-    assert all(session.scope.conversation_id == "conv-1" for session in loaded)
+    assert save.call_args.kwargs["binding_id"] == "binding-1"
+    assert save.call_args.kwargs["session_snapshot"] == snapshot.model_dump_json()

api/tests/unit_tests/core/app/workflow/layers/test_persistence.py

@@ -34,6 +34,7 @@ def test_update_node_execution_prefers_event_finished_at(monkeypatch: pytest.Mon
     node_execution = Mock()
     node_execution.id = "node-exec-1"
     node_execution.created_at = datetime(2024, 1, 1, 0, 0, 0, tzinfo=UTC).replace(tzinfo=None)
+    node_execution.process_data = None
     node_execution.update_from_mapping = Mock()
 
     layer._node_snapshots[node_execution.id] = _NodeRuntimeSnapshot(
@@ -66,6 +67,7 @@ def test_update_node_execution_projects_start_outputs() -> None:
     node_execution.id = "node-exec-2"
     node_execution.node_type = BuiltinNodeTypes.START
     node_execution.created_at = datetime(2024, 1, 1, 0, 0, 0, tzinfo=UTC).replace(tzinfo=None)
+    node_execution.process_data = None
     node_execution.update_from_mapping = Mock()
 
     layer._node_snapshots[node_execution.id] = _NodeRuntimeSnapshot(

api/tests/unit_tests/core/app/workflow/test_persistence_layer.py

@@ -39,11 +39,15 @@ from graphon.runtime import GraphRuntimeState, ReadOnlyGraphRuntimeStateWrapper,
 class _RepoRecorder:
     def __init__(self) -> None:
         self.saved: list[object] = []
+        self.synchronously_saved: list[object] = []
         self.saved_exec_data: list[object] = []
 
     def save(self, entity):
         self.saved.append(entity)
 
+    def save_synchronously(self, entity):
+        self.synchronously_saved.append(entity)
+
     def save_execution_data(self, entity):
         self.saved_exec_data.append(entity)
 
@@ -331,6 +335,24 @@ class TestWorkflowPersistenceLayer:
         layer._handle_node_retry(retry_event)
         assert node_repo.saved_exec_data
 
+    def test_agent_v2_caller_row_is_saved_synchronously_before_node_run(self):
+        layer, _, node_repo, _ = _make_layer()
+        layer._handle_graph_run_started()
+
+        layer._handle_node_started(
+            NodeRunStartedEvent(
+                id="agent-exec",
+                node_id="agent-node",
+                node_type=BuiltinNodeTypes.AGENT,
+                node_version="2",
+                node_title="Agent",
+                start_at=_naive_utc_now(),
+            )
+        )
+
+        assert [execution.id for execution in node_repo.synchronously_saved] == ["agent-exec"]
+        assert node_repo.saved == []
+
     def test_retry_history_is_preserved_after_node_succeeds(self):
         layer, _, node_repo, _ = _make_layer()
         layer._handle_graph_run_started()
@@ -501,7 +523,12 @@ class TestWorkflowPersistenceLayer:
         domain_execution = layer._node_execution_cache["exec"]
         domain_execution.inputs = {"old": True}
 
-        result = NodeRunResult(inputs={"new": True}, outputs={"out": 1}, process_data={"p": 1}, metadata={})
+        result = NodeRunResult(
+            inputs={"new": True},
+            outputs={"out": 1},
+            process_data={"p": 1, "workflow_agent_binding_id": "workflow-binding-1"},
+            metadata={},
+        )
         pause_event = NodeRunPauseRequestedEvent(
             id="exec",
             node_id="node",
@@ -513,6 +540,38 @@ class TestWorkflowPersistenceLayer:
 
         assert domain_execution.status == WorkflowNodeExecutionStatus.PAUSED
         assert domain_execution.inputs == {"old": True}
+        assert domain_execution.process_data == {"workflow_agent_binding_id": "workflow-binding-1"}
+
+    def test_handle_node_retry_preserves_workflow_agent_binding_identity(self):
+        layer, _, _, _ = _make_layer()
+        layer._handle_graph_run_started()
+        started_at = _naive_utc_now()
+        layer._handle_node_started(
+            NodeRunStartedEvent(
+                id="exec",
+                node_id="node",
+                node_type=BuiltinNodeTypes.AGENT,
+                node_title="Agent",
+                start_at=started_at,
+            )
+        )
+
+        layer._handle_node_retry(
+            NodeRunRetryEvent(
+                id="exec",
+                node_id="node",
+                node_type=BuiltinNodeTypes.AGENT,
+                node_title="Agent",
+                start_at=started_at,
+                error="retry",
+                retry_index=1,
+                node_run_result=NodeRunResult(
+                    process_data={"workflow_agent_binding_id": "workflow-binding-1"},
+                ),
+            )
+        )
+
+        assert layer._node_execution_cache["exec"].process_data["workflow_agent_binding_id"] == "workflow-binding-1"
 
     def test_get_node_execution_raises_for_missing(self):
         layer, _, _, _ = _make_layer()

api/tests/unit_tests/core/repositories/test_celery_workflow_node_execution_repository.py

@@ -186,6 +186,23 @@ class TestCeleryWorkflowNodeExecutionRepository:
             in repo._workflow_execution_mapping[sample_workflow_node_execution.workflow_execution_id]
         )
 
+    def test_save_synchronously_uses_sql_repository_without_queueing(
+        self, mock_session_factory, mock_account, sample_workflow_node_execution
+    ):
+        repo = CeleryWorkflowNodeExecutionRepository(
+            session_factory=mock_session_factory,
+            tenant_id=RESOURCE_TENANT_ID,
+            user=mock_account,
+            app_id="test-app",
+            triggered_from=WorkflowNodeExecutionTriggeredFrom.WORKFLOW_RUN,
+        )
+        repo._sql_repository.save_synchronously = Mock()
+
+        repo.save_synchronously(sample_workflow_node_execution)
+
+        repo._sql_repository.save_synchronously.assert_called_once_with(sample_workflow_node_execution)
+        assert repo._execution_cache[sample_workflow_node_execution.id] is sample_workflow_node_execution
+
     @patch("core.repositories.celery_workflow_node_execution_repository.save_workflow_node_execution_task")
     def test_save_handles_celery_failure(
         self, mock_task, mock_session_factory, mock_account, sample_workflow_node_execution

api/tests/unit_tests/core/repositories/test_sqlalchemy_workflow_node_execution_repository.py

@@ -238,7 +238,11 @@ def test_to_db_model_requires_constructor_context(monkeypatch: pytest.MonkeyPatc
         app_id=None,
         triggered_from=WorkflowNodeExecutionTriggeredFrom.WORKFLOW_RUN,
     )
-    execution = _execution(inputs={"b": 1, "a": 2}, metadata={WorkflowNodeExecutionMetadataKey.TOTAL_TOKENS: 1})
+    execution = _execution(
+        inputs={"b": 1, "a": 2},
+        process_data={"agent_workspace_binding_id": "participant-1"},
+        metadata={WorkflowNodeExecutionMetadataKey.TOTAL_TOKENS: 1},
+    )
 
     # Happy path: deterministic json dump should be sorted
     db_model = repo._to_db_model(execution)
@@ -247,6 +251,7 @@ def test_to_db_model_requires_constructor_context(monkeypatch: pytest.MonkeyPatc
     assert db_model.created_by_role.value == "account"
     assert json.loads(db_model.inputs or "{}") == {"a": 2, "b": 1}
     assert json.loads(db_model.execution_metadata or "{}")["total_tokens"] == 1
+    assert db_model.agent_workspace_binding_id is None
 
     repo._triggered_from = None
     with pytest.raises(ValueError, match="triggered_from is required"):
@@ -330,7 +335,7 @@ def test_persist_to_database_updates_existing_and_inserts_new(monkeypatch: pytes
     db_model.foo = "bar"  # type: ignore[attr-defined]
     db_model.__dict__["_private"] = "x"
 
-    existing = SimpleNamespace()
+    existing = SimpleNamespace(process_data=None, process_data_dict=None)
     session.get.return_value = existing
     repo._persist_to_database(db_model)
     assert existing.foo == "bar"
@@ -523,7 +528,9 @@ def test_save_execution_data_handles_existing_db_model_and_truncation(monkeypatc
         offload_data=[WorkflowNodeExecutionOffload(type_=ExecutionOffLoadType.INPUTS)],
         inputs=None,
         outputs=None,
-        process_data=None,
+        process_data='{"workflow_agent_binding_id": "workflow-binding-1"}',
+        process_data_dict={"workflow_agent_binding_id": "workflow-binding-1"},
+        agent_workspace_binding_id="authoritative-participant",
     )
     session.merge = Mock()
     session.flush = Mock()
@@ -538,7 +545,11 @@ def test_save_execution_data_handles_existing_db_model_and_truncation(monkeypatc
         triggered_from=WorkflowNodeExecutionTriggeredFrom.WORKFLOW_RUN,
     )
 
-    execution = _execution(inputs={"a": 1}, outputs={"b": 2}, process_data={"c": 3})
+    execution = _execution(
+        inputs={"a": 1},
+        outputs={"b": 2},
+        process_data={"c": 3},
+    )
 
     trunc_result = SimpleNamespace(
         truncated_value={"trunc": True},
@@ -554,7 +565,11 @@ def test_save_execution_data_handles_existing_db_model_and_truncation(monkeypatc
     db_model = session.merge.call_args.args[0]
     assert json.loads(db_model.inputs) == {"trunc": True}
     assert json.loads(db_model.outputs) == {"b": 2}
-    assert json.loads(db_model.process_data) == {"c": 3}
+    assert json.loads(db_model.process_data) == {
+        "c": 3,
+        "workflow_agent_binding_id": "workflow-binding-1",
+    }
+    assert db_model.agent_workspace_binding_id == "authoritative-participant"
     assert any(off.type_ == ExecutionOffLoadType.INPUTS for off in db_model.offload_data)
     assert execution.get_truncated_inputs() == {"trunc": True}
 
@@ -570,6 +585,7 @@ def test_save_execution_data_truncates_outputs_and_process_data(monkeypatch: pyt
         inputs=None,
         outputs=None,
         process_data=None,
+        process_data_dict=None,
     )
     session = MagicMock()
     session.execute.return_value.scalars.return_value.first.return_value = existing
@@ -633,7 +649,14 @@ def test_save_execution_data_handles_missing_db_model(monkeypatch: pytest.Monkey
     )
 
     execution = _execution(inputs={"a": 1})
-    fake_db_model = SimpleNamespace(id=execution.id, offload_data=[], inputs=None, outputs=None, process_data=None)
+    fake_db_model = SimpleNamespace(
+        id=execution.id,
+        offload_data=[],
+        inputs=None,
+        outputs=None,
+        process_data=None,
+        process_data_dict=None,
+    )
     monkeypatch.setattr(repo, "_to_db_model", lambda *_: fake_db_model)
     monkeypatch.setattr(repo, "_truncate_and_upload", lambda *_args, **_kwargs: None)
     monkeypatch.setattr(repo, "_json_encode", lambda values: json.dumps(values))

api/tests/unit_tests/core/repositories/test_workflow_node_execution_conflict_handling.py

@@ -99,6 +99,7 @@ class TestWorkflowNodeExecutionConflictHandling:
 
         # Mock existing record
         mock_existing = MagicMock()
+        mock_existing.process_data_dict = None
         mock_session.get.return_value = mock_existing
         mock_session.commit.return_value = None
 

api/tests/unit_tests/core/workflow/nodes/agent_v2/test_agent_node.py

@@ -3,6 +3,7 @@ from datetime import UTC, datetime
 from types import SimpleNamespace
 from typing import cast
 from unittest.mock import MagicMock, patch
+from uuid import UUID
 
 from agenton.compositor import CompositorSessionSnapshot
 from dify_agent.layers.ask_human import AskHumanToolResult
@@ -24,7 +25,6 @@ from clients.agent_backend import (
     AgentBackendStreamInternalEvent,
     FakeAgentBackendRunClient,
     FakeAgentBackendScenario,
-    RuntimeLayerSpec,
 )
 from core.app.entities.app_invoke_entities import DIFY_RUN_CONTEXT_KEY, DifyRunContext, InvokeFrom, UserFrom
 from core.workflow.file_reference import build_file_reference
@@ -33,18 +33,22 @@ from core.workflow.nodes.agent_v2.ask_human_resume import AskHumanResumeOutcome
 from core.workflow.nodes.agent_v2.binding_resolver import WorkflowAgentBindingBundle, WorkflowAgentBindingResolver
 from core.workflow.nodes.agent_v2.entities import DifyAgentNodeData
 from core.workflow.nodes.agent_v2.output_adapter import WorkflowAgentOutputAdapter
-from core.workflow.nodes.agent_v2.runtime_request_builder import WorkflowAgentRuntimeRequestBuilder
+from core.workflow.nodes.agent_v2.runtime_request_builder import (
+    WorkflowAgentRuntimeBuildContext,
+    WorkflowAgentRuntimeRequestBuilder,
+)
 from core.workflow.nodes.agent_v2.session_store import (
     StoredWorkflowAgentSession,
-    WorkflowAgentRuntimeSessionStore,
     WorkflowAgentSessionScope,
+    WorkflowAgentWorkspaceStore,
 )
 from core.workflow.nodes.human_input.pause_reason import HumanInputRequired
 from graphon.entities import GraphInitParams
 from graphon.entities.pause_reason import HitlRequired
 from graphon.enums import BuiltinNodeTypes, WorkflowNodeExecutionMetadataKey, WorkflowNodeExecutionStatus
 from graphon.file import File, FileTransferMethod, FileType
-from graphon.node_events import PauseRequestedEvent, StreamCompletedEvent
+from graphon.graph_events import NodeRunPauseRequestedEvent
+from graphon.node_events import StreamCompletedEvent
 from graphon.runtime import GraphRuntimeState
 from graphon.variables.segments import ArrayFileSegment, FileSegment, StringSegment
 from models.agent import Agent, AgentConfigSnapshot, WorkflowAgentNodeBinding
@@ -54,6 +58,7 @@ from models.agent_config_entities import (
     DeclaredOutputType,
     WorkflowNodeJobConfig,
 )
+from services.agent.workspace_service import AgentWorkspaceNotFoundError
 
 
 class FakeCredentialsProvider:
@@ -95,12 +100,14 @@ class FakeVariablePool:
 
 class FakeBindingResolver(WorkflowAgentBindingResolver):
     def __init__(self):
+        self.calls: list[dict[str, object]] = []
         self.agent = Agent(id="agent-1", tenant_id="tenant-1", name="Agent")
         self.snapshot = AgentConfigSnapshot(
             id="snapshot-1",
             tenant_id="tenant-1",
             agent_id="agent-1",
             version=1,
+            home_snapshot_id="home-1",
             config_snapshot=AgentSoulConfig(
                 prompt={"system_prompt": "You are careful."},
                 model=AgentSoulModelConfig(
@@ -127,13 +134,29 @@ class FakeBindingResolver(WorkflowAgentBindingResolver):
             ),
         )
 
-    def resolve(self, **_kwargs):
+    def resolve(self, **kwargs):
+        self.calls.append(kwargs)
+        snapshot_id = kwargs.get("snapshot_id")
+        if isinstance(snapshot_id, str):
+            self.snapshot.id = snapshot_id
         return WorkflowAgentBindingBundle(binding=self.binding, agent=self.agent, snapshot=self.snapshot)
 
 
 class FakeSessionStore:
-    def __init__(self, snapshot: CompositorSessionSnapshot | None = None) -> None:
+    def __init__(
+        self,
+        snapshot: CompositorSessionSnapshot | None = None,
+        *,
+        binding_id: str = "binding-1",
+        workspace_id: str = "workspace-1",
+        backend_binding_ref: str = "backend-binding-1",
+    ) -> None:
         self.loaded_snapshot = snapshot
+        self.binding_id = binding_id
+        self.workspace_id = workspace_id
+        self.backend_binding_ref = backend_binding_ref
+        self.resolved_scopes: list[WorkflowAgentSessionScope] = []
+        self.existing_scope_lookups: list[dict[str, object]] = []
         # ENG-638: set to simulate resume after a submitted/timed-out form.
         self.loaded_session: StoredWorkflowAgentSession | None = None
         self.saved: list[
@@ -141,40 +164,43 @@ class FakeSessionStore:
                 WorkflowAgentSessionScope,
                 str,
                 CompositorSessionSnapshot | None,
-                list[RuntimeLayerSpec],
                 str | None,
                 str | None,
             ]
         ] = []
-        self.cleaned: list[tuple[WorkflowAgentSessionScope, str | None]] = []
 
-    def load_active_snapshot(self, scope: WorkflowAgentSessionScope) -> CompositorSessionSnapshot | None:
-        return self.loaded_snapshot
+    def load_existing_node_execution_scope(self, **kwargs: object) -> WorkflowAgentSessionScope | None:
+        self.existing_scope_lookups.append(kwargs)
+        return self.loaded_session.scope if self.loaded_session is not None else None
 
-    def load_active_session(self, scope: WorkflowAgentSessionScope) -> StoredWorkflowAgentSession | None:
-        return self.loaded_session
+    def load_or_create_node_execution_session(
+        self,
+        scope: WorkflowAgentSessionScope,
+        *,
+        home_snapshot_id: str,
+    ) -> StoredWorkflowAgentSession:
+        assert home_snapshot_id == "home-1"
+        self.resolved_scopes.append(scope)
+        if self.loaded_session is not None:
+            return self.loaded_session
+        return StoredWorkflowAgentSession(
+            scope=scope,
+            binding_id=self.binding_id,
+            workspace_id=self.workspace_id,
+            backend_binding_ref=self.backend_binding_ref,
+            session_snapshot=self.loaded_snapshot,
+        )
 
     def save_active_snapshot(
         self,
         *,
         scope: WorkflowAgentSessionScope,
-        backend_run_id: str,
+        binding_id: str,
         snapshot: CompositorSessionSnapshot | None,
-        runtime_layer_specs: list[RuntimeLayerSpec],
         pending_form_id: str | None = None,
         pending_tool_call_id: str | None = None,
     ) -> None:
-        self.saved.append(
-            (scope, backend_run_id, snapshot, list(runtime_layer_specs), pending_form_id, pending_tool_call_id)
-        )
-
-    def mark_cleaned(
-        self,
-        *,
-        scope: WorkflowAgentSessionScope,
-        backend_run_id: str | None = None,
-    ) -> None:
-        self.cleaned.append((scope, backend_run_id))
+        self.saved.append((scope, binding_id, snapshot, pending_form_id, pending_tool_call_id))
 
 
 class FileOutputBackendClient(FakeAgentBackendRunClient):
@@ -286,6 +312,8 @@ def _node(
     session_store: FakeSessionStore | None = None,
     declared_outputs: list[dict[str, object]] | None = None,
     agent_backend_client: FakeAgentBackendRunClient | None = None,
+    binding_resolver: FakeBindingResolver | None = None,
+    runtime_request_builder: WorkflowAgentRuntimeRequestBuilder | None = None,
 ) -> DifyAgentNode:
     graph_init_params = GraphInitParams(
         workflow_id="workflow-1",
@@ -309,7 +337,7 @@ def _node(
             return True
 
     client = agent_backend_client or FakeAgentBackendRunClient(scenario=scenario)
-    binding_resolver = FakeBindingResolver()
+    binding_resolver = binding_resolver or FakeBindingResolver()
     if declared_outputs is not None:
         binding_resolver.binding.node_job_config = WorkflowNodeJobConfig.model_validate(
             {
@@ -323,15 +351,22 @@ def _node(
         node_id="agent-node",
         data=DifyAgentNodeData.model_validate({"type": BuiltinNodeTypes.AGENT, "version": "2"}),
         graph_init_params=graph_init_params,
-        graph_runtime_state=cast(GraphRuntimeState, SimpleNamespace(variable_pool=FakeVariablePool())),
+        graph_runtime_state=cast(
+            GraphRuntimeState,
+            SimpleNamespace(
+                variable_pool=FakeVariablePool(),
+                graph_execution=SimpleNamespace(node_executions={}),
+            ),
+        ),
         binding_resolver=binding_resolver,
-        runtime_request_builder=WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()),
+        runtime_request_builder=runtime_request_builder
+        or WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()),
         agent_backend_client=client,
         event_adapter=AgentBackendRunEventAdapter(),
         output_adapter=WorkflowAgentOutputAdapter(),
         type_checker=PerOutputTypeChecker(file_validator=_AlwaysAllowFileValidator()),
         failure_orchestrator=OutputFailureOrchestrator(),
-        session_store=cast(WorkflowAgentRuntimeSessionStore | None, session_store),
+        session_store=cast(WorkflowAgentWorkspaceStore, session_store or FakeSessionStore()),
     )
 
 
@@ -367,6 +402,85 @@ def test_agent_node_run_maps_successful_agent_backend_run_to_node_result():
     assert layers["llm"]["config"]["credentials"] == "[REDACTED]"
 
 
+def test_agent_node_uses_resolved_backend_binding_before_backend_invocation() -> None:
+    client = FakeAgentBackendRunClient()
+    store = FakeSessionStore(binding_id="binding-2", backend_binding_ref="backend-binding-2")
+
+    events = list(_node(agent_backend_client=client, session_store=store)._run())
+
+    assert len(events) == 1
+    assert client.request is not None
+    layers = {layer["name"]: layer for layer in client.request.model_dump(mode="json")["composition"]["layers"]}
+    assert layers["runtime"]["config"]["backend_binding_ref"] == "backend-binding-2"
+    assert store.saved[0][1] == "binding-2"
+    assert len(store.resolved_scopes) == 1
+
+
+def test_agent_node_resume_resolves_the_generation_from_the_persisted_execution() -> None:
+    binding_resolver = FakeBindingResolver()
+    store = FakeSessionStore()
+    store.loaded_session = StoredWorkflowAgentSession(
+        scope=WorkflowAgentSessionScope(
+            tenant_id="tenant-1",
+            app_id="app-1",
+            workflow_id="workflow-1",
+            workflow_run_id="workflow-run-1",
+            node_id="agent-node",
+            node_execution_id="exec-1",
+            workflow_agent_binding_id="binding-1",
+            agent_id="agent-1",
+            agent_config_snapshot_id="snapshot-pinned",
+        ),
+        binding_id="workspace-binding-1",
+        workspace_id="workspace-1",
+        backend_binding_ref="backend-binding-1",
+        session_snapshot=None,
+    )
+    node = _node(binding_resolver=binding_resolver, session_store=store)
+
+    events = list(node._run())
+
+    assert len(events) == 1
+    assert store.existing_scope_lookups[0]["node_execution_id"] == node.execution_id
+    assert binding_resolver.calls[0]["binding_id"] == "binding-1"
+    assert binding_resolver.calls[0]["snapshot_id"] == "snapshot-pinned"
+
+
+def test_agent_node_maps_persisted_participant_lookup_error_to_node_failure() -> None:
+    class _UnavailableParticipantStore(FakeSessionStore):
+        def load_existing_node_execution_scope(self, **kwargs: object) -> WorkflowAgentSessionScope | None:
+            del kwargs
+            raise AgentWorkspaceNotFoundError("Workflow node participant Binding is unavailable")
+
+    events = list(_node(session_store=_UnavailableParticipantStore())._run())
+
+    assert len(events) == 1
+    result = cast(StreamCompletedEvent, events[0]).node_run_result
+    assert result.status == WorkflowNodeExecutionStatus.FAILED
+    assert result.error == "Workflow node participant Binding is unavailable"
+    assert result.error_type == "agent_workflow_node_runtime_error"
+
+
+def test_agent_node_passes_execution_id_to_session_store_and_runtime_request_builder() -> None:
+    store = FakeSessionStore()
+    request_builder = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider())
+    node = _node(session_store=store, runtime_request_builder=request_builder)
+    execution_id = node.ensure_execution_id()
+
+    with patch.object(request_builder, "build", wraps=request_builder.build) as build:
+        list(node._run())
+
+    assert str(UUID(execution_id)) == execution_id
+    assert execution_id != node.id
+    scope = store.resolved_scopes[0]
+    build.assert_called_once()
+    context = cast(WorkflowAgentRuntimeBuildContext, build.call_args.args[0])
+    assert scope.node_id == node.id
+    assert scope.node_execution_id == execution_id
+    assert context.node_id == node.id
+    assert context.node_execution_id == execution_id
+
+
 def test_agent_node_run_ignores_agent_message_delta_until_terminal_result():
     events = list(_node(agent_backend_client=AgentMessageDeltaBackendClient())._run())
 
@@ -491,23 +605,6 @@ def test_agent_node_run_maps_failed_agent_backend_run_to_node_result():
     assert result.error_type == "unit_test"
 
 
-def test_agent_node_failed_run_marks_session_cleaned_to_prevent_stale_reuse():
-    """A failed agent run must retire the local ACTIVE session row so a workflow
-    loop back into the same Agent node does not resume from a stale snapshot."""
-    existing_snapshot = CompositorSessionSnapshot(layers=[])
-    store = FakeSessionStore(snapshot=existing_snapshot)
-
-    events = list(_node(scenario=FakeAgentBackendScenario.FAILED, session_store=store)._run())
-
-    assert len(events) == 1
-    assert store.cleaned, "failed agent run should mark the session cleaned"
-    cleaned_scope, cleaned_backend_run_id = store.cleaned[0]
-    assert cleaned_scope.workflow_run_id == "workflow-run-1"
-    assert cleaned_backend_run_id == "fake-run-1"
-    # A failed run does not produce a fresh snapshot to persist.
-    assert store.saved == []
-
-
 def test_agent_node_saves_success_snapshot_and_reuses_existing_snapshot():
     existing_snapshot = CompositorSessionSnapshot(layers=[])
     store = FakeSessionStore(snapshot=existing_snapshot)
@@ -518,21 +615,15 @@ def test_agent_node_saves_success_snapshot_and_reuses_existing_snapshot():
 
     assert len(events) == 1
     assert store.saved
-    scope, backend_run_id, saved_snapshot, saved_specs, pending_form_id, pending_tool_call_id = store.saved[0]
+    scope, binding_id, saved_snapshot, pending_form_id, pending_tool_call_id = store.saved[0]
     assert scope.workflow_run_id == "workflow-run-1"
-    assert backend_run_id == "fake-run-1"
+    assert binding_id == "binding-1"
     assert saved_snapshot is not None
     # A successful terminal carries no ask_human pause correlation.
     assert pending_form_id is None
     assert pending_tool_call_id is None
     assert client.request is not None
     assert client.request.session_snapshot is existing_snapshot
-    # Persist enough composition shape to replay a cleanup run; plugin layers
-    # (which would carry credentials) are intentionally absent.
-    saved_layer_names = [spec.name for spec in saved_specs]
-    assert saved_layer_names, "cleanup specs must persist at least the non-plugin layers"
-    plugin_types = {"dify.plugin.llm", "dify.plugin.tools"}
-    assert not {spec.type for spec in saved_specs} & plugin_types
 
 
 def test_agent_node_run_when_session_store_save_raises_records_persist_error_in_metadata():
@@ -553,83 +644,7 @@ def test_agent_node_run_when_session_store_save_raises_records_persist_error_in_
     assert result.status == WorkflowNodeExecutionStatus.SUCCEEDED
     agent_backend = result.metadata[WorkflowNodeExecutionMetadataKey.AGENT_LOG]["agent_backend"]
     assert agent_backend["session_snapshot_persisted"] is False
-    assert agent_backend["session_snapshot_persist_error"] == "workflow_agent_runtime_session_store_error"
-
-
-def test_agent_node_failed_run_when_mark_cleaned_raises_records_cleanup_error_in_metadata():
-    """Same defensive pattern: a DB-side mark_cleaned failure must surface as
-    a ``session_snapshot_cleanup_error`` in metadata, not as a node crash."""
-
-    class _ExplodingMarkCleanedStore(FakeSessionStore):
-        def mark_cleaned(self, **kwargs):  # type: ignore[override]
-            del kwargs
-            raise RuntimeError("simulated DB failure")
-
-    store = _ExplodingMarkCleanedStore()
-    events = list(_node(scenario=FakeAgentBackendScenario.FAILED, session_store=store)._run())
-
-    assert len(events) == 1
-    result = cast(StreamCompletedEvent, events[0]).node_run_result
-    assert result.status == WorkflowNodeExecutionStatus.FAILED
-    agent_backend = result.metadata[WorkflowNodeExecutionMetadataKey.AGENT_LOG]["agent_backend"]
-    assert agent_backend["session_snapshot_cleaned_on_failure"] is False
-    assert agent_backend["session_snapshot_cleanup_error"] == "workflow_agent_runtime_session_store_error"
-
-
-def test_agent_node_success_run_without_session_store_skips_persistence():
-    """When ``session_store`` is None the node still completes successfully —
-    the lifecycle branch is a no-op and the run result is unaffected."""
-    events = list(_node(session_store=None)._run())
-
-    assert len(events) == 1
-    result = cast(StreamCompletedEvent, events[0]).node_run_result
-    assert result.status == WorkflowNodeExecutionStatus.SUCCEEDED
-    agent_backend = result.metadata[WorkflowNodeExecutionMetadataKey.AGENT_LOG]["agent_backend"]
-    # No persistence metadata is attached when the store is missing.
-    assert "session_snapshot_persisted" not in agent_backend
-
-
-def test_agent_node_failed_run_without_session_store_skips_mark_cleaned():
-    """``session_store=None`` + failed terminal must remain a no-op for
-    the cleanup branch — the node failure path still surfaces correctly."""
-    events = list(_node(scenario=FakeAgentBackendScenario.FAILED, session_store=None)._run())
-
-    assert len(events) == 1
-    result = cast(StreamCompletedEvent, events[0]).node_run_result
-    assert result.status == WorkflowNodeExecutionStatus.FAILED
-    agent_backend = result.metadata[WorkflowNodeExecutionMetadataKey.AGENT_LOG]["agent_backend"]
-    assert "session_snapshot_cleaned_on_failure" not in agent_backend
-
-
-def test_agent_node_failed_run_enqueues_backend_cleanup_before_local_retirement(monkeypatch):
-    store = FakeSessionStore()
-    store.loaded_session = StoredWorkflowAgentSession(
-        scope=_pending_session(CompositorSessionSnapshot(layers=[])).scope,
-        session_snapshot=CompositorSessionSnapshot(layers=[]),
-        backend_run_id="stored-run-1",
-        runtime_layer_specs=[RuntimeLayerSpec(name="history", type="pydantic_ai.history")],
-    )
-    queued_payloads: list[dict[str, object]] = []
-    monkeypatch.setattr(
-        "core.workflow.nodes.agent_v2.agent_node.cleanup_workflow_agent_runtime_session.delay",
-        lambda payload: queued_payloads.append(payload),
-    )
-
-    events = list(_node(scenario=FakeAgentBackendScenario.FAILED, session_store=store)._run())
-
-    assert len(events) == 1
-    result = cast(StreamCompletedEvent, events[0]).node_run_result
-    assert result.status == WorkflowNodeExecutionStatus.FAILED
-    assert store.cleaned[0][1] == "fake-run-1"
-    assert store.cleaned[0][0].workflow_run_id == "workflow-run-1"
-    assert store.cleaned[0][0].node_id == "agent-node"
-    assert len(queued_payloads) == 1
-    assert (
-        queued_payloads[0]["idempotency_key"]
-        == "tenant-1:workflow-run-1:agent-node:binding-1:workflow-agent-failure-cleanup:stored-run-1:fake-run-1"
-    )
-    assert queued_payloads[0]["metadata"]["previous_agent_backend_run_id"] == "stored-run-1"
-    assert queued_payloads[0]["metadata"]["failed_agent_backend_run_id"] == "fake-run-1"
+    assert agent_backend["session_snapshot_persist_error"] == "workflow_agent_workspace_store_error"
 
 
 def test_agent_node_paused_run_requests_workflow_pause_and_persists_snapshot():
@@ -646,17 +661,21 @@ def test_agent_node_paused_run_requests_workflow_pause_and_persists_snapshot():
     events = list(node._run())
 
     assert len(events) == 1
-    assert isinstance(events[0], PauseRequestedEvent)
+    assert isinstance(events[0], NodeRunPauseRequestedEvent)
     assert isinstance(events[0].reason, HitlRequired)
     assert events[0].reason.session_id == "form-1"
     assert events[0].reason.node_id == "agent-node"
+    assert events[0].node_run_result.process_data == {
+        "agent_id": "agent-1",
+        "agent_config_snapshot_id": "snapshot-1",
+        "workflow_agent_binding_id": "binding-1",
+    }
     fake_repo.create_form.assert_called_once()
     assert store.saved
-    assert store.saved[0][1] == "fake-run-1"
-    assert store.saved[0][3], "paused agent run should still persist replayable layer specs"
+    assert store.saved[0][1] == "binding-1"
     # ENG-637: the awaiting form + deferred tool_call correlation is persisted.
-    assert store.saved[0][4] == "form-1"
-    assert store.saved[0][5] == "fake-ask-human-1"
+    assert store.saved[0][3] == "form-1"
+    assert store.saved[0][4] == "fake-ask-human-1"
 
 
 def _pending_session(snapshot: CompositorSessionSnapshot) -> StoredWorkflowAgentSession:
@@ -668,12 +687,14 @@ def _pending_session(snapshot: CompositorSessionSnapshot) -> StoredWorkflowAgent
             workflow_run_id="workflow-run-1",
             node_id="agent-node",
             node_execution_id="exec-1",
-            binding_id="binding-1",
+            workflow_agent_binding_id="binding-1",
             agent_id="agent-1",
             agent_config_snapshot_id="snapshot-1",
         ),
+        binding_id="binding-1",
+        workspace_id="workspace-1",
+        backend_binding_ref="backend-binding-1",
         session_snapshot=snapshot,
-        backend_run_id="run-0",
         pending_form_id="form-1",
         pending_tool_call_id="call-1",
     )
@@ -727,19 +748,75 @@ def test_agent_node_repauses_when_resumed_form_still_waiting(monkeypatch):
     events = list(node._run())
 
     assert len(events) == 1
-    assert isinstance(events[0], PauseRequestedEvent)
+    assert isinstance(events[0], NodeRunPauseRequestedEvent)
     assert isinstance(events[0].reason, HitlRequired)
+    assert events[0].node_run_result.process_data["workflow_agent_binding_id"] == "binding-1"
     assert client.request is None  # no second Agent run was created
 
 
+def test_agent_node_expired_ask_human_failure_keeps_binding_identity(monkeypatch):
+    snapshot = CompositorSessionSnapshot(layers=[])
+    store = FakeSessionStore(snapshot=snapshot)
+    store.loaded_session = _pending_session(snapshot)
+
+    def _raise_expired_form(**_kwargs):
+        raise AssertionError("cannot resume globally expired ask_human form, form_id=form-1")
+
+    monkeypatch.setattr(
+        "core.workflow.nodes.agent_v2.agent_node.resolve_ask_human_form",
+        _raise_expired_form,
+    )
+
+    events = list(_node(session_store=store)._run())
+
+    assert len(events) == 1
+    result = cast(StreamCompletedEvent, events[0]).node_run_result
+    assert result.status == WorkflowNodeExecutionStatus.FAILED
+    assert result.error == "cannot resume globally expired ask_human form, form_id=form-1"
+    assert result.error_type == "agent_workflow_node_runtime_error"
+    assert result.process_data["workflow_agent_binding_id"] == "binding-1"
+    assert "agent_workspace_binding_id" not in result.process_data
+
+
+def test_agent_node_unexpected_post_resolution_failure_keeps_binding_identity():
+    class _FailingSessionStore(FakeSessionStore):
+        def load_or_create_node_execution_session(
+            self,
+            scope: WorkflowAgentSessionScope,
+            *,
+            home_snapshot_id: str,
+        ) -> StoredWorkflowAgentSession:
+            del scope, home_snapshot_id
+            raise RuntimeError("session store failed")
+
+    events = list(_node(session_store=_FailingSessionStore())._run())
+
+    assert len(events) == 1
+    result = cast(StreamCompletedEvent, events[0]).node_run_result
+    assert result.status == WorkflowNodeExecutionStatus.FAILED
+    assert result.error == "session store failed"
+    assert result.error_type == "agent_workflow_node_runtime_error"
+    assert result.process_data == {
+        "agent_id": "agent-1",
+        "agent_config_snapshot_id": "snapshot-1",
+        "workflow_agent_binding_id": "binding-1",
+    }
+
+
 def test_agent_node_cancels_backend_run_when_stream_fails():
     client = FailingStreamBackendClient()
     node = _node(agent_backend_client=client)
 
-    terminal, failure = node._consume_event_stream("run-1", {"agent_backend": {}})
+    terminal, failure = node._consume_event_stream(
+        "run-1",
+        inputs={},
+        process_data={"workflow_agent_binding_id": "binding-1"},
+        metadata={"agent_backend": {}},
+    )
 
     assert terminal is None
     assert failure is not None
+    assert failure.node_run_result.process_data == {"workflow_agent_binding_id": "binding-1"}
     assert len(client.cancel_requests) == 1
     assert client.cancel_requests[0] is not None
     assert client.cancel_requests[0].reason == "event_stream_failed"
@@ -749,7 +826,12 @@ def test_agent_node_cancels_backend_run_when_stream_ends_without_terminal_event(
     client = EmptyStreamBackendClient()
     node = _node(agent_backend_client=client)
 
-    terminal, failure = node._consume_event_stream("run-1", {"agent_backend": {}})
+    terminal, failure = node._consume_event_stream(
+        "run-1",
+        inputs={},
+        process_data={"workflow_agent_binding_id": "binding-1"},
+        metadata={"agent_backend": {}},
+    )
 
     assert terminal is None
     assert failure is None
@@ -761,11 +843,17 @@ def test_agent_node_cancels_backend_run_when_stream_raises_unexpected_error():
     client = GenericFailingStreamBackendClient()
     node = _node(agent_backend_client=client)
 
-    terminal, failure = node._consume_event_stream("run-1", {"agent_backend": {}})
+    terminal, failure = node._consume_event_stream(
+        "run-1",
+        inputs={},
+        process_data={"workflow_agent_binding_id": "binding-1"},
+        metadata={"agent_backend": {}},
+    )
 
     assert terminal is None
     assert failure is not None
     assert failure.node_run_result.error == "unexpected stream failure"
+    assert failure.node_run_result.process_data == {"workflow_agent_binding_id": "binding-1"}
     assert client.cancel_requests[0] is not None
     assert client.cancel_requests[0].reason == "event_stream_failed"
 
@@ -775,7 +863,12 @@ def test_agent_node_uses_graph_abort_reason_when_cancel_request_fails(caplog):
     node = _node(agent_backend_client=client)
     node.graph_runtime_state.graph_execution = SimpleNamespace(aborted=True)
 
-    terminal, failure = node._consume_event_stream("run-1", {"agent_backend": {}})
+    terminal, failure = node._consume_event_stream(
+        "run-1",
+        inputs={},
+        process_data={"workflow_agent_binding_id": "binding-1"},
+        metadata={"agent_backend": {}},
+    )
 
     assert terminal is None
     assert failure is not None
@@ -794,13 +887,19 @@ def test_agent_node_cancels_backend_run_for_unexpected_internal_event():
         return_value=[SimpleNamespace(type=AgentBackendInternalEventType.RUN_FAILED)]
     )
 
-    terminal, failure = node._consume_event_stream("run-1", {"agent_backend": {}})
+    terminal, failure = node._consume_event_stream(
+        "run-1",
+        inputs={},
+        process_data={"workflow_agent_binding_id": "binding-1"},
+        metadata={"agent_backend": {}},
+    )
 
     assert terminal is None
     assert failure is not None
     assert failure.node_run_result.error == (
         "Unexpected internal event type <AgentBackendInternalEventType.RUN_FAILED: 'run_failed'>"
     )
+    assert failure.node_run_result.process_data == {"workflow_agent_binding_id": "binding-1"}
     node._agent_backend_client.cancel_run.assert_called_once()
 
 

api/tests/unit_tests/core/workflow/nodes/agent_v2/test_binding_resolver.py

@@ -23,6 +23,7 @@ class FakeSession:
     def __init__(self, scalar_results):
         self._scalar_results = list(scalar_results)
         self.expunge_calls = []
+        self.scalar_statements = []
 
     def __enter__(self):
         return self
@@ -30,7 +31,8 @@ class FakeSession:
     def __exit__(self, exc_type, exc, tb):
         return False
 
-    def scalar(self, _stmt):
+    def scalar(self, stmt):
+        self.scalar_statements.append(stmt)
         if not self._scalar_results:
             return None
         return self._scalar_results.pop(0)
@@ -62,6 +64,7 @@ def _snapshot() -> AgentConfigSnapshot:
         tenant_id="tenant-1",
         agent_id="agent-1",
         version=1,
+        home_snapshot_id="home-1",
         config_snapshot=AgentSoulConfig(
             model=AgentSoulModelConfig(
                 plugin_id="langgenius/openai",
@@ -115,6 +118,73 @@ def test_binding_resolver_uses_active_snapshot_for_roster_agent(monkeypatch: pyt
     assert bundle.snapshot.id == "active-snapshot"
 
 
+@pytest.mark.parametrize(
+    "binding_type",
+    [WorkflowAgentBindingType.ROSTER_AGENT, WorkflowAgentBindingType.INLINE_AGENT],
+)
+def test_binding_resolver_uses_pinned_snapshot_for_existing_node_execution(
+    monkeypatch: pytest.MonkeyPatch,
+    binding_type: WorkflowAgentBindingType,
+):
+    binding = _binding()
+    binding.binding_type = binding_type
+    agent = _agent()
+    agent.active_config_snapshot_id = "active-snapshot"
+    snapshot = _snapshot()
+    snapshot.id = "pinned-snapshot"
+    fake_session = FakeSession([binding, agent, snapshot])
+    monkeypatch.setattr(
+        "core.workflow.nodes.agent_v2.binding_resolver.session_factory.create_session",
+        lambda: fake_session,
+    )
+
+    bundle = WorkflowAgentBindingResolver().resolve(
+        **_resolve(),
+        binding_id="binding-1",
+        snapshot_id="pinned-snapshot",
+    )
+
+    assert bundle.snapshot.id == "pinned-snapshot"
+    assert "binding-1" in fake_session.scalar_statements[0].compile().params.values()
+    assert "pinned-snapshot" in fake_session.scalar_statements[-1].compile().params.values()
+
+
+def test_binding_resolver_does_not_fallback_from_an_explicit_empty_snapshot(monkeypatch: pytest.MonkeyPatch):
+    binding = _binding()
+    binding.binding_type = WorkflowAgentBindingType.ROSTER_AGENT
+    agent = _agent()
+    agent.active_config_snapshot_id = "active-snapshot"
+    fake_session = FakeSession([binding, agent, None])
+    monkeypatch.setattr(
+        "core.workflow.nodes.agent_v2.binding_resolver.session_factory.create_session",
+        lambda: fake_session,
+    )
+
+    with pytest.raises(WorkflowAgentBindingError) as exc_info:
+        WorkflowAgentBindingResolver().resolve(**_resolve(), binding_id="binding-1", snapshot_id="")
+
+    assert exc_info.value.error_code == "agent_config_snapshot_not_found"
+    assert "" in fake_session.scalar_statements[-1].compile().params.values()
+
+
+@pytest.mark.parametrize(
+    ("binding_id", "snapshot_id"),
+    [("binding-1", None), (None, "snapshot-1")],
+)
+def test_binding_resolver_rejects_half_pinned_generation(
+    binding_id: str | None,
+    snapshot_id: str | None,
+) -> None:
+    with pytest.raises(WorkflowAgentBindingError) as exc_info:
+        WorkflowAgentBindingResolver().resolve(
+            **_resolve(),
+            binding_id=binding_id,
+            snapshot_id=snapshot_id,
+        )
+
+    assert exc_info.value.error_code == "agent_binding_generation_invalid"
+
+
 def test_binding_resolver_rejects_unpublished_roster_agent(monkeypatch: pytest.MonkeyPatch):
     binding = _binding()
     binding.binding_type = WorkflowAgentBindingType.ROSTER_AGENT