fix: block workflow DSL export with agent nodes

[autofix.ci] apply automated fixes
chore: fs download show
2026-07-06 17:48:26 +08:00 · 2026-07-06 09:33:21 +00:00 · 2026-07-06 17:27:38 +08:00 · 2026-07-06 16:42:59 +08:00 · 2026-07-06 16:37:26 +08:00 · 2026-07-06 16:35:57 +08:00
148 changed files with 3318 additions and 909 deletions
@@ -29,13 +29,13 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          fetch-depth: 0
          persist-credentials: false

      - name: Setup UV and Python
-        uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39 # v8.2.0
+        uses: astral-sh/setup-uv@d31148d669074a8d0a63714ba94f3201e7020bc3 # v8.3.0
        with:
          enable-cache: true
          python-version: ${{ matrix.python-version }}
@@ -91,13 +91,13 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          fetch-depth: 0
          persist-credentials: false

      - name: Setup UV and Python
-        uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39 # v8.2.0
+        uses: astral-sh/setup-uv@d31148d669074a8d0a63714ba94f3201e7020bc3 # v8.3.0
        with:
          enable-cache: true
          python-version: ${{ matrix.python-version }}
@@ -142,13 +142,13 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          fetch-depth: 0
          persist-credentials: false

      - name: Setup UV and Python
-        uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39 # v8.2.0
+        uses: astral-sh/setup-uv@d31148d669074a8d0a63714ba94f3201e7020bc3 # v8.3.0
        with:
          enable-cache: true
          python-version: "3.12"
@@ -20,7 +20,7 @@ jobs:
        run: echo "autofix.ci updates pull request branches, not merge group refs."

      - if: github.event_name != 'merge_group'
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0

      - name: Check Docker Compose inputs
        if: github.event_name != 'merge_group'
@@ -73,12 +73,12 @@ jobs:
            dify-agent/pyproject.toml
            dify-agent/uv.lock
      - if: github.event_name != 'merge_group'
-        uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6.2.0
+        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
        with:
          python-version: "3.11"

      - if: github.event_name != 'merge_group'
-        uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39 # v8.2.0
+        uses: astral-sh/setup-uv@d31148d669074a8d0a63714ba94f3201e7020bc3 # v8.3.0

      - name: Generate Docker Compose
        if: github.event_name != 'merge_group' && steps.docker-compose-changes.outputs.any_changed == 'true'
@@ -97,7 +97,7 @@ jobs:
          echo "PLATFORM_PAIR=${platform//\//-}" >> $GITHUB_ENV

      - name: Login to Docker Hub
-        uses: docker/login-action@650006c6eb7dba73a995cc03b0b2d7f5ca915bee # v4.2.0
+        uses: docker/login-action@af1e73f918a031802d376d3c8bbc3fe56130a9b0 # v4.4.0
        with:
          username: ${{ env.DOCKERHUB_USER }}
          password: ${{ env.DOCKERHUB_TOKEN }}
@@ -107,7 +107,7 @@ jobs:

      - name: Extract metadata for Docker
        id: meta
-        uses: docker/metadata-action@80c7e94dd9b9319bd5eb7a0e0fe9291e23a2a2e9 # v6.1.0
+        uses: docker/metadata-action@dc802804100637a589fabce1cb79ff13a1411302 # v6.2.0
        with:
          images: ${{ env[matrix.image_name_env] }}

@@ -159,10 +159,10 @@ jobs:
            file: "docker/local-sandbox/Dockerfile"
    steps:
      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@d7f5e7f509e45cec5c76c4d5afdd7de93d0b3df5 # v4.1.0
+        uses: docker/setup-buildx-action@bb05f3f5519dd87d3ba754cc423b652a5edd6d2c # v4.2.0

      - name: Validate Docker image
-        uses: docker/build-push-action@f9f3042f7e2789586610d6e8b85c8f03e5195baf # v7.2.0
+        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7.3.0
        with:
          push: false
          context: ${{ matrix.build_context }}
@@ -197,14 +197,14 @@ jobs:
          merge-multiple: true

      - name: Login to Docker Hub
-        uses: docker/login-action@650006c6eb7dba73a995cc03b0b2d7f5ca915bee # v4.2.0
+        uses: docker/login-action@af1e73f918a031802d376d3c8bbc3fe56130a9b0 # v4.4.0
        with:
          username: ${{ env.DOCKERHUB_USER }}
          password: ${{ env.DOCKERHUB_TOKEN }}

      - name: Extract metadata for Docker
        id: meta
-        uses: docker/metadata-action@80c7e94dd9b9319bd5eb7a0e0fe9291e23a2a2e9 # v6.1.0
+        uses: docker/metadata-action@dc802804100637a589fabce1cb79ff13a1411302 # v6.2.0
        with:
          images: ${{ env[matrix.image_name_env] }}
          tags: |
@@ -79,7 +79,7 @@ jobs:
      ws2_app_id:                 ${{ steps.out.outputs.DIFY_E2E_WS2_APP_ID }}

    steps:
-      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v4
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v4
        with:
          ref: ${{ inputs.cli_ref || github.ref }}
          persist-credentials: false
@@ -123,7 +123,7 @@ jobs:
        shell: bash

    steps:
-      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v4
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v4
        with:
          ref: ${{ inputs.cli_ref || github.ref }}
          persist-credentials: false
@@ -170,7 +170,7 @@ jobs:
        shell: bash

    steps:
-      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v4
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v4
        with:
          ref: ${{ inputs.cli_ref || github.ref }}
          persist-credentials: false
@@ -233,7 +233,7 @@ jobs:
        shell: bash

    steps:
-      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v4
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v4
        with:
          ref: ${{ inputs.cli_ref || github.ref }}
          persist-credentials: false
@@ -295,7 +295,7 @@ jobs:
        shell: bash

    steps:
-      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v4
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v4
        with:
          ref: ${{ inputs.cli_ref || github.ref }}
          persist-credentials: false
@@ -351,7 +351,7 @@ jobs:
        shell: bash

    steps:
-      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v4
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v4
        with:
          ref: ${{ inputs.cli_ref || github.ref }}
          persist-credentials: false
@@ -23,7 +23,7 @@ jobs:
        working-directory: ./cli
    steps:
      - name: Checkout
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false
          fetch-depth: 0
@@ -35,7 +35,7 @@ jobs:
      dify_tag: ${{ steps.resolve.outputs.dify_tag }}
    steps:
      - name: Checkout
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false

@@ -98,7 +98,7 @@ jobs:
      DIFY_TAG: ${{ needs.validate.outputs.dify_tag }}
    steps:
      - name: Checkout
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false
          fetch-depth: 1
@@ -24,7 +24,7 @@ jobs:
        shell: bash
    steps:
      - name: Checkout cli ref
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          ref: ${{ inputs.cli_ref || github.ref }}
          persist-credentials: false
@@ -30,7 +30,7 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false

@@ -13,13 +13,13 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          fetch-depth: 0
          persist-credentials: false

      - name: Setup UV and Python
-        uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39 # v8.2.0
+        uses: astral-sh/setup-uv@d31148d669074a8d0a63714ba94f3201e7020bc3 # v8.3.0
        with:
          enable-cache: true
          python-version: "3.12"
@@ -63,13 +63,13 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          fetch-depth: 0
          persist-credentials: false

      - name: Setup UV and Python
-        uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39 # v8.2.0
+        uses: astral-sh/setup-uv@d31148d669074a8d0a63714ba94f3201e7020bc3 # v8.3.0
        with:
          enable-cache: true
          python-version: "3.12"
@@ -77,10 +77,10 @@ jobs:
            file: "web/Dockerfile"
    steps:
      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@d7f5e7f509e45cec5c76c4d5afdd7de93d0b3df5 # v4.1.0
+        uses: docker/setup-buildx-action@bb05f3f5519dd87d3ba754cc423b652a5edd6d2c # v4.2.0

      - name: Build Docker Image
-        uses: docker/build-push-action@f9f3042f7e2789586610d6e8b85c8f03e5195baf # v7.2.0
+        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7.3.0
        with:
          push: false
          context: ${{ matrix.context }}
@@ -24,7 +24,7 @@ jobs:
    name: Require cherry-pick provenance
    runs-on: depot-ubuntu-24.04
    steps:
-      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          fetch-depth: 0

@@ -46,8 +46,8 @@ jobs:
      vdb-changed: ${{ steps.changes.outputs.vdb }}
      migration-changed: ${{ steps.changes.outputs.migration }}
    steps:
-      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
-      - uses: dorny/paths-filter@fbd0ab8f3e69293af611ebaee6363fc25e6d187d # v4.0.1
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+      - uses: dorny/paths-filter@7b450fff21473bca461d4b92ce414b9d0420d706 # v4.0.2
        id: changes
        with:
          filters: |
@@ -18,8 +18,8 @@ jobs:
    outputs:
      external-e2e-changed: ${{ steps.changes.outputs.external_e2e }}
    steps:
-      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
-      - uses: dorny/paths-filter@fbd0ab8f3e69293af611ebaee6363fc25e6d187d # v4.0.1
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+      - uses: dorny/paths-filter@7b450fff21473bca461d4b92ce414b9d0420d706 # v4.0.2
        id: changes
        with:
          filters: |
@@ -17,12 +17,12 @@ jobs:
      pull-requests: write
    steps:
      - name: Checkout PR branch
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          fetch-depth: 0

      - name: Setup Python & UV
-        uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39 # v8.2.0
+        uses: astral-sh/setup-uv@d31148d669074a8d0a63714ba94f3201e7020bc3 # v8.3.0
        with:
          enable-cache: true

@@ -21,10 +21,10 @@ jobs:
    if: ${{ github.event.workflow_run.conclusion == 'success' && github.event.workflow_run.pull_requests[0].head.repo.full_name != github.repository }}
    steps:
      - name: Checkout default branch (trusted code)
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0

      - name: Setup Python & UV
-        uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39 # v8.2.0
+        uses: astral-sh/setup-uv@d31148d669074a8d0a63714ba94f3201e7020bc3 # v8.3.0
        with:
          enable-cache: true

@@ -17,12 +17,12 @@ jobs:
      pull-requests: write
    steps:
      - name: Checkout PR branch
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          fetch-depth: 0

      - name: Setup Python & UV
-        uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39 # v8.2.0
+        uses: astral-sh/setup-uv@d31148d669074a8d0a63714ba94f3201e7020bc3 # v8.3.0
        with:
          enable-cache: true

@@ -19,7 +19,7 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false
          fetch-depth: 0
@@ -36,7 +36,7 @@ jobs:

      - name: Setup UV and Python
        if: steps.changed-files.outputs.any_changed == 'true'
-        uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39 # v8.2.0
+        uses: astral-sh/setup-uv@d31148d669074a8d0a63714ba94f3201e7020bc3 # v8.3.0
        with:
          enable-cache: false
          python-version: "3.12"
@@ -88,7 +88,7 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false

@@ -139,7 +139,7 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false

@@ -168,7 +168,7 @@ jobs:
      - name: Restore ESLint cache
        if: steps.changed-files.outputs.any_changed == 'true'
        id: eslint-cache-restore
-        uses: actions/cache/restore@27d5ce7f107fe9357f9df03efb73ab90386fccae # v5.0.5
+        uses: actions/cache/restore@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
        with:
          path: .eslintcache
          key: ${{ runner.os }}-eslint-${{ hashFiles('pnpm-lock.yaml', 'eslint.config.mjs', 'web/eslint.config.mjs', 'web/eslint.constants.mjs', 'web/plugins/eslint/**') }}-${{ github.sha }}
@@ -185,7 +185,7 @@ jobs:

      - name: Save ESLint cache
        if: steps.changed-files.outputs.any_changed == 'true' && success() && steps.eslint-cache-restore.outputs.cache-hit != 'true'
-        uses: actions/cache/save@27d5ce7f107fe9357f9df03efb73ab90386fccae # v5.0.5
+        uses: actions/cache/save@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
        with:
          path: .eslintcache
          key: ${{ steps.eslint-cache-restore.outputs.cache-primary-key }}
@@ -196,7 +196,7 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          fetch-depth: 0
          persist-credentials: false
@@ -214,7 +214,7 @@ jobs:
            .editorconfig

      - name: Super-linter
-        uses: super-linter/super-linter/slim@9e863354e3ff62e0727d37183162c4a88873df41 # v8.6.0
+        uses: super-linter/super-linter/slim@4ce20838b8ab83717e78138c5b3a1407148e0918 # v8.7.0
        if: steps.changed-files.outputs.any_changed == 'true'
        env:
          BASH_SEVERITY: warning
@@ -24,7 +24,7 @@ jobs:
        working-directory: sdks/nodejs-client

    steps:
-      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false

@@ -40,7 +40,7 @@ jobs:

    steps:
      - name: Checkout repository
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          fetch-depth: 0
          token: ${{ secrets.GITHUB_TOKEN }}
@@ -158,7 +158,7 @@ jobs:

      - name: Run Claude Code for Translation Sync
        if: steps.context.outputs.CHANGED_FILES != ''
-        uses: anthropics/claude-code-action@806af32823ef69c8ef357086c573a902af641307 # v1.0.151
+        uses: anthropics/claude-code-action@558b1d6cab4085c7753fe402c10bef0fbb92ac7a # v1.0.165
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          github_token: ${{ secrets.GITHUB_TOKEN }}
@@ -21,7 +21,7 @@ jobs:

    steps:
      - name: Checkout repository
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          fetch-depth: 0

@@ -24,7 +24,7 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false

@@ -36,7 +36,7 @@ jobs:
          remove_tool_cache: true

      - name: Setup UV and Python
-        uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39 # v8.2.0
+        uses: astral-sh/setup-uv@d31148d669074a8d0a63714ba94f3201e7020bc3 # v8.3.0
        with:
          enable-cache: true
          python-version: ${{ matrix.python-version }}
@@ -21,7 +21,7 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false

@@ -33,7 +33,7 @@ jobs:
          remove_tool_cache: true

      - name: Setup UV and Python
-        uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39 # v8.2.0
+        uses: astral-sh/setup-uv@d31148d669074a8d0a63714ba94f3201e7020bc3 # v8.3.0
        with:
          enable-cache: true
          python-version: ${{ matrix.python-version }}
@@ -25,7 +25,7 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false

@@ -33,7 +33,7 @@ jobs:
        uses: ./.github/actions/setup-web

      - name: Setup UV and Python
-        uses: astral-sh/setup-uv@fac544c07dec837d0ccb6301d7b5580bf5edae39 # v8.2.0
+        uses: astral-sh/setup-uv@d31148d669074a8d0a63714ba94f3201e7020bc3 # v8.3.0
        with:
          enable-cache: true
          python-version: "3.12"
@@ -31,7 +31,7 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false

@@ -64,7 +64,7 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false

@@ -102,7 +102,7 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false

@@ -134,7 +134,7 @@ jobs:

    steps:
      - name: Checkout code
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
        with:
          persist-credentials: false

@@ -0,0 +1,51 @@
+from typing import Any, cast
+
+from sqlalchemy import select
+
+from core.app.apps.agent_app.app_variable_projection import agent_app_variables_to_user_input_form
+from core.app.apps.agent_app.errors import AgentAppGeneratorError, AgentAppNotPublishedError
+from extensions.ext_database import db
+from models.agent import Agent, AgentConfigSnapshot, AgentStatus
+from models.agent_config_entities import AgentSoulConfig
+from models.model import App
+
+
+def get_published_agent_app_feature_dict_and_user_input_form(
+    app_model: App,
+) -> tuple[dict[str, Any], list[dict[str, Any]]]:
+    """Return public Agent App parameters backed by the published Agent Soul."""
+    app_model_config = app_model.app_model_config
+    features_dict = cast(dict[str, Any], app_model_config.to_dict()) if app_model_config is not None else {}
+
+    agent_id = app_model.bound_agent_id
+    if not agent_id:
+        raise AgentAppGeneratorError("Agent App has no bound Agent")
+
+    agent = db.session.scalar(
+        select(Agent)
+        .where(
+            Agent.tenant_id == app_model.tenant_id,
+            Agent.id == agent_id,
+            Agent.status == AgentStatus.ACTIVE,
+        )
+        .limit(1)
+    )
+    if agent is None:
+        raise AgentAppGeneratorError("Agent App has no bound Agent")
+    if not agent.active_config_snapshot_id or not agent.active_config_is_published:
+        raise AgentAppNotPublishedError("Agent has not been published")
+
+    snapshot = db.session.scalar(
+        select(AgentConfigSnapshot)
+        .where(
+            AgentConfigSnapshot.tenant_id == app_model.tenant_id,
+            AgentConfigSnapshot.agent_id == agent.id,
+            AgentConfigSnapshot.id == agent.active_config_snapshot_id,
+        )
+        .limit(1)
+    )
+    if snapshot is None:
+        raise AgentAppGeneratorError("Agent published version not found")
+
+    agent_soul = AgentSoulConfig.model_validate(snapshot.config_snapshot_dict)
+    return features_dict, agent_app_variables_to_user_input_form(agent_soul.app_variables)
@@ -2,19 +2,16 @@ from typing import Any, cast

 from flask_restx import Resource
 from pydantic import Field
-from sqlalchemy import select

+from controllers.common.agent_app_parameters import get_published_agent_app_feature_dict_and_user_input_form
 from controllers.common.fields import Parameters
 from controllers.common.schema import register_response_schema_models
 from controllers.service_api import service_api_ns
-from controllers.service_api.app.error import AppUnavailableError
+from controllers.service_api.app.error import AgentNotPublishedError, AppUnavailableError
 from controllers.service_api.wraps import validate_app_token
 from core.app.app_config.common.parameters_mapping import get_parameters_from_feature_dict
-from core.app.apps.agent_app.app_variable_projection import agent_app_variables_to_user_input_form
-from extensions.ext_database import db
+from core.app.apps.agent_app.errors import AgentAppGeneratorError, AgentAppNotPublishedError
 from fields.base import ResponseModel
-from models.agent import Agent, AgentConfigSnapshot, AgentScope, AgentSource, AgentStatus
-from models.agent_config_entities import AgentSoulConfig
 from models.model import App, AppMode
 from services.app_service import AppService

@@ -35,38 +32,13 @@ register_response_schema_models(service_api_ns, Parameters, AppMetaResponse, App


 def _get_agent_app_feature_dict_and_user_input_form(app_model: App) -> tuple[dict[str, Any], list[dict[str, Any]]]:
-    app_model_config = app_model.app_model_config
-    features_dict = cast(dict[str, Any], app_model_config.to_dict()) if app_model_config is not None else {}
-
-    agent = db.session.scalar(
-        select(Agent)
-        .where(
-            Agent.tenant_id == app_model.tenant_id,
-            Agent.app_id == app_model.id,
-            Agent.scope == AgentScope.ROSTER,
-            Agent.source == AgentSource.AGENT_APP,
-            Agent.status == AgentStatus.ACTIVE,
-        )
-        .limit(1)
-    )
-    if agent is None or not agent.active_config_snapshot_id:
+    try:
+        return get_published_agent_app_feature_dict_and_user_input_form(app_model)
+    except AgentAppNotPublishedError:
+        raise AgentNotPublishedError()
+    except AgentAppGeneratorError:
        raise AppUnavailableError()

-    snapshot = db.session.scalar(
-        select(AgentConfigSnapshot)
-        .where(
-            AgentConfigSnapshot.tenant_id == app_model.tenant_id,
-            AgentConfigSnapshot.agent_id == agent.id,
-            AgentConfigSnapshot.id == agent.active_config_snapshot_id,
-        )
-        .limit(1)
-    )
-    if snapshot is None:
-        raise AppUnavailableError()
-
-    agent_soul = AgentSoulConfig.model_validate(snapshot.config_snapshot_dict)
-    return features_dict, agent_app_variables_to_user_input_form(agent_soul.app_variables)
-

@service_api_ns.route("/parameters")
 class AppParameterApi(Resource):
@@ -15,6 +15,7 @@ from controllers.common.schema import register_response_schema_models, register_
 from controllers.console.app.wraps import with_session
 from controllers.service_api import service_api_ns
 from controllers.service_api.app.error import (
+    AgentNotPublishedError,
    AppUnavailableError,
    CompletionRequestError,
    ConversationCompletedError,
@@ -31,6 +32,7 @@ from controllers.service_api.schema import (
 )
 from controllers.service_api.wraps import FetchUserArg, WhereisUserArg, validate_app_token
 from controllers.web.error import InvokeRateLimitError as InvokeRateLimitHttpError
+from core.app.apps.agent_app.errors import AgentAppNotPublishedError
 from core.app.entities.app_invoke_entities import InvokeFrom
 from core.errors.error import (
    ModelCurrentlyNotSupportError,
@@ -248,6 +250,8 @@ class CompletionApi(Resource):
        except services.errors.app_model_config.AppModelConfigBrokenError:
            logger.exception("App model config broken.")
            raise AppUnavailableError()
+        except AgentAppNotPublishedError:
+            raise AgentNotPublishedError()
        except ProviderTokenNotInitError as ex:
            raise ProviderNotInitializeError(ex.description)
        except QuotaExceededError:
@@ -403,6 +407,8 @@ class ChatApi(Resource):
        except services.errors.app_model_config.AppModelConfigBrokenError:
            logger.exception("App model config broken.")
            raise AppUnavailableError()
+        except AgentAppNotPublishedError:
+            raise AgentNotPublishedError()
        except ProviderTokenNotInitError as ex:
            raise ProviderNotInitializeError(ex.description)
        except QuotaExceededError:
@@ -7,6 +7,12 @@ class AppUnavailableError(BaseHTTPException):
    code = 400


+class AgentNotPublishedError(BaseHTTPException):
+    error_code = "agent_not_published"
+    description = "Agent has not been published. Please publish the Agent before using the API."
+    code = 400
+
+
 class NotCompletionAppError(BaseHTTPException):
    error_code = "not_completion_app"
    description = "Please check if your Completion app mode matches the right API route."
@@ -8,8 +8,10 @@ from werkzeug.exceptions import Unauthorized

 from constants import HEADER_NAME_APP_CODE
 from controllers.common import fields
+from controllers.common.agent_app_parameters import get_published_agent_app_feature_dict_and_user_input_form
 from controllers.common.schema import query_params_from_model, register_response_schema_models, register_schema_models
 from core.app.app_config.common.parameters_mapping import get_parameters_from_feature_dict
+from core.app.apps.agent_app.errors import AgentAppGeneratorError, AgentAppNotPublishedError
 from libs.passport import PassportService
 from libs.token import extract_webapp_passport
 from models.model import App, AppMode, EndUser
@@ -19,7 +21,7 @@ from services.feature_service import FeatureService
 from services.webapp_auth_service import WebAppAuthService

 from . import web_ns
-from .error import AppUnavailableError
+from .error import AgentNotPublishedError, AppUnavailableError
 from .wraps import WebApiResource

 logger = logging.getLogger(__name__)
@@ -74,12 +76,21 @@ class AppParameterApi(WebApiResource):
    @web_ns.response(200, "Success", web_ns.models[fields.Parameters.__name__])
    def get(self, app_model: App, end_user: EndUser):
        """Retrieve app parameters."""
-        if app_model.mode in {AppMode.ADVANCED_CHAT, AppMode.WORKFLOW}:
+        features_dict: dict[str, Any]
+        user_input_form: list[dict[str, Any]]
+        if app_model.mode == AppMode.AGENT:
+            try:
+                features_dict, user_input_form = get_published_agent_app_feature_dict_and_user_input_form(app_model)
+            except AgentAppNotPublishedError:
+                raise AgentNotPublishedError()
+            except AgentAppGeneratorError:
+                raise AppUnavailableError()
+        elif app_model.mode in {AppMode.ADVANCED_CHAT, AppMode.WORKFLOW}:
            workflow = app_model.workflow
            if workflow is None:
                raise AppUnavailableError()

-            features_dict: dict[str, Any] = workflow.features_dict
+            features_dict = workflow.features_dict
            user_input_form = workflow.user_input_form(to_old_structure=True)
        else:
            app_model_config = app_model.app_model_config
@@ -11,6 +11,7 @@ from controllers.common.schema import register_response_schema_models, register_
 from controllers.console.app.wraps import with_session
 from controllers.web import web_ns
 from controllers.web.error import (
+    AgentNotPublishedError,
    AppUnavailableError,
    CompletionRequestError,
    ConversationCompletedError,
@@ -22,6 +23,7 @@ from controllers.web.error import (
 )
 from controllers.web.error import InvokeRateLimitError as InvokeRateLimitHttpError
 from controllers.web.wraps import WebApiResource
+from core.app.apps.agent_app.errors import AgentAppNotPublishedError
 from core.app.entities.app_invoke_entities import InvokeFrom
 from core.errors.error import (
    ModelCurrentlyNotSupportError,
@@ -138,6 +140,8 @@ class CompletionApi(WebApiResource):
        except services.errors.app_model_config.AppModelConfigBrokenError:
            logger.exception("App model config broken.")
            raise AppUnavailableError()
+        except AgentAppNotPublishedError:
+            raise AgentNotPublishedError()
        except ProviderTokenNotInitError as ex:
            raise ProviderNotInitializeError(ex.description)
        except QuotaExceededError:
@@ -235,6 +239,8 @@ class ChatApi(WebApiResource):
        except services.errors.app_model_config.AppModelConfigBrokenError:
            logger.exception("App model config broken.")
            raise AppUnavailableError()
+        except AgentAppNotPublishedError:
+            raise AgentNotPublishedError()
        except ProviderTokenNotInitError as ex:
            raise ProviderNotInitializeError(ex.description)
        except QuotaExceededError:
@@ -7,6 +7,12 @@ class AppUnavailableError(BaseHTTPException):
    code = 400


+class AgentNotPublishedError(BaseHTTPException):
+    error_code = "agent_not_published"
+    description = "Agent has not been published. Please publish the Agent before using the web app."
+    code = 400
+
+
 class NotCompletionAppError(BaseHTTPException):
    error_code = "not_completion_app"
    description = "Please check if your Completion app mode matches the right API route."
@@ -61,10 +61,17 @@ _sensitive_word_avoidance_adapter: TypeAdapter[SensitiveWordAvoidanceConfig] = T

 def _normalize_raw(raw: Any) -> Any:
    if isinstance(raw, dict):
-        if raw.get("enabled") is None:
+        enabled = raw.get("enabled")
+        if enabled is None:
            raw = {**raw, "enabled": False}
-        elif raw.get("enabled") is True and raw.get("config") is None:
-            raw = {**raw, "config": {}}
+        elif enabled is True:
+            if raw.get("config") is None:
+                raw = {**raw, "config": {}}
+        else:
+            # enabled is False or any falsy value —
+            # drop extra fields (type, config) so they don't
+            # violate SensitiveWordAvoidanceDisabledConfig.extra="forbid"
+            raw = {"enabled": False}
    return raw


@@ -3,9 +3,9 @@
 An Agent App has no legacy ``app_model_config``: its model / prompt live in the
 bound Agent Soul snapshot. To ride the existing chat message + SSE pipeline we
 synthesize an ``app_model_config``-shaped dict from the Soul (model + system
-prompt) plus any app-level feature flags (opening statement, follow-up, …)
-stored on ``app_model_config`` when present, then reuse the same sub-managers
-the chat app type uses.
+prompt) plus app-level feature flags from Agent Soul, while preserving any
+legacy ``app_model_config`` feature flags when present. Then we reuse the same
+sub-managers the chat app type uses.
 """

 from typing import Any, cast
@@ -79,12 +79,14 @@ class AgentAppConfigManager(BaseAppConfigManager):
    ) -> dict[str, Any]:
        """Shape a Soul + feature flags into an ``app_model_config``-style dict.

-        Feature flags (opening statement / follow-up / tts / stt / citations /
-        moderation / annotation) come from ``app_model_config`` when present
-        (Q3: stored there), otherwise defaults; model + prompt always come from
+        Feature flags come from Agent Soul and fill gaps in the legacy
+        ``app_model_config`` when one exists; model + prompt always come from
        the Agent Soul (the single source of truth for those).
        """
        base: dict[str, Any] = dict(app_model_config.to_dict()) if app_model_config else {}
+        soul_features = agent_soul.app_features.model_dump(mode="json", exclude_none=True)
+        for key, value in soul_features.items():
+            base.setdefault(key, value)

        model = agent_soul.model
        if model is not None:
@@ -32,6 +32,7 @@ from constants import UUID_NIL
 from core.app.app_config.easy_ui_based_app.model_config.converter import ModelConfigConverter
 from core.app.apps.agent_app.app_config_manager import AgentAppConfigManager
 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
@@ -64,10 +65,6 @@ from services.conversation_service import ConversationService
 logger = logging.getLogger(__name__)


-class AgentAppGeneratorError(ValueError):
-    """Raised when an Agent App turn cannot be set up."""
-
-
 def _append_prompt_file_mappings(query: str, prompt_file_mappings: Sequence[JsonValue]) -> str:
    """Append raw request file references to the backend user prompt."""
    if not prompt_file_mappings:
@@ -614,6 +611,8 @@ class AgentAppGenerator(MessageBasedAppGenerator):
                "build_draft" if draft.draft_type == AgentConfigDraftType.DEBUG_BUILD else "draft"
            )
            return agent, draft.id, config_version_kind, agent_soul
+        if not agent.active_config_snapshot_id or not agent.active_config_is_published:
+            raise AgentAppNotPublishedError("Agent has not been published")
        _, snapshot, agent_soul = self._resolve_agent_by_id(
            tenant_id=app_model.tenant_id,
            agent_id=agent.id,
@@ -709,4 +708,4 @@ class AgentAppGenerator(MessageBasedAppGenerator):
        return agent, draft, agent_soul


-__all__ = ["AgentAppGenerator", "AgentAppGeneratorError"]
+__all__ = ["AgentAppGenerator", "AgentAppGeneratorError", "AgentAppNotPublishedError"]
@@ -372,14 +372,14 @@ class _AgentProcessRecorder:
        row = MessageAgentThought(
            message_id=self._message_id,
            message_chain_id=None,
-            thought=thought,
-            tool=tool,
+            thought=thought or "",
+            tool=tool or "",
            tool_labels_str=_tool_labels(tool),
            tool_meta_str="{}",
-            tool_input=tool_input,
-            observation=None,
+            tool_input=tool_input or "",
+            observation="",
            tool_process_data=None,
-            message=None,
+            message="",
            message_token=0,
            message_unit_price=Decimal(0),
            message_price_unit=Decimal("0.001"),
@@ -0,0 +1,6 @@
+class AgentAppGeneratorError(ValueError):
+    """Raised when an Agent App turn cannot be set up."""
+
+
+class AgentAppNotPublishedError(AgentAppGeneratorError):
+    """Raised when a public Agent App runtime is requested before publish."""
@@ -5,6 +5,8 @@ from collections.abc import Generator, Mapping, Sequence
 from mimetypes import guess_extension
 from typing import TYPE_CHECKING, Any, Union

+from sqlalchemy.orm import sessionmaker
+
 from core.app.app_config.entities import ExternalDataVariableEntity, PromptTemplateEntity
 from core.app.apps.base_app_queue_manager import AppQueueManager, PublishFrom
 from core.app.apps.exc import GenerateTaskStoppedError
@@ -423,7 +425,9 @@ class AppRunner:
            _logger.exception("Failed to save image file")
            return

-        # Create MessageFile record
+        # Create MessageFile record.
+        # Use an independent session so this side-effect write does not
+        # commit or close the caller's request-scoped session.
        message_file = MessageFile(
            message_id=message_id,
            type=FileType.IMAGE,
@@ -437,9 +441,8 @@ class AppRunner:
            created_by=user_id,
        )

-        db.session.add(message_file)
-        db.session.commit()
-        db.session.refresh(message_file)
+        with sessionmaker(bind=db.engine, expire_on_commit=False).begin() as session:
+            session.add(message_file)

        # Publish QueueMessageFileEvent
        queue_manager.publish(
@@ -122,7 +122,7 @@ class MessageStreamResponse(StreamResponse):
    event: StreamEvent = StreamEvent.MESSAGE
    id: str
    answer: str
-    from_variable_selector: list[str] | None = None
+    from_variable_selector: list[str] = Field(default_factory=list)


 class MessageAudioStreamResponse(StreamResponse):
@@ -151,7 +151,7 @@ class MessageEndStreamResponse(StreamResponse):
    event: StreamEvent = StreamEvent.MESSAGE_END
    id: str
    metadata: Mapping[str, object] = Field(default_factory=dict)
-    files: Sequence[Mapping[str, Any]] | None = None
+    files: Sequence[Mapping[str, Any]] = Field(default_factory=list)


 class MessageFileStreamResponse(StreamResponse):
@@ -1,6 +1,6 @@
 import logging
 import time
-from collections.abc import Generator
+from collections.abc import Generator, Mapping, Sequence
 from threading import Thread
 from typing import Any, cast

@@ -44,7 +44,7 @@ from core.app.entities.task_entities import (
 )
 from core.app.task_pipeline.based_generate_task_pipeline import BasedGenerateTaskPipeline
 from core.app.task_pipeline.message_cycle_manager import MessageCycleManager
-from core.app.task_pipeline.message_file_utils import prepare_file_dict
+from core.app.task_pipeline.message_file_utils import MessageFileInfoDict, prepare_file_dict
 from core.base.tts import AppGeneratorTTSPublisher, AudioTrunk
 from core.model_manager import ModelInstance
 from core.ops.entities.trace_entity import TraceTaskName
@@ -466,10 +466,10 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline[EasyUIAppGenerat
        :return:
        """
        self._task_state.metadata.usage = self._task_state.llm_result.usage
-        metadata_dict = self._task_state.metadata.model_dump()
+        metadata_dict = self._task_state.metadata.model_dump(exclude_none=True)

        # Fetch files associated with this message
-        files = None
+        files: list[MessageFileInfoDict] = []
        with Session(db.engine, expire_on_commit=False) as session:
            message_files = session.scalars(select(MessageFile).where(MessageFile.message_id == self._message_id)).all()

@@ -492,13 +492,13 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline[EasyUIAppGenerat
                    file_dict = prepare_file_dict(message_file, upload_files_map)
                    files_list.append(file_dict)

-                files = files_list or None
+                files = files_list

        return MessageEndStreamResponse(
            task_id=self._application_generate_entity.task_id,
            id=self._message_id,
            metadata=metadata_dict,
-            files=files,
+            files=cast(Sequence[Mapping[str, Any]], files),
        )

    def _agent_message_to_stream_response(self, answer: str, message_id: str) -> AgentMessageStreamResponse:
@@ -528,11 +528,11 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline[EasyUIAppGenerat
                task_id=self._application_generate_entity.task_id,
                id=agent_thought.id,
                position=agent_thought.position,
-                thought=agent_thought.thought,
-                observation=agent_thought.observation,
-                tool=agent_thought.tool,
+                thought=agent_thought.thought or "",
+                observation=agent_thought.observation or "",
+                tool=agent_thought.tool or "",
                tool_labels=agent_thought.tool_labels,
-                tool_input=agent_thought.tool_input,
+                tool_input=agent_thought.tool_input or "",
                message_files=agent_thought.files,
            )

@@ -257,7 +257,7 @@ class MessageCycleManager:
            task_id=self._application_generate_entity.task_id,
            id=message_id,
            answer=answer,
-            from_variable_selector=from_variable_selector,
+            from_variable_selector=from_variable_selector or [],
            event=event_type or StreamEvent.MESSAGE,
        )

@@ -2,7 +2,7 @@ import logging
 from collections.abc import Sequence

 from sqlalchemy import select, update
-from sqlalchemy.orm import scoped_session
+from sqlalchemy.orm import Session, scoped_session, sessionmaker

 from core.app.apps.base_app_queue_manager import AppQueueManager, PublishFrom
 from core.app.entities.app_invoke_entities import InvokeFrom
@@ -10,6 +10,7 @@ from core.app.entities.queue_entities import QueueRetrieverResourcesEvent
 from core.rag.entities import RetrievalSourceMetadata
 from core.rag.index_processor.constant.index_type import IndexStructureType
 from core.rag.models.document import Document
+from extensions.ext_database import db
 from models.dataset import ChildChunk, DatasetQuery, DocumentSegment
 from models.dataset import Document as DatasetDocument
 from models.enums import CreatorUserRole, DatasetQuerySource
@@ -46,47 +47,52 @@ class DatasetIndexToolCallbackHandler:
            created_by=self._user_id,
        )

-        session.add(dataset_query)
-        session.commit()
+        # Use an independent session so this audit-log side effect does
+        # not commit or close the caller's request-scoped session.
+        with sessionmaker(bind=db.engine, expire_on_commit=False).begin() as independent_session:
+            independent_session.add(dataset_query)

    def on_tool_end(self, documents: list[Document], session: scoped_session):
        """Handle tool end."""
-        for document in documents:
-            if document.metadata is not None:
-                document_id = document.metadata["document_id"]
-                dataset_document_stmt = select(DatasetDocument).where(DatasetDocument.id == document_id)
-                dataset_document = session.scalar(dataset_document_stmt)
-                if not dataset_document:
-                    _logger.warning(
-                        "Expected DatasetDocument record to exist, but none was found, document_id=%s",
-                        document_id,
-                    )
-                    continue
-                if dataset_document.doc_form == IndexStructureType.PARENT_CHILD_INDEX:
-                    child_chunk_stmt = select(ChildChunk).where(
-                        ChildChunk.index_node_id == document.metadata["doc_id"],
-                        ChildChunk.dataset_id == dataset_document.dataset_id,
-                        ChildChunk.document_id == dataset_document.id,
-                    )
-                    child_chunk = session.scalar(child_chunk_stmt)
-                    if child_chunk:
-                        session.execute(
-                            update(DocumentSegment)
-                            .where(DocumentSegment.id == child_chunk.segment_id)
-                            .values(hit_count=DocumentSegment.hit_count + 1)
+        # Use an independent session so hit-count updates do not
+        # interfere with the caller's request-scoped session.
+        with Session(db.engine, expire_on_commit=False) as independent_session:
+            for document in documents:
+                if document.metadata is not None:
+                    document_id = document.metadata["document_id"]
+                    dataset_document_stmt = select(DatasetDocument).where(DatasetDocument.id == document_id)
+                    dataset_document = independent_session.scalar(dataset_document_stmt)
+                    if not dataset_document:
+                        _logger.warning(
+                            "Expected DatasetDocument record to exist, but none was found, document_id=%s",
+                            document_id,
                        )
-                else:
-                    conditions = [DocumentSegment.index_node_id == document.metadata["doc_id"]]
+                        continue
+                    if dataset_document.doc_form == IndexStructureType.PARENT_CHILD_INDEX:
+                        child_chunk_stmt = select(ChildChunk).where(
+                            ChildChunk.index_node_id == document.metadata["doc_id"],
+                            ChildChunk.dataset_id == dataset_document.dataset_id,
+                            ChildChunk.document_id == dataset_document.id,
+                        )
+                        child_chunk = independent_session.scalar(child_chunk_stmt)
+                        if child_chunk:
+                            independent_session.execute(
+                                update(DocumentSegment)
+                                .where(DocumentSegment.id == child_chunk.segment_id)
+                                .values(hit_count=DocumentSegment.hit_count + 1)
+                            )
+                    else:
+                        conditions = [DocumentSegment.index_node_id == document.metadata["doc_id"]]

-                    if "dataset_id" in document.metadata:
-                        conditions.append(DocumentSegment.dataset_id == document.metadata["dataset_id"])
+                        if "dataset_id" in document.metadata:
+                            conditions.append(DocumentSegment.dataset_id == document.metadata["dataset_id"])

-                    # add hit count to document segment
-                    session.execute(
-                        update(DocumentSegment).where(*conditions).values(hit_count=DocumentSegment.hit_count + 1)
-                    )
+                        # add hit count to document segment
+                        independent_session.execute(
+                            update(DocumentSegment).where(*conditions).values(hit_count=DocumentSegment.hit_count + 1)
+                        )

-                session.commit()
+            independent_session.commit()

    # TODO(-LAN-): Improve type check
    def return_retriever_resource_info(self, resource: Sequence[RetrievalSourceMetadata]):
@@ -48,7 +48,7 @@ from clients.agent_backend import (
 )
 from configs import dify_config
 from core.app.entities.app_invoke_entities import DifyRunContext, InvokeFrom
-from core.workflow.system_variables import SystemVariableKey, get_system_text
+from core.workflow.system_variables import SystemVariableKey, get_system_text, get_system_value
 from graphon.file import File, FileTransferMethod
 from graphon.variables.segments import Segment
 from models.agent import Agent, AgentConfigSnapshot, WorkflowAgentNodeBinding
@@ -354,17 +354,22 @@ class WorkflowAgentRuntimeRequestBuilder:
    ) -> str:
        lines: list[str] = []
        query = get_system_text(context.variable_pool, SystemVariableKey.QUERY)
+        uploaded_files = self._summarize_uploaded_workflow_files(context.variable_pool)
        resolved_outputs = self._resolve_previous_node_outputs(
            context.variable_pool,
            node_job.previous_node_output_refs,
        )
-        if not query and not resolved_outputs:
+        if not query and uploaded_files is None and not resolved_outputs:
            return ""

        lines.append("Workflow context loaded for this run:")
        if query:
            lines.append(f"- User query: {query}")

+        if uploaded_files is not None:
+            lines.append("- Uploaded workflow files:")
+            lines.append(f"  - sys.files: {uploaded_files}")
+
        if resolved_outputs:
            lines.append("- Previous node outputs:")
            for item in resolved_outputs:
@@ -373,6 +378,14 @@ class WorkflowAgentRuntimeRequestBuilder:
        lines.append("The above workflow context is run-specific. Do not treat it as Agent Soul or persistent memory.")
        return "\n".join(lines)

+    def _summarize_uploaded_workflow_files(self, variable_pool: VariablePoolReader) -> str | None:
+        files = get_system_value(variable_pool, SystemVariableKey.FILES)
+        if files is None:
+            return None
+        if isinstance(files, list | tuple) and not files:
+            return None
+        return self._summarize_value(files)
+
    def _build_workflow_task_prompt(
        self,
        context: WorkflowAgentRuntimeBuildContext,
@@ -8,7 +8,7 @@ from pydantic import BaseModel, ConfigDict, Field, WithJsonSchema, field_validat

 from core.rag.entities.metadata_entities import ConditionValue, SupportedComparisonOperator
 from core.workflow.file_reference import is_canonical_file_reference
-from graphon.file import FileTransferMethod
+from graphon.file import FileTransferMethod, FileType


 class AgentKnowledgeQueryMode(StrEnum):
@@ -314,8 +314,9 @@ class AgentKnowledgeQueryConfig(BaseModel):

    Agent v2 stores knowledge as explicit ``knowledge.sets`` rather than the
    legacy flat ``datasets`` / ``query_mode`` / ``query_config`` shape. Each
-    set owns its own query policy, so ``user_query`` must carry an explicit
-    ``value`` while ``generated_query`` leaves that value empty.
+    set owns its own query policy. Mode-dependent completeness, such as
+    requiring ``value`` for ``user_query``, is enforced by composer publish
+    validation so draft saves can persist partially configured knowledge sets.
    """

    model_config = ConfigDict(extra="forbid")
@@ -323,12 +324,6 @@ class AgentKnowledgeQueryConfig(BaseModel):
    mode: AgentKnowledgeQueryMode
    value: str | None = None

-    @model_validator(mode="after")
-    def validate_query(self) -> Self:
-        if self.mode == AgentKnowledgeQueryMode.USER_QUERY and not (self.value or "").strip():
-            raise ValueError("knowledge query.value is required for user_query mode")
-        return self
-

 class AgentKnowledgeModelConfig(BaseModel):
    model_config = ConfigDict(extra="forbid")
@@ -356,8 +351,9 @@ class AgentKnowledgeRetrievalConfig(BaseModel):
    """Per-set retrieval policy for Agent v2 knowledge retrieval.

    Retrieval settings now live on each knowledge set instead of one shared
-    flat config. A set may use either ``multiple`` retrieval with ``top_k`` or
-    ``single`` retrieval with a required model config.
+    flat config. Mode-dependent completeness, such as requiring ``top_k`` for
+    ``multiple`` or a model for ``single``, is enforced by composer publish
+    validation so draft saves can persist partially configured knowledge sets.
    """

    model_config = ConfigDict(extra="forbid")
@@ -371,14 +367,6 @@ class AgentKnowledgeRetrievalConfig(BaseModel):
    weights: AgentKnowledgeWeightedScoreConfig | None = None
    model: AgentKnowledgeModelConfig | None = None

-    @model_validator(mode="after")
-    def validate_mode_fields(self) -> Self:
-        if self.mode == "multiple" and self.top_k is None:
-            raise ValueError("knowledge retrieval.top_k is required for multiple mode")
-        if self.mode == "single" and self.model is None:
-            raise ValueError("knowledge retrieval.model is required for single mode")
-        return self
-

 class AgentKnowledgeMetadataCondition(BaseModel):
    model_config = ConfigDict(extra="forbid")
@@ -401,6 +389,8 @@ class AgentKnowledgeMetadataFilteringConfig(BaseModel):
    The Python attribute uses ``metadata_model_config`` for clarity because the
    model belongs to metadata filtering specifically, while the external API and
    generated schema keep the historical ``model_config`` field name via alias.
+    Mode-dependent completeness is enforced by composer publish validation so
+    draft saves can persist partially configured metadata filters.
    """

    model_config = ConfigDict(extra="forbid", populate_by_name=True)
@@ -410,14 +400,6 @@ class AgentKnowledgeMetadataFilteringConfig(BaseModel):
    metadata_model_config: AgentKnowledgeModelConfig | None = Field(default=None, alias="model_config")
    conditions: AgentKnowledgeMetadataConditions | None = None

-    @model_validator(mode="after")
-    def validate_mode_fields(self) -> Self:
-        if self.mode == "automatic" and self.metadata_model_config is None:
-            raise ValueError("metadata_filtering.model_config is required for automatic mode")
-        if self.mode == "manual" and (self.conditions is None or not self.conditions.conditions):
-            raise ValueError("metadata_filtering.conditions is required for manual mode")
-        return self
-

 class AgentKnowledgeSetConfig(BaseModel):
    """One explicit knowledge set in Agent v2.
@@ -547,6 +529,23 @@ class AgentSensitiveWordAvoidanceFeatureConfig(AgentFeatureToggleConfig):
    config: AgentModerationProviderConfig | None = None


+class AgentFileUploadImageFeatureConfig(AgentFeatureToggleConfig):
+    enabled: bool = True
+
+
+class AgentFileUploadFeatureConfig(AgentFeatureToggleConfig):
+    enabled: bool = True
+    allowed_file_extensions: list[str] = Field(default_factory=lambda: ["JPG", "JPEG", "PNG", "GIF", "WEBP", "SVG"])
+    allowed_file_types: list[FileType] = Field(
+        default_factory=lambda: [FileType.DOCUMENT, FileType.IMAGE, FileType.AUDIO, FileType.VIDEO]
+    )
+    allowed_file_upload_methods: list[FileTransferMethod] = Field(
+        default_factory=lambda: [FileTransferMethod.LOCAL_FILE, FileTransferMethod.REMOTE_URL]
+    )
+    image: AgentFileUploadImageFeatureConfig = Field(default_factory=AgentFileUploadImageFeatureConfig)
+    number_limits: int = 3
+
+
 class AgentSoulAppFeaturesConfig(AgentFlexibleConfig):
    opening_statement: str | None = None
    suggested_questions: list[str] | None = None
@@ -555,6 +554,7 @@ class AgentSoulAppFeaturesConfig(AgentFlexibleConfig):
    text_to_speech: AgentTextToSpeechFeatureConfig | None = None
    retriever_resource: AgentFeatureToggleConfig | None = None
    sensitive_word_avoidance: AgentSensitiveWordAvoidanceFeatureConfig | None = None
+    file_upload: AgentFileUploadFeatureConfig = Field(default_factory=AgentFileUploadFeatureConfig)


 class WorkflowPreviousNodeOutputRef(AgentFlexibleConfig):
@@ -13747,6 +13747,23 @@ Stable Agent Soul reference to one normalized skill archive.
 | upload_file_id | string |  | No |
 | url | string |  | No |

+#### AgentFileUploadFeatureConfig
+
+| Name | Type | Description | Required |
+| ---- | ---- | ----------- | -------- |
+| allowed_file_extensions | [ string ] |  | No |
+| allowed_file_types | [ [FileType](#filetype) ] |  | No |
+| allowed_file_upload_methods | [ [FileTransferMethod](#filetransfermethod) ] |  | No |
+| enabled | boolean, <br>**Default:** true |  | No |
+| image | [AgentFileUploadImageFeatureConfig](#agentfileuploadimagefeatureconfig) |  | No |
+| number_limits | integer, <br>**Default:** 3 |  | No |
+
+#### AgentFileUploadImageFeatureConfig
+
+| Name | Type | Description | Required |
+| ---- | ---- | ----------- | -------- |
+| enabled | boolean, <br>**Default:** true |  | No |
+
 #### AgentHumanContactConfig

 | Name | Type | Description | Required |
@@ -13890,6 +13907,8 @@ Per-set metadata filtering policy.
 The Python attribute uses ``metadata_model_config`` for clarity because the
 model belongs to metadata filtering specifically, while the external API and
 generated schema keep the historical ``model_config`` field name via alias.
+Mode-dependent completeness is enforced by composer publish validation so
+draft saves can persist partially configured metadata filters.

 | Name | Type | Description | Required |
 | ---- | ---- | ----------- | -------- |
@@ -13912,8 +13931,9 @@ Per-set query policy for Agent v2 knowledge retrieval.

 Agent v2 stores knowledge as explicit ``knowledge.sets`` rather than the
 legacy flat ``datasets`` / ``query_mode`` / ``query_config`` shape. Each
-set owns its own query policy, so ``user_query`` must carry an explicit
-``value`` while ``generated_query`` leaves that value empty.
+set owns its own query policy. Mode-dependent completeness, such as
+requiring ``value`` for ``user_query``, is enforced by composer publish
+validation so draft saves can persist partially configured knowledge sets.

 | Name | Type | Description | Required |
 | ---- | ---- | ----------- | -------- |
@@ -13938,8 +13958,9 @@ set owns its own query policy, so ``user_query`` must carry an explicit
 Per-set retrieval policy for Agent v2 knowledge retrieval.

 Retrieval settings now live on each knowledge set instead of one shared
-flat config. A set may use either ``multiple`` retrieval with ``top_k`` or
-``single`` retrieval with a required model config.
+flat config. Mode-dependent completeness, such as requiring ``top_k`` for
+``multiple`` or a model for ``single``, is enforced by composer publish
+validation so draft saves can persist partially configured knowledge sets.

 | Name | Type | Description | Required |
 | ---- | ---- | ----------- | -------- |
@@ -14338,6 +14359,7 @@ Visibility and lifecycle scope of an Agent record.

 | Name | Type | Description | Required |
 | ---- | ---- | ----------- | -------- |
+| file_upload | [AgentFileUploadFeatureConfig](#agentfileuploadfeatureconfig) |  | No |
 | opening_statement | string |  | No |
 | retriever_resource | [AgentFeatureToggleConfig](#agentfeaturetoggleconfig) |  | No |
 | sensitive_word_avoidance | [AgentSensitiveWordAvoidanceFeatureConfig](#agentsensitivewordavoidancefeatureconfig) |  | No |
@@ -33,6 +33,7 @@ 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 (
+    AgentModelNotConfiguredError,
    AgentNameConflictError,
    AgentNotFoundError,
    AgentVersionConflictError,
@@ -168,7 +169,8 @@ class AgentComposerService:

        _backfill_cli_tool_ids(payload.agent_soul)
        _validate_composer_payload_for_strategy(payload)
-        cls.validate_knowledge_datasets(tenant_id=tenant_id, agent_soul=payload.agent_soul)
+        if payload.save_strategy in _PUBLISH_SAVE_STRATEGIES:
+            cls.validate_knowledge_datasets(tenant_id=tenant_id, agent_soul=payload.agent_soul)
        workflow = cls._get_draft_workflow(tenant_id=tenant_id, app_id=app_id)
        binding = cls._get_workflow_binding(tenant_id=tenant_id, workflow_id=workflow.id, node_id=node_id)

@@ -357,7 +359,6 @@ class AgentComposerService:
            raise ValueError("agent_soul is required")
        _backfill_cli_tool_ids(payload.agent_soul)
        _validate_composer_payload_for_strategy(payload)
-        cls.validate_knowledge_datasets(tenant_id=tenant_id, agent_soul=payload.agent_soul)

        agent = cls._get_agent_app_agent(tenant_id=tenant_id, app_id=app_id)
        if not agent:
@@ -401,7 +402,6 @@ class AgentComposerService:
            raise ValueError("agent_soul is required")
        _backfill_cli_tool_ids(payload.agent_soul)
        _validate_composer_payload_for_strategy(payload)
-        cls.validate_knowledge_datasets(tenant_id=tenant_id, agent_soul=payload.agent_soul)
        agent = cls._require_agent(tenant_id=tenant_id, agent_id=agent_id)
        return cls._save_agent_composer_for_agent(
            tenant_id=tenant_id,
@@ -511,6 +511,8 @@ class AgentComposerService:
                version_note=version_note,
            )
        )
+        if not agent_soul_has_model(agent_soul):
+            raise AgentModelNotConfiguredError()
        cls.validate_knowledge_datasets(tenant_id=tenant_id, agent_soul=agent_soul)
        version = cls._create_config_version(
            tenant_id=tenant_id,
@@ -591,7 +593,6 @@ class AgentComposerService:
            raise ValueError("agent_soul is required")
        _backfill_cli_tool_ids(payload.agent_soul)
        ComposerConfigValidator.validate_draft_save_payload(payload)
-        cls.validate_knowledge_datasets(tenant_id=tenant_id, agent_soul=payload.agent_soul)
        agent = cls._require_agent(tenant_id=tenant_id, agent_id=agent_id)
        build_draft = cls._save_agent_draft(
            tenant_id=tenant_id,
@@ -3,6 +3,7 @@ from typing import Any

 from pydantic import ValidationError

+from models.agent_config_entities import AgentKnowledgeQueryMode
 from services.agent.errors import AgentSoulLockedError, InvalidComposerConfigError, PlaintextSecretNotAllowedError
 from services.agent.prompt_mentions import (
    MAX_MENTIONS_PER_PROMPT,
@@ -228,9 +229,40 @@ class ComposerConfigValidator:
    @classmethod
    def validate_agent_soul(cls, agent_soul: AgentSoulConfig) -> None:
        dumped = agent_soul.model_dump(mode="json")
+        cls._validate_knowledge_runtime_config(agent_soul)
        cls._reject_plaintext_secrets(dumped, path="agent_soul")
        cls._validate_shell_config(dumped)

+    @classmethod
+    def _validate_knowledge_runtime_config(cls, agent_soul: AgentSoulConfig) -> None:
+        """Validate knowledge settings that are required only for publish/run.
+
+        Draft composer saves must be able to persist partially configured
+        knowledge sets while a user is still editing the panel. These checks
+        stay in the publish validator so invalid runtime configs are still
+        blocked before a version can be published or executed.
+        """
+        for knowledge_set in agent_soul.knowledge.sets:
+            if (
+                knowledge_set.query.mode == AgentKnowledgeQueryMode.USER_QUERY
+                and not (knowledge_set.query.value or "").strip()
+            ):
+                raise InvalidComposerConfigError("knowledge query.value is required for user_query mode")
+
+            retrieval = knowledge_set.retrieval
+            if retrieval.mode == "multiple" and retrieval.top_k is None:
+                raise InvalidComposerConfigError("knowledge retrieval.top_k is required for multiple mode")
+            if retrieval.mode == "single" and retrieval.model is None:
+                raise InvalidComposerConfigError("knowledge retrieval.model is required for single mode")
+
+            metadata_filtering = knowledge_set.metadata_filtering
+            if metadata_filtering.mode == "automatic" and metadata_filtering.metadata_model_config is None:
+                raise InvalidComposerConfigError("metadata_filtering.model_config is required for automatic mode")
+            if metadata_filtering.mode == "manual" and (
+                metadata_filtering.conditions is None or not metadata_filtering.conditions.conditions
+            ):
+                raise InvalidComposerConfigError("metadata_filtering.conditions is required for manual mode")
+
    @classmethod
    def validate_node_job(cls, node_job: WorkflowNodeJobConfig) -> None:
        cls._reject_plaintext_secrets(node_job.model_dump(mode="json"), path="node_job")
@@ -1,5 +1,7 @@
 from werkzeug.exceptions import BadRequest, Conflict, NotFound

+from libs.exception import BaseHTTPException
+

 class AgentNotFoundError(NotFound):
    description = "Agent not found."
@@ -21,6 +23,12 @@ class AgentVersionConflictError(Conflict):
    description = "Agent config version changed. Please reload and try again."


+class AgentModelNotConfiguredError(BaseHTTPException):
+    error_code = "agent_model_not_configured"
+    description = "Agent App requires the Agent Soul model to be configured."
+    code = 400
+
+
 class AgentSoulLockedError(BadRequest):
    description = "Agent Soul is locked for this workflow node."

@@ -24,6 +24,7 @@ from core.trigger.constants import (
    TRIGGER_SCHEDULE_NODE_TYPE,
    TRIGGER_WEBHOOK_NODE_TYPE,
 )
+from core.workflow.nodes.agent_v2.validators import WorkflowAgentNodeValidator
 from core.workflow.nodes.knowledge_retrieval.entities import KnowledgeRetrievalNodeData
 from core.workflow.nodes.trigger_schedule.trigger_schedule_node import TriggerScheduleNode
 from events.app_event import app_model_config_was_updated, app_was_created
@@ -41,7 +42,7 @@ from models.model import AppModelConfig, AppModelConfigDict, IconType
 from models.workflow import Workflow
 from services.dsl_version import check_version_compatibility
 from services.entities.dsl_entities import CheckDependenciesResult, ImportMode, ImportStatus
-from services.errors.app import WorkflowNotFoundError
+from services.errors.app import WorkflowAgentNodeDslExportUnsupportedError, WorkflowNotFoundError
 from services.plugin.dependencies_analysis import DependenciesAnalysisService
 from services.workflow_draft_variable_service import WorkflowDraftVariableService
 from services.workflow_service import WorkflowService
@@ -563,6 +564,11 @@ class AppDslService:
            raise WorkflowNotFoundError("Missing draft workflow configuration, please check.")

        workflow_dict = workflow.to_dict(include_secret=include_secret)
+        if any(WorkflowAgentNodeValidator.iter_agent_v2_nodes(workflow_dict.get("graph", {}))):
+            raise WorkflowAgentNodeDslExportUnsupportedError(
+                "Workflow DSL export does not support Agent nodes yet. Remove Agent nodes before exporting."
+            )
+
        # TODO: refactor: we need a better way to filter workspace related data from nodes
        for node in workflow_dict.get("graph", {}).get("nodes", []):
            node_data = node.get("data", {})
@@ -14,6 +14,15 @@ class WorkflowNotFoundError(Exception):
    pass


+class WorkflowAgentNodeDslExportUnsupportedError(ValueError):
+    """Raised when a workflow DSL export contains Agent v2 nodes.
+
+    Agent v2 node configuration is not fully portable until Agent DSL
+    export/import is supported, because its runtime config lives outside the
+    workflow graph in agent bindings and config snapshots.
+    """
+
+
 class WorkflowIdFormatError(Exception):
    pass

@@ -73,7 +73,7 @@ class TagService:
        target must be bound to all requested tags.
        """
        # Check if tag_ids is not empty to avoid WHERE false condition
-        if not tag_ids or len(tag_ids) == 0:
+        if not tag_ids:
            return []
        # Deduplicate repeated query params so match_all counts each requested tag once.
        requested_tag_ids = list(dict.fromkeys(tag_ids))
@@ -88,7 +88,7 @@ class TagService:
            return []
        tag_ids = list(tags)
        # Check if tag_ids is not empty to avoid WHERE false condition
-        if not tag_ids or len(tag_ids) == 0:
+        if not tag_ids:
            return []
        if match_all:
            if len(tag_ids) != len(requested_tag_ids):
@@ -35,6 +35,7 @@ from controllers.openapi._errors import (
    RecipientSurfaceMismatch,
 )
 from controllers.service_api.app.error import (
+    AgentNotPublishedError,
    AppUnavailableError,
    CompletionRequestError,
    ConversationCompletedError,
@@ -306,6 +307,7 @@ ERROR_MATRIX = [
    (InternalServerError(), 500, "internal_server_error"),
    (BadGateway("x"), 502, "bad_gateway"),
    (AppUnavailableError(), 400, "app_unavailable"),
+    (AgentNotPublishedError(), 400, "agent_not_published"),
    (ConversationCompletedError(), 400, "conversation_completed"),
    (ProviderNotInitializeError(), 400, "provider_not_initialize"),
    (ProviderQuotaExceededError(), 400, "provider_quota_exceeded"),
@@ -9,7 +9,8 @@ import pytest
 from flask import Flask

 from controllers.service_api.app.app import AppInfoApi, AppMetaApi, AppParameterApi
-from controllers.service_api.app.error import AppUnavailableError
+from controllers.service_api.app.error import AgentNotPublishedError, AppUnavailableError
+from core.app.apps.agent_app.errors import AgentAppNotPublishedError
 from models.account import TenantStatus
 from models.model import App, AppMode
 from tests.unit_tests.conftest import setup_mock_tenant_owner_execute_result
@@ -185,6 +186,41 @@ class TestAppParameterApi:
        ]
        mock_get_agent_parameters.assert_called_once_with(mock_app_model)

+    @patch("controllers.service_api.wraps.user_logged_in")
+    @patch("controllers.service_api.wraps.current_app")
+    @patch("controllers.service_api.wraps.validate_and_get_api_token")
+    @patch("controllers.service_api.wraps.db")
+    @patch(
+        "controllers.service_api.app.app.get_published_agent_app_feature_dict_and_user_input_form",
+        side_effect=AgentAppNotPublishedError("Agent has not been published"),
+    )
+    def test_get_parameters_for_unpublished_agent_app_raises_friendly_error(
+        self,
+        mock_get_agent_parameters,
+        mock_db,
+        mock_validate_token,
+        mock_current_app,
+        mock_user_logged_in,
+        app: Flask,
+        mock_app_model,
+    ):
+        _configure_current_app_mock(mock_current_app)
+
+        mock_app_model.mode = AppMode.AGENT
+        mock_api_token = Mock()
+        mock_api_token.app_id = mock_app_model.id
+        mock_api_token.tenant_id = mock_app_model.tenant_id
+        mock_validate_token.return_value = mock_api_token
+
+        mock_tenant = Mock()
+        mock_tenant.status = TenantStatus.NORMAL
+        mock_db.session.get.side_effect = [mock_app_model, mock_tenant]
+        setup_mock_tenant_owner_execute_result(mock_db, mock_tenant, Mock(current_tenant=mock_tenant))
+
+        with app.test_request_context("/parameters", method="GET", headers={"Authorization": "Bearer test_token"}):
+            with pytest.raises(AgentNotPublishedError):
+                AppParameterApi().get()
+
    @patch("controllers.service_api.wraps.user_logged_in")
    @patch("controllers.service_api.wraps.current_app")
    @patch("controllers.service_api.wraps.validate_and_get_api_token")
@@ -31,10 +31,12 @@ from controllers.service_api.app.completion import (
    CompletionStopApi,
 )
 from controllers.service_api.app.error import (
+    AgentNotPublishedError,
    AppUnavailableError,
    ConversationCompletedError,
    NotChatAppError,
 )
+from core.app.apps.agent_app.errors import AgentAppNotPublishedError
 from core.errors.error import QuotaExceededError
 from graphon.model_runtime.errors.invoke import InvokeError
 from models.model import App, AppMode, EndUser
@@ -516,6 +518,22 @@ class TestChatApiController:
            with pytest.raises(BadRequest):
                handler(api, session=Mock(), app_model=app_model, end_user=end_user)

+    def test_agent_not_published_error_mapped(self, app: Flask, monkeypatch: pytest.MonkeyPatch) -> None:
+        monkeypatch.setattr(
+            AppGenerateService,
+            "generate",
+            lambda *_args, **_kwargs: (_ for _ in ()).throw(AgentAppNotPublishedError("Agent has not been published")),
+        )
+
+        api = ChatApi()
+        handler = unwrap(api.post)
+        app_model = SimpleNamespace(mode=AppMode.AGENT.value)
+        end_user = SimpleNamespace()
+
+        with app.test_request_context("/chat-messages", method="POST", json={"inputs": {}, "query": "hi"}):
+            with pytest.raises(AgentNotPublishedError):
+                handler(api, session=Mock(), app_model=app_model, end_user=end_user)
+

 class TestChatStopApiController:
    def test_wrong_mode(self, app: Flask) -> None:
@@ -9,7 +9,8 @@ import pytest
 from flask import Flask

 from controllers.web.app import AppAccessMode, AppMeta, AppParameterApi, AppWebAuthPermission
-from controllers.web.error import AppUnavailableError
+from controllers.web.error import AgentNotPublishedError, AppUnavailableError
+from core.app.apps.agent_app.errors import AgentAppNotPublishedError


 # ---------------------------------------------------------------------------
@@ -80,6 +81,18 @@ class TestAppParameterApi:
            with pytest.raises(AppUnavailableError):
                AppParameterApi().get(app_model, SimpleNamespace())

+    def test_agent_mode_unpublished_raises_friendly_error(self, app: Flask) -> None:
+        app_model = SimpleNamespace(mode="agent")
+        with (
+            app.test_request_context("/parameters"),
+            patch(
+                "controllers.web.app.get_published_agent_app_feature_dict_and_user_input_form",
+                side_effect=AgentAppNotPublishedError("Agent has not been published"),
+            ),
+        ):
+            with pytest.raises(AgentNotPublishedError):
+                AppParameterApi().get(app_model, SimpleNamespace())
+

 # ---------------------------------------------------------------------------
 # AppMeta
@@ -10,6 +10,7 @@ from flask import Flask

 from controllers.web.completion import ChatApi, ChatStopApi, CompletionApi, CompletionStopApi
 from controllers.web.error import (
+    AgentNotPublishedError,
    CompletionRequestError,
    NotChatAppError,
    NotCompletionAppError,
@@ -17,6 +18,7 @@ from controllers.web.error import (
    ProviderNotInitializeError,
    ProviderQuotaExceededError,
 )
+from core.app.apps.agent_app.errors import AgentAppNotPublishedError
 from core.errors.error import ModelCurrentlyNotSupportError, ProviderTokenNotInitError, QuotaExceededError
 from graphon.model_runtime.errors.invoke import InvokeError

@@ -142,6 +144,19 @@ class TestChatApi:
            with pytest.raises(CompletionRequestError):
                ChatApi().post(_chat_app(), _end_user())

+    @patch(
+        "controllers.web.completion.AppGenerateService.generate",
+        side_effect=AgentAppNotPublishedError("Agent has not been published"),
+    )
+    @patch("controllers.web.completion.web_ns")
+    def test_agent_not_published_error_mapped(self, mock_ns: MagicMock, mock_gen: MagicMock, app: Flask) -> None:
+        mock_ns.payload = {"inputs": {}, "query": "x"}
+        app_model = SimpleNamespace(id="app-1", mode="agent")
+
+        with app.test_request_context("/chat-messages", method="POST"):
+            with pytest.raises(AgentNotPublishedError):
+                ChatApi().post(app_model, _end_user())
+

 # ---------------------------------------------------------------------------
 # ChatStopApi
@@ -6,6 +6,7 @@ import pytest

 from controllers.common.errors import InvalidArgumentError, NotFoundError
 from controllers.web.error import (
+    AgentNotPublishedError,
    AppMoreLikeThisDisabledError,
    AppSuggestedQuestionsAfterAnswerDisabledError,
    AppUnavailableError,
@@ -29,6 +30,7 @@ from controllers.web.error import (

 _ERROR_SPECS: list[tuple[type, str, int]] = [
    (AppUnavailableError, "app_unavailable", 400),
+    (AgentNotPublishedError, "agent_not_published", 400),
    (NotCompletionAppError, "not_completion_app", 400),
    (NotChatAppError, "not_chat_app", 400),
    (NotWorkflowAppError, "not_workflow_app", 400),
@@ -65,6 +65,32 @@ def test_missing_soul_model_leaves_no_model_key():
    d = AgentAppConfigManager._synthesize_config_dict(AgentSoulConfig(), None)
    assert "model" not in d
    assert d["pre_prompt"] == ""
+    assert d["file_upload"] == {
+        "allowed_file_extensions": ["JPG", "JPEG", "PNG", "GIF", "WEBP", "SVG"],
+        "allowed_file_types": ["document", "image", "audio", "video"],
+        "allowed_file_upload_methods": ["local_file", "remote_url"],
+        "enabled": True,
+        "image": {"enabled": True},
+        "number_limits": 3,
+    }
+
+
+def test_legacy_app_model_config_file_upload_takes_precedence():
+    fake_amc = SimpleNamespace(
+        to_dict=lambda: {
+            "file_upload": {
+                "enabled": False,
+                "image": {"enabled": False},
+            },
+        }
+    )
+
+    d = AgentAppConfigManager._synthesize_config_dict(AgentSoulConfig(), fake_amc)  # type: ignore[arg-type]
+
+    assert d["file_upload"] == {
+        "enabled": False,
+        "image": {"enabled": False},
+    }


 def test_prompt_type_defaults_to_simple():
@@ -13,7 +13,7 @@ from typing import Any
 import pytest

 from core.app.apps.agent_app import app_generator as gen_mod
-from core.app.apps.agent_app.app_generator import AgentAppGenerator, AgentAppGeneratorError
+from core.app.apps.agent_app.app_generator import AgentAppGenerator, AgentAppGeneratorError, AgentAppNotPublishedError
 from core.app.entities.app_invoke_entities import InvokeFrom

 _SOUL_DICT = {
@@ -78,7 +78,7 @@ class TestResolveAgentById:

 class TestResolveAgent:
    def test_success_chains_to_resolve_by_id(self, monkeypatch: pytest.MonkeyPatch):
-        bound_agent = SimpleNamespace(id="agent-1", active_config_snapshot_id="snap-1")
+        bound_agent = SimpleNamespace(id="agent-1", active_config_snapshot_id="snap-1", active_config_is_published=True)
        inner_agent = SimpleNamespace(id="agent-1")
        snapshot = _snapshot()
        # scalar order: bound agent (in _resolve_agent), then agent + snapshot (in _resolve_agent_by_id)
@@ -97,6 +97,23 @@ class TestResolveAgent:
        assert config_version_kind == "snapshot"
        assert soul.model is not None

+    def test_unpublished_agent_raises_before_model_resolution(self, monkeypatch: pytest.MonkeyPatch):
+        bound_agent = SimpleNamespace(
+            id="agent-1",
+            active_config_snapshot_id="snap-1",
+            active_config_is_published=False,
+        )
+        _patch_session(monkeypatch, [bound_agent])
+        app_model = SimpleNamespace(id="app-1", tenant_id="t1")
+
+        with pytest.raises(AgentAppNotPublishedError, match="not been published"):
+            AgentAppGenerator()._resolve_agent(
+                app_model,
+                invoke_from=InvokeFrom.WEB_APP,
+                draft_type=None,
+                user=SimpleNamespace(id="user-1"),
+            )  # type: ignore[arg-type]
+
    def test_unbound_app_raises(self, monkeypatch: pytest.MonkeyPatch):
        _patch_session(monkeypatch, [None])
        app_model = SimpleNamespace(id="app-1", tenant_id="t1")
@@ -5,7 +5,6 @@ from uuid import uuid4

 import pytest

-from core.app.apps.base_app_queue_manager import PublishFrom
 from core.app.apps.base_app_runner import AppRunner
 from core.app.entities.app_invoke_entities import InvokeFrom
 from core.app.entities.queue_entities import QueueMessageFileEvent
@@ -81,59 +80,55 @@ class TestBaseAppRunnerMultimodal:
                # Setup mock message file
                mock_msg_file_class.return_value = mock_message_file

-                with patch("core.app.apps.base_app_runner.db.session", autospec=True) as mock_session:
-                    mock_session.add = MagicMock()
-                    mock_session.commit = MagicMock()
-                    mock_session.refresh = MagicMock()
+                file_session = MagicMock()
+                mock_session_factory = MagicMock()
+                mock_session_factory.begin.return_value.__enter__ = MagicMock(return_value=file_session)
+                mock_session_factory.begin.return_value.__exit__ = MagicMock(return_value=False)

-                    # Act
-                    # Create a mock runner with the method bound
-                    runner = MagicMock()
+                with patch("core.app.apps.base_app_runner.sessionmaker", return_value=mock_session_factory) as mock_sm:
+                    with patch("core.app.apps.base_app_runner.db") as mock_db:
+                        # Act
+                        runner = MagicMock()
+                        method = AppRunner._handle_multimodal_image_content
+                        runner._handle_multimodal_image_content = lambda *args, **kwargs: method(
+                            runner, *args, **kwargs
+                        )

-                    method = AppRunner._handle_multimodal_image_content
-                    runner._handle_multimodal_image_content = lambda *args, **kwargs: method(runner, *args, **kwargs)
+                        runner._handle_multimodal_image_content(
+                            content=content,
+                            message_id=mock_message_id,
+                            user_id=mock_user_id,
+                            tenant_id=mock_tenant_id,
+                            queue_manager=mock_queue_manager,
+                        )

-                    runner._handle_multimodal_image_content(
-                        content=content,
-                        message_id=mock_message_id,
-                        user_id=mock_user_id,
-                        tenant_id=mock_tenant_id,
-                        queue_manager=mock_queue_manager,
-                    )
+                        # Assert
+                        mock_mgr.create_file_by_url.assert_called_once_with(
+                            user_id=mock_user_id,
+                            tenant_id=mock_tenant_id,
+                            file_url=image_url,
+                            conversation_id=None,
+                        )

-                    # Assert
-                    # Verify tool file was created from URL
-                    mock_mgr.create_file_by_url.assert_called_once_with(
-                        user_id=mock_user_id,
-                        tenant_id=mock_tenant_id,
-                        file_url=image_url,
-                        conversation_id=None,
-                    )
+                        mock_msg_file_class.assert_called_once()
+                        call_kwargs = mock_msg_file_class.call_args[1]
+                        assert call_kwargs["message_id"] == mock_message_id
+                        assert call_kwargs["type"] == FileType.IMAGE
+                        assert call_kwargs["transfer_method"] == FileTransferMethod.TOOL_FILE
+                        assert call_kwargs["belongs_to"] == "assistant"
+                        assert call_kwargs["created_by"] == mock_user_id

-                    # Verify message file was created with correct parameters
-                    mock_msg_file_class.assert_called_once()
-                    call_kwargs = mock_msg_file_class.call_args[1]
-                    assert call_kwargs["message_id"] == mock_message_id
-                    assert call_kwargs["type"] == FileType.IMAGE
-                    assert call_kwargs["transfer_method"] == FileTransferMethod.TOOL_FILE
-                    assert call_kwargs["belongs_to"] == "assistant"
-                    assert call_kwargs["created_by"] == mock_user_id
+                        # Verify independent session was used (not db.session)
+                        mock_sm.assert_called_once_with(bind=mock_db.engine, expire_on_commit=False)
+                        file_session.add.assert_called_once_with(mock_message_file)
+                        mock_db.session.commit.assert_not_called()
+                        mock_db.session.close.assert_not_called()

-                    # Verify database operations
-                    mock_session.add.assert_called_once_with(mock_message_file)
-                    mock_session.commit.assert_called_once()
-                    mock_session.refresh.assert_called_once_with(mock_message_file)
-
-                    # Verify event was published
-                    mock_queue_manager.publish.assert_called_once()
-                    publish_call = mock_queue_manager.publish.call_args
-                    assert isinstance(publish_call[0][0], QueueMessageFileEvent)
-                    assert publish_call[0][0].message_file_id == mock_message_file.id
-                    # publish_from might be passed as positional or keyword argument
-                    assert (
-                        publish_call[0][1] == PublishFrom.APPLICATION_MANAGER
-                        or publish_call.kwargs.get("publish_from") == PublishFrom.APPLICATION_MANAGER
-                    )
+                        # Verify event was published
+                        mock_queue_manager.publish.assert_called_once()
+                        publish_call = mock_queue_manager.publish.call_args
+                        assert isinstance(publish_call[0][0], QueueMessageFileEvent)
+                        assert publish_call[0][0].message_file_id == mock_message_file.id

    def test_handle_multimodal_image_content_with_base64(
        self,
@@ -165,50 +160,44 @@ class TestBaseAppRunnerMultimodal:
            mock_mgr_class.return_value = mock_mgr

            with patch("core.app.apps.base_app_runner.MessageFile", autospec=True) as mock_msg_file_class:
-                # Setup mock message file
                mock_msg_file_class.return_value = mock_message_file

-                with patch("core.app.apps.base_app_runner.db.session", autospec=True) as mock_session:
-                    mock_session.add = MagicMock()
-                    mock_session.commit = MagicMock()
-                    mock_session.refresh = MagicMock()
+                file_session = MagicMock()
+                mock_session_factory = MagicMock()
+                mock_session_factory.begin.return_value.__enter__ = MagicMock(return_value=file_session)
+                mock_session_factory.begin.return_value.__exit__ = MagicMock(return_value=False)

-                    # Act
-                    # Create a mock runner with the method bound
-                    runner = MagicMock()
-                    method = AppRunner._handle_multimodal_image_content
-                    runner._handle_multimodal_image_content = lambda *args, **kwargs: method(runner, *args, **kwargs)
+                with patch("core.app.apps.base_app_runner.sessionmaker", return_value=mock_session_factory):
+                    with patch("core.app.apps.base_app_runner.db") as mock_db:
+                        runner = MagicMock()
+                        method = AppRunner._handle_multimodal_image_content
+                        runner._handle_multimodal_image_content = lambda *args, **kwargs: method(
+                            runner, *args, **kwargs
+                        )

-                    runner._handle_multimodal_image_content(
-                        content=content,
-                        message_id=mock_message_id,
-                        user_id=mock_user_id,
-                        tenant_id=mock_tenant_id,
-                        queue_manager=mock_queue_manager,
-                    )
+                        runner._handle_multimodal_image_content(
+                            content=content,
+                            message_id=mock_message_id,
+                            user_id=mock_user_id,
+                            tenant_id=mock_tenant_id,
+                            queue_manager=mock_queue_manager,
+                        )

-                    # Assert
-                    # Verify tool file was created from base64
-                    mock_mgr.create_file_by_raw.assert_called_once()
-                    call_kwargs = mock_mgr.create_file_by_raw.call_args[1]
-                    assert call_kwargs["user_id"] == mock_user_id
-                    assert call_kwargs["tenant_id"] == mock_tenant_id
-                    assert call_kwargs["conversation_id"] is None
-                    assert "file_binary" in call_kwargs
-                    assert call_kwargs["mimetype"] == "image/png"
-                    assert call_kwargs["filename"].startswith("generated_image")
-                    assert call_kwargs["filename"].endswith(".png")
+                        mock_mgr.create_file_by_raw.assert_called_once()
+                        call_kwargs = mock_mgr.create_file_by_raw.call_args[1]
+                        assert call_kwargs["user_id"] == mock_user_id
+                        assert call_kwargs["tenant_id"] == mock_tenant_id
+                        assert call_kwargs["conversation_id"] is None
+                        assert "file_binary" in call_kwargs
+                        assert call_kwargs["mimetype"] == "image/png"
+                        assert call_kwargs["filename"].startswith("generated_image")
+                        assert call_kwargs["filename"].endswith(".png")

-                    # Verify message file was created
-                    mock_msg_file_class.assert_called_once()
+                        mock_msg_file_class.assert_called_once()
+                        file_session.add.assert_called_once()
+                        mock_db.session.commit.assert_not_called()

-                    # Verify database operations
-                    mock_session.add.assert_called_once()
-                    mock_session.commit.assert_called_once()
-                    mock_session.refresh.assert_called_once()
-
-                    # Verify event was published
-                    mock_queue_manager.publish.assert_called_once()
+                        mock_queue_manager.publish.assert_called_once()

    def test_handle_multimodal_image_content_with_base64_data_uri(
        self,
@@ -238,33 +227,32 @@ class TestBaseAppRunnerMultimodal:
            mock_mgr_class.return_value = mock_mgr

            with patch("core.app.apps.base_app_runner.MessageFile", autospec=True) as mock_msg_file_class:
-                # Setup mock message file
                mock_msg_file_class.return_value = mock_message_file

-                with patch("core.app.apps.base_app_runner.db.session", autospec=True) as mock_session:
-                    mock_session.add = MagicMock()
-                    mock_session.commit = MagicMock()
-                    mock_session.refresh = MagicMock()
+                file_session = MagicMock()
+                mock_session_factory = MagicMock()
+                mock_session_factory.begin.return_value.__enter__ = MagicMock(return_value=file_session)
+                mock_session_factory.begin.return_value.__exit__ = MagicMock(return_value=False)

-                    # Act
-                    # Create a mock runner with the method bound
-                    runner = MagicMock()
-                    method = AppRunner._handle_multimodal_image_content
-                    runner._handle_multimodal_image_content = lambda *args, **kwargs: method(runner, *args, **kwargs)
+                with patch("core.app.apps.base_app_runner.sessionmaker", return_value=mock_session_factory):
+                    with patch("core.app.apps.base_app_runner.db"):
+                        runner = MagicMock()
+                        method = AppRunner._handle_multimodal_image_content
+                        runner._handle_multimodal_image_content = lambda *args, **kwargs: method(
+                            runner, *args, **kwargs
+                        )

-                    runner._handle_multimodal_image_content(
-                        content=content,
-                        message_id=mock_message_id,
-                        user_id=mock_user_id,
-                        tenant_id=mock_tenant_id,
-                        queue_manager=mock_queue_manager,
-                    )
+                        runner._handle_multimodal_image_content(
+                            content=content,
+                            message_id=mock_message_id,
+                            user_id=mock_user_id,
+                            tenant_id=mock_tenant_id,
+                            queue_manager=mock_queue_manager,
+                        )

-                    # Assert - verify that base64 data was extracted correctly (without prefix)
-                    mock_mgr.create_file_by_raw.assert_called_once()
-                    call_kwargs = mock_mgr.create_file_by_raw.call_args[1]
-                    # The base64 data should be decoded, so we check the binary was passed
-                    assert "file_binary" in call_kwargs
+                        mock_mgr.create_file_by_raw.assert_called_once()
+                        call_kwargs = mock_mgr.create_file_by_raw.call_args[1]
+                        assert "file_binary" in call_kwargs

    def test_handle_multimodal_image_content_without_url_or_base64(
        self,
@@ -284,9 +272,7 @@ class TestBaseAppRunnerMultimodal:

        with patch("core.app.apps.base_app_runner.ToolFileManager", autospec=True) as mock_mgr_class:
            with patch("core.app.apps.base_app_runner.MessageFile", autospec=True) as mock_msg_file_class:
-                with patch("core.app.apps.base_app_runner.db.session", autospec=True) as mock_session:
-                    # Act
-                    # Create a mock runner with the method bound
+                with patch("core.app.apps.base_app_runner.db"):
                    runner = MagicMock()
                    method = AppRunner._handle_multimodal_image_content
                    runner._handle_multimodal_image_content = lambda *args, **kwargs: method(runner, *args, **kwargs)
@@ -299,10 +285,8 @@ class TestBaseAppRunnerMultimodal:
                        queue_manager=mock_queue_manager,
                    )

-                    # Assert - should not create any files or publish events
                    mock_mgr_class.assert_not_called()
                    mock_msg_file_class.assert_not_called()
-                    mock_session.add.assert_not_called()
                    mock_queue_manager.publish.assert_not_called()

    def test_handle_multimodal_image_content_with_error(
@@ -322,20 +306,16 @@ class TestBaseAppRunnerMultimodal:
        )

        with patch("core.app.apps.base_app_runner.ToolFileManager", autospec=True) as mock_mgr_class:
-            # Setup mock to raise exception
            mock_mgr = MagicMock()
            mock_mgr.create_file_by_url.side_effect = Exception("Network error")
            mock_mgr_class.return_value = mock_mgr

            with patch("core.app.apps.base_app_runner.MessageFile", autospec=True) as mock_msg_file_class:
-                with patch("core.app.apps.base_app_runner.db.session", autospec=True) as mock_session:
-                    # Act
-                    # Create a mock runner with the method bound
+                with patch("core.app.apps.base_app_runner.db"):
                    runner = MagicMock()
                    method = AppRunner._handle_multimodal_image_content
                    runner._handle_multimodal_image_content = lambda *args, **kwargs: method(runner, *args, **kwargs)

-                    # Should not raise exception, just log it
                    runner._handle_multimodal_image_content(
                        content=content,
                        message_id=mock_message_id,
@@ -344,9 +324,7 @@ class TestBaseAppRunnerMultimodal:
                        queue_manager=mock_queue_manager,
                    )

-                    # Assert - should not create message file or publish event on error
                    mock_msg_file_class.assert_not_called()
-                    mock_session.add.assert_not_called()
                    mock_queue_manager.publish.assert_not_called()

    def test_handle_multimodal_image_content_debugger_mode(
@@ -369,37 +347,36 @@ class TestBaseAppRunnerMultimodal:
        mock_queue_manager.invoke_from = InvokeFrom.DEBUGGER

        with patch("core.app.apps.base_app_runner.ToolFileManager", autospec=True) as mock_mgr_class:
-            # Setup mock tool file manager
            mock_mgr = MagicMock()
            mock_mgr.create_file_by_url.return_value = mock_tool_file
            mock_mgr_class.return_value = mock_mgr

            with patch("core.app.apps.base_app_runner.MessageFile", autospec=True) as mock_msg_file_class:
-                # Setup mock message file
                mock_msg_file_class.return_value = mock_message_file

-                with patch("core.app.apps.base_app_runner.db.session", autospec=True) as mock_session:
-                    mock_session.add = MagicMock()
-                    mock_session.commit = MagicMock()
-                    mock_session.refresh = MagicMock()
+                file_session = MagicMock()
+                mock_session_factory = MagicMock()
+                mock_session_factory.begin.return_value.__enter__ = MagicMock(return_value=file_session)
+                mock_session_factory.begin.return_value.__exit__ = MagicMock(return_value=False)

-                    # Act
-                    # Create a mock runner with the method bound
-                    runner = MagicMock()
-                    method = AppRunner._handle_multimodal_image_content
-                    runner._handle_multimodal_image_content = lambda *args, **kwargs: method(runner, *args, **kwargs)
+                with patch("core.app.apps.base_app_runner.sessionmaker", return_value=mock_session_factory):
+                    with patch("core.app.apps.base_app_runner.db"):
+                        runner = MagicMock()
+                        method = AppRunner._handle_multimodal_image_content
+                        runner._handle_multimodal_image_content = lambda *args, **kwargs: method(
+                            runner, *args, **kwargs
+                        )

-                    runner._handle_multimodal_image_content(
-                        content=content,
-                        message_id=mock_message_id,
-                        user_id=mock_user_id,
-                        tenant_id=mock_tenant_id,
-                        queue_manager=mock_queue_manager,
-                    )
+                        runner._handle_multimodal_image_content(
+                            content=content,
+                            message_id=mock_message_id,
+                            user_id=mock_user_id,
+                            tenant_id=mock_tenant_id,
+                            queue_manager=mock_queue_manager,
+                        )

-                    # Assert - verify created_by_role is ACCOUNT for debugger mode
-                    call_kwargs = mock_msg_file_class.call_args[1]
-                    assert call_kwargs["created_by_role"] == CreatorUserRole.ACCOUNT
+                        call_kwargs = mock_msg_file_class.call_args[1]
+                        assert call_kwargs["created_by_role"] == CreatorUserRole.ACCOUNT

    def test_handle_multimodal_image_content_service_api_mode(
        self,
@@ -421,34 +398,33 @@ class TestBaseAppRunnerMultimodal:
        mock_queue_manager.invoke_from = InvokeFrom.SERVICE_API

        with patch("core.app.apps.base_app_runner.ToolFileManager", autospec=True) as mock_mgr_class:
-            # Setup mock tool file manager
            mock_mgr = MagicMock()
            mock_mgr.create_file_by_url.return_value = mock_tool_file
            mock_mgr_class.return_value = mock_mgr

            with patch("core.app.apps.base_app_runner.MessageFile", autospec=True) as mock_msg_file_class:
-                # Setup mock message file
                mock_msg_file_class.return_value = mock_message_file

-                with patch("core.app.apps.base_app_runner.db.session", autospec=True) as mock_session:
-                    mock_session.add = MagicMock()
-                    mock_session.commit = MagicMock()
-                    mock_session.refresh = MagicMock()
+                file_session = MagicMock()
+                mock_session_factory = MagicMock()
+                mock_session_factory.begin.return_value.__enter__ = MagicMock(return_value=file_session)
+                mock_session_factory.begin.return_value.__exit__ = MagicMock(return_value=False)

-                    # Act
-                    # Create a mock runner with the method bound
-                    runner = MagicMock()
-                    method = AppRunner._handle_multimodal_image_content
-                    runner._handle_multimodal_image_content = lambda *args, **kwargs: method(runner, *args, **kwargs)
+                with patch("core.app.apps.base_app_runner.sessionmaker", return_value=mock_session_factory):
+                    with patch("core.app.apps.base_app_runner.db"):
+                        runner = MagicMock()
+                        method = AppRunner._handle_multimodal_image_content
+                        runner._handle_multimodal_image_content = lambda *args, **kwargs: method(
+                            runner, *args, **kwargs
+                        )

-                    runner._handle_multimodal_image_content(
-                        content=content,
-                        message_id=mock_message_id,
-                        user_id=mock_user_id,
-                        tenant_id=mock_tenant_id,
-                        queue_manager=mock_queue_manager,
-                    )
+                        runner._handle_multimodal_image_content(
+                            content=content,
+                            message_id=mock_message_id,
+                            user_id=mock_user_id,
+                            tenant_id=mock_tenant_id,
+                            queue_manager=mock_queue_manager,
+                        )

-                    # Assert - verify created_by_role is END_USER for service API
-                    call_kwargs = mock_msg_file_class.call_args[1]
-                    assert call_kwargs["created_by_role"] == CreatorUserRole.END_USER
+                        call_kwargs = mock_msg_file_class.call_args[1]
+                        assert call_kwargs["created_by_role"] == CreatorUserRole.END_USER
@@ -721,6 +721,56 @@ class TestEasyUiBasedGenerateTaskPipeline:
        assert response is not None
        assert response.id == "thought"

+    def test_agent_thought_to_stream_response_normalizes_null_display_fields(self, monkeypatch: pytest.MonkeyPatch):
+        conversation = _make_conversation(AppMode.CHAT)
+        message = _make_message()
+
+        pipeline = EasyUIBasedGenerateTaskPipeline(
+            application_generate_entity=_make_entity(ChatAppGenerateEntity, AppMode.CHAT),
+            queue_manager=_FakeQueueManager(),
+            conversation=conversation,
+            message=message,
+            stream=True,
+        )
+
+        agent_thought = _agent_thought()
+        agent_thought.thought = None
+        agent_thought.observation = None
+        agent_thought.tool = None
+        agent_thought.tool_input = None
+        agent_thought.message_files = None
+
+        class _Session:
+            def __init__(self, *args, **kwargs):
+                pass
+
+            def __enter__(self):
+                return self
+
+            def __exit__(self, exc_type, exc, tb):
+                return False
+
+            def scalar(self, *args, **kwargs):
+                return agent_thought
+
+        monkeypatch.setattr(
+            "core.app.task_pipeline.easy_ui_based_generate_task_pipeline.Session",
+            _Session,
+        )
+        monkeypatch.setattr(
+            "core.app.task_pipeline.easy_ui_based_generate_task_pipeline.db",
+            _FakeDb(),
+        )
+
+        response = pipeline._agent_thought_to_stream_response(QueueAgentThoughtEvent(agent_thought_id="thought"))
+
+        assert response is not None
+        assert response.thought == ""
+        assert response.observation == ""
+        assert response.tool == ""
+        assert response.tool_input == ""
+        assert response.model_dump(mode="json")["message_files"] == []
+
    def test_process_routes_to_stream_and_starts_conversation_name_generation(self):
        conversation = _make_conversation(AppMode.CHAT)
        message = _make_message()
@@ -1280,7 +1330,7 @@ class TestEasyUiBasedGenerateTaskPipeline:
        usage_metadata = cast(dict[str, object], response.metadata["usage"])
        assert usage_metadata["prompt_tokens"] == 1

-    def test_record_files_returns_none_when_message_has_no_files(self, monkeypatch: pytest.MonkeyPatch):
+    def test_record_files_returns_empty_list_when_message_has_no_files(self, monkeypatch: pytest.MonkeyPatch):
        conversation = _make_conversation(AppMode.CHAT)
        message = _make_message()
        pipeline = EasyUIBasedGenerateTaskPipeline(
@@ -1316,7 +1366,7 @@ class TestEasyUiBasedGenerateTaskPipeline:

        response = pipeline._message_end_to_stream_response()

-        assert response.files is None
+        assert response.files == []

    def test_record_files_handles_local_fallback_and_tool_url_variants(self, monkeypatch: pytest.MonkeyPatch):
        conversation = _make_conversation(AppMode.CHAT)
@@ -6,7 +6,7 @@ SSE event, which is critical for vision/image chat responses to render correctly

 Test Coverage:
 - Files array populated when MessageFile records exist
- Files array is None when no MessageFile records exist
+- Files array is empty when no MessageFile records exist
 - Correct signed URL generation for LOCAL_FILE transfer method
 - Correct URL handling for REMOTE_URL transfer method
 - Correct URL handling for TOOL_FILE transfer method
@@ -90,7 +90,7 @@ class TestMessageEndStreamResponseFiles:
        return upload_file

    def test_message_end_with_no_files(self, mock_pipeline):
-        """Test that files array is None when no MessageFile records exist."""
+        """Test that files array is empty when no MessageFile records exist."""
        # Arrange
        with (
            patch("core.app.task_pipeline.easy_ui_based_generate_task_pipeline.db") as mock_db,
@@ -108,9 +108,10 @@ class TestMessageEndStreamResponseFiles:

            # Assert
            assert isinstance(result, MessageEndStreamResponse)
-            assert result.files is None
+            assert result.files == []
            assert result.id == mock_pipeline._message_id
            assert result.metadata == {"test": "metadata"}
+            mock_pipeline._task_state.metadata.model_dump.assert_called_once_with(exclude_none=True)

    def test_message_end_with_local_file(self, mock_pipeline, mock_message_file_local, mock_upload_file):
        """Test that files array is populated correctly for LOCAL_FILE transfer method."""
@@ -14,6 +14,9 @@ def mock_queue_manager(mocker: MockerFixture):

@pytest.fixture
 def handler(mock_queue_manager, mocker: MockerFixture):
+    mocker.patch(
+        "core.callback_handler.index_tool_callback_handler.db",
+    )
    return DatasetIndexToolCallbackHandler(
        queue_manager=mock_queue_manager,
        app_id="app-1",
@@ -33,8 +36,18 @@ class TestOnQuery:
        ],
    )
    def test_on_query_success_roles(self, mocker: MockerFixture, mock_queue_manager, invoke_from, expected_role):
-        # Arrange
-        mock_session = mocker.Mock()
+        # Arrange — the caller passes a session, but our fix uses an independent one
+        caller_session = mocker.Mock()
+
+        independent_session = mocker.MagicMock()
+        mock_session_factory = mocker.MagicMock()
+        mock_session_factory.begin.return_value.__enter__ = mocker.MagicMock(return_value=independent_session)
+        mock_session_factory.begin.return_value.__exit__ = mocker.MagicMock(return_value=False)
+        mocker.patch(
+            "core.callback_handler.index_tool_callback_handler.sessionmaker",
+            return_value=mock_session_factory,
+        )
+        mocker.patch("core.callback_handler.index_tool_callback_handler.db")

        handler = DatasetIndexToolCallbackHandler(
            queue_manager=mock_queue_manager,
@@ -46,17 +59,28 @@ class TestOnQuery:

        handler._invoke_from = invoke_from

-        # Act
-        handler.on_query("test query", "dataset-1", mock_session)
+        # Act — pass caller_session as required by signature
+        handler.on_query("test query", "dataset-1", caller_session)

-        # Assert
-        mock_session.add.assert_called_once()
-        dataset_query = mock_session.add.call_args.args[0]
+        # Assert — independent session used, not the caller's session
+        independent_session.add.assert_called_once()
+        dataset_query = independent_session.add.call_args.args[0]
        assert dataset_query.created_by_role == expected_role
-        mock_session.commit.assert_called_once()
+        caller_session.add.assert_not_called()
+        caller_session.commit.assert_not_called()

    def test_on_query_none_values(self, mocker: MockerFixture, mock_queue_manager):
-        mock_session = mocker.Mock()
+        caller_session = mocker.Mock()
+
+        independent_session = mocker.MagicMock()
+        mock_session_factory = mocker.MagicMock()
+        mock_session_factory.begin.return_value.__enter__ = mocker.MagicMock(return_value=independent_session)
+        mock_session_factory.begin.return_value.__exit__ = mocker.MagicMock(return_value=False)
+        mocker.patch(
+            "core.callback_handler.index_tool_callback_handler.sessionmaker",
+            return_value=mock_session_factory,
+        )
+        mocker.patch("core.callback_handler.index_tool_callback_handler.db")

        handler = DatasetIndexToolCallbackHandler(
            queue_manager=mock_queue_manager,
@@ -66,40 +90,67 @@ class TestOnQuery:
            invoke_from=None,
        )

-        handler.on_query(None, None, mock_session)
+        handler.on_query(None, None, caller_session)

-        mock_session.add.assert_called_once()
-        mock_session.commit.assert_called_once()
+        independent_session.add.assert_called_once()
+        caller_session.add.assert_not_called()


 class TestOnToolEnd:
    def test_on_tool_end_no_metadata(self, handler: DatasetIndexToolCallbackHandler, mocker: MockerFixture):
-        mock_session = mocker.Mock()
+        caller_session = mocker.Mock()
+
+        independent_session = mocker.MagicMock()
+        mocker.patch(
+            "core.callback_handler.index_tool_callback_handler.Session",
+            return_value=independent_session,
+        )
+        independent_session.__enter__ = mocker.MagicMock(return_value=independent_session)
+        independent_session.__exit__ = mocker.MagicMock(return_value=False)

        document = mocker.Mock()
        document.metadata = None

-        handler.on_tool_end([document], mock_session)
+        handler.on_tool_end([document], caller_session)

-        mock_session.commit.assert_not_called()
+        independent_session.commit.assert_called_once()
+        independent_session.execute.assert_not_called()
+        caller_session.commit.assert_not_called()

    def test_on_tool_end_dataset_document_not_found(
        self, handler: DatasetIndexToolCallbackHandler, mocker: MockerFixture
    ):
-        mock_session = mocker.Mock()
-        mock_session.scalar.return_value = None
+        caller_session = mocker.Mock()
+
+        independent_session = mocker.MagicMock()
+        mocker.patch(
+            "core.callback_handler.index_tool_callback_handler.Session",
+            return_value=independent_session,
+        )
+        independent_session.__enter__ = mocker.MagicMock(return_value=independent_session)
+        independent_session.__exit__ = mocker.MagicMock(return_value=False)
+        independent_session.scalar.return_value = None

        document = mocker.Mock()
        document.metadata = {"document_id": "doc-1", "doc_id": "node-1"}

-        handler.on_tool_end([document], mock_session)
+        handler.on_tool_end([document], caller_session)

-        mock_session.scalar.assert_called_once()
+        independent_session.scalar.assert_called_once()
+        caller_session.scalar.assert_not_called()

    def test_on_tool_end_parent_child_index_with_child(
        self, handler: DatasetIndexToolCallbackHandler, mocker: MockerFixture
    ):
-        mock_session = mocker.Mock()
+        caller_session = mocker.Mock()
+
+        independent_session = mocker.MagicMock()
+        mocker.patch(
+            "core.callback_handler.index_tool_callback_handler.Session",
+            return_value=independent_session,
+        )
+        independent_session.__enter__ = mocker.MagicMock(return_value=independent_session)
+        independent_session.__exit__ = mocker.MagicMock(return_value=False)

        mock_dataset_doc = mocker.Mock()
        from core.callback_handler.index_tool_callback_handler import IndexStructureType
@@ -111,23 +162,32 @@ class TestOnToolEnd:
        mock_child_chunk = mocker.Mock()
        mock_child_chunk.segment_id = "segment-1"

-        mock_session.scalar.side_effect = [mock_dataset_doc, mock_child_chunk]
+        independent_session.scalar.side_effect = [mock_dataset_doc, mock_child_chunk]

        document = mocker.Mock()
        document.metadata = {"document_id": "doc-1", "doc_id": "node-1"}

-        handler.on_tool_end([document], mock_session)
+        handler.on_tool_end([document], caller_session)

-        mock_session.execute.assert_called_once()
-        mock_session.commit.assert_called_once()
+        independent_session.execute.assert_called_once()
+        independent_session.commit.assert_called_once()
+        caller_session.execute.assert_not_called()

    def test_on_tool_end_non_parent_child_index(self, handler: DatasetIndexToolCallbackHandler, mocker: MockerFixture):
-        mock_session = mocker.Mock()
+        caller_session = mocker.Mock()
+
+        independent_session = mocker.MagicMock()
+        mocker.patch(
+            "core.callback_handler.index_tool_callback_handler.Session",
+            return_value=independent_session,
+        )
+        independent_session.__enter__ = mocker.MagicMock(return_value=independent_session)
+        independent_session.__exit__ = mocker.MagicMock(return_value=False)

        mock_dataset_doc = mocker.Mock()
        mock_dataset_doc.doc_form = "OTHER"

-        mock_session.scalar.return_value = mock_dataset_doc
+        independent_session.scalar.return_value = mock_dataset_doc

        document = mocker.Mock()
        document.metadata = {
@@ -136,14 +196,24 @@ class TestOnToolEnd:
            "dataset_id": "dataset-1",
        }

-        handler.on_tool_end([document], mock_session)
+        handler.on_tool_end([document], caller_session)

-        mock_session.execute.assert_called_once()
-        mock_session.commit.assert_called_once()
+        independent_session.execute.assert_called_once()
+        independent_session.commit.assert_called_once()
+        caller_session.execute.assert_not_called()

    def test_on_tool_end_empty_documents(self, handler: DatasetIndexToolCallbackHandler, mocker: MockerFixture):
-        mock_session = mocker.Mock()
-        handler.on_tool_end([], mock_session)
+        caller_session = mocker.Mock()
+
+        independent_session = mocker.MagicMock()
+        mocker.patch(
+            "core.callback_handler.index_tool_callback_handler.Session",
+            return_value=independent_session,
+        )
+        independent_session.__enter__ = mocker.MagicMock(return_value=independent_session)
+        independent_session.__exit__ = mocker.MagicMock(return_value=False)
+
+        handler.on_tool_end([], caller_session)


 class TestReturnRetrieverResourceInfo:
@@ -227,6 +227,15 @@ def _previous_node_prompt_payload(result, selector: str) -> object:
    raise AssertionError(f"missing prompt payload for {selector}")


+def _uploaded_workflow_files_prompt_payload(result) -> object:
+    prefix = "  - sys.files: "
+    user_prompt = _workflow_user_prompt(result)
+    for line in user_prompt.splitlines():
+        if line.startswith(prefix):
+            return json.loads(line.removeprefix(prefix))
+    raise AssertionError("missing prompt payload for sys.files")
+
+
 def test_builds_create_run_request_from_agent_soul_and_node_job():
    result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(_context())

@@ -1252,6 +1261,48 @@ def test_previous_node_file_array_uses_agent_stub_download_mappings_in_workflow_
    ]


+def test_uploaded_workflow_files_are_included_without_prompt_marker():
+    file_reference = build_file_reference(record_id="uploaded-file-1")
+
+    class UploadedFilesVariablePool(FakeVariablePool):
+        def get(self, selector):
+            if list(selector) == ["sys", "files"]:
+                return ArrayFileSegment(
+                    value=[
+                        File(
+                            type=FileType.DOCUMENT,
+                            transfer_method=FileTransferMethod.LOCAL_FILE,
+                            reference=file_reference,
+                            remote_url=None,
+                            filename="requirements.pdf",
+                            extension=".pdf",
+                            mime_type="application/pdf",
+                            size=12,
+                        )
+                    ]
+                )
+            return super().get(selector)
+
+    context = replace(_context(), variable_pool=UploadedFilesVariablePool())
+    context.binding.node_job_config = WorkflowNodeJobConfig.model_validate(
+        {
+            "workflow_prompt": "Answer the user's question.",
+        }
+    )
+
+    result = WorkflowAgentRuntimeRequestBuilder(credentials_provider=FakeCredentialsProvider()).build(context)
+
+    user_prompt = _workflow_user_prompt(result)
+    assert "- Uploaded workflow files:" in user_prompt
+    assert _uploaded_workflow_files_prompt_payload(result) == [
+        {
+            "transfer_method": "local_file",
+            "reference": file_reference,
+        }
+    ]
+    assert "Previous node outputs:" not in user_prompt
+
+
 def test_previous_node_remote_url_file_mapping_is_not_truncated_in_workflow_context():
    remote_url = "https://example.com/" + ("a" * 2100) + ".pdf"

@@ -14,6 +14,23 @@ from services.entities.agent_entities import (
 )


+def test_default_agent_soul_enables_file_upload_feature():
+    agent_soul = AgentSoulConfig()
+
+    file_upload = agent_soul.model_dump(mode="json")["app_features"]["file_upload"]
+    assert file_upload == {
+        "allowed_file_extensions": ["JPG", "JPEG", "PNG", "GIF", "WEBP", "SVG"],
+        "allowed_file_types": ["document", "image", "audio", "video"],
+        "allowed_file_upload_methods": ["local_file", "remote_url"],
+        "enabled": True,
+        "image": {"enabled": True},
+        "number_limits": 3,
+    }
+    # The product default should be visible in API responses, but it must not
+    # make workflow-only payload validation treat app_features as user-authored.
+    assert bool(agent_soul.app_features) is False
+
+
 def test_workflow_variant_rejects_agent_app_only_fields():
    with pytest.raises(ValueError):
        ComposerSavePayload.model_validate(
@@ -257,6 +274,16 @@ def test_knowledge_query_mode_uses_stable_backend_enums():
            },
            "knowledge set dataset ids must be unique",
        ),
+    ],
+)
+def test_knowledge_sets_contract_rejects_invalid_configs(knowledge_payload, match: str):
+    with pytest.raises(ValidationError, match=match):
+        AgentSoulConfig.model_validate({"knowledge": knowledge_payload})
+
+
+@pytest.mark.parametrize(
+    ("knowledge_payload", "match"),
+    [
        (
            {
                "sets": [
@@ -317,9 +344,25 @@ def test_knowledge_query_mode_uses_stable_backend_enums():
        ),
    ],
 )
-def test_knowledge_sets_contract_rejects_invalid_configs(knowledge_payload, match: str):
-    with pytest.raises(ValidationError, match=match):
-        AgentSoulConfig.model_validate({"knowledge": knowledge_payload})
+def test_knowledge_runtime_requirements_block_publish_but_not_draft_save(knowledge_payload, match: str):
+    draft_payload = ComposerSavePayload.model_validate(
+        {
+            "variant": ComposerVariant.AGENT_APP,
+            "save_strategy": ComposerSaveStrategy.SAVE_TO_CURRENT_VERSION,
+            "agent_soul": {"knowledge": knowledge_payload},
+        }
+    )
+    ComposerConfigValidator.validate_draft_save_payload(draft_payload)
+
+    publish_payload = ComposerSavePayload.model_validate(
+        {
+            "variant": ComposerVariant.AGENT_APP,
+            "save_strategy": ComposerSaveStrategy.SAVE_AS_NEW_VERSION,
+            "agent_soul": {"knowledge": knowledge_payload},
+        }
+    )
+    with pytest.raises(InvalidComposerConfigError, match=match):
+        ComposerConfigValidator.validate_publish_payload(publish_payload)


 def test_agent_soul_model_config_is_first_class_without_credentials():
@@ -36,6 +36,7 @@ from services.agent.agent_soul_state import agent_soul_has_model
 from services.agent.composer_service import AgentComposerService
 from services.agent.composer_validator import ComposerConfigValidator
 from services.agent.errors import (
+    AgentModelNotConfiguredError,
    AgentNameConflictError,
    AgentNotFoundError,
    AgentVersionConflictError,
@@ -576,6 +577,55 @@ def test_save_agent_app_composer_keeps_published_when_draft_matches_active_snaps
    assert fake_session.commits == 1


+def test_publish_agent_app_draft_rejects_missing_model(monkeypatch: pytest.MonkeyPatch):
+    agent = Agent(
+        id="agent-1",
+        tenant_id="tenant-1",
+        name="Iris",
+        description="",
+        agent_kind=AgentKind.DIFY_AGENT,
+        scope=AgentScope.ROSTER,
+        source=AgentSource.AGENT_APP,
+        status=AgentStatus.ACTIVE,
+        active_config_snapshot_id="version-1",
+        active_config_is_published=False,
+    )
+    draft = AgentConfigDraft(
+        tenant_id="tenant-1",
+        agent_id="agent-1",
+        draft_type=AgentConfigDraftType.DRAFT,
+        draft_owner_key="",
+        base_snapshot_id="version-1",
+        config_snapshot=AgentSoulConfig(),
+    )
+    fake_session = FakeSession(scalar=[agent, draft])
+
+    def fail_create_config_version(**_kwargs):
+        raise AssertionError("config version must not be created when Agent Soul has no model")
+
+    def fail_validate_knowledge_datasets(**_kwargs):
+        raise AssertionError("knowledge datasets must not be validated when Agent Soul has no model")
+
+    monkeypatch.setattr(composer_service.db, "session", fake_session)
+    monkeypatch.setattr(composer_service.ComposerConfigValidator, "validate_publish_payload", lambda payload: None)
+    monkeypatch.setattr(AgentComposerService, "validate_knowledge_datasets", fail_validate_knowledge_datasets)
+    monkeypatch.setattr(AgentComposerService, "_create_config_version", fail_create_config_version)
+
+    with pytest.raises(AgentModelNotConfiguredError) as exc_info:
+        AgentComposerService.publish_agent_app_draft(
+            tenant_id="tenant-1",
+            agent_id="agent-1",
+            account_id="account-1",
+            version_note="ship it",
+        )
+
+    assert exc_info.value.error_code == "agent_model_not_configured"
+    assert agent.active_config_snapshot_id == "version-1"
+    assert agent.active_config_is_published is False
+    assert draft.base_snapshot_id == "version-1"
+    assert fake_session.commits == 0
+
+
 def test_publish_agent_app_draft_creates_published_snapshot(monkeypatch: pytest.MonkeyPatch):
    agent = Agent(
        id="agent-1",
@@ -4257,31 +4307,7 @@ def test_dataset_rows_filters_malformed_ids(monkeypatch: pytest.MonkeyPatch):
    assert captured == {}


-@pytest.mark.parametrize(
-    ("variant", "save_call"),
-    [
-        (
-            ComposerVariant.AGENT_APP,
-            lambda payload: AgentComposerService.save_agent_app_composer(
-                tenant_id="tenant-1",
-                app_id="app-1",
-                account_id="account-1",
-                payload=payload,
-            ),
-        ),
-        (
-            ComposerVariant.WORKFLOW,
-            lambda payload: AgentComposerService.save_workflow_composer(
-                tenant_id="tenant-1",
-                app_id="app-1",
-                node_id="node-1",
-                account_id="account-1",
-                payload=payload,
-            ),
-        ),
-    ],
-)
-def test_composer_save_rejects_malformed_knowledge_dataset_ids(monkeypatch: pytest.MonkeyPatch, variant, save_call):
+def test_validate_knowledge_datasets_rejects_malformed_ids_without_dataset_lookup(monkeypatch: pytest.MonkeyPatch):
    captured = {"calls": 0}

    def fake_get_datasets_by_ids(ids, tenant_id):
@@ -4294,60 +4320,29 @@ def test_composer_save_rejects_malformed_knowledge_dataset_ids(monkeypatch: pyte

    monkeypatch.setattr(dataset_service_module.DatasetService, "get_datasets_by_ids", fake_get_datasets_by_ids)

-    payload = ComposerSavePayload.model_validate(
+    agent_soul = AgentSoulConfig.model_validate(
        {
-            "variant": variant.value,
-            "save_strategy": ComposerSaveStrategy.SAVE_TO_CURRENT_VERSION.value,
-            "soul_lock": {"locked": False},
-            "agent_soul": {
-                "knowledge": {
-                    "sets": [
-                        {
-                            "id": "support",
-                            "name": "Support KB",
-                            "datasets": [{"id": "not-a-uuid"}],
-                            "query": {"mode": "generated_query"},
-                            "retrieval": {"mode": "multiple", "top_k": 4},
-                        }
-                    ]
-                }
+            "knowledge": {
+                "sets": [
+                    {
+                        "id": "support",
+                        "name": "Support KB",
+                        "datasets": [{"id": "not-a-uuid"}],
+                        "query": {"mode": "generated_query"},
+                        "retrieval": {"mode": "multiple", "top_k": 4},
+                    }
+                ]
            },
        }
    )

    with pytest.raises(InvalidComposerConfigError, match="not-a-uuid"):
-        save_call(payload)
+        AgentComposerService.validate_knowledge_datasets(tenant_id="tenant-1", agent_soul=agent_soul)

    assert captured == {"calls": 0}


-@pytest.mark.parametrize(
-    ("variant", "save_call"),
-    [
-        (
-            ComposerVariant.AGENT_APP,
-            lambda payload: AgentComposerService.save_agent_app_composer(
-                tenant_id="tenant-1",
-                app_id="app-1",
-                account_id="account-1",
-                payload=payload,
-            ),
-        ),
-        (
-            ComposerVariant.WORKFLOW,
-            lambda payload: AgentComposerService.save_workflow_composer(
-                tenant_id="tenant-1",
-                app_id="app-1",
-                node_id="node-1",
-                account_id="account-1",
-                payload=payload,
-            ),
-        ),
-    ],
-)
-def test_composer_save_rejects_missing_or_out_of_scope_knowledge_datasets(
-    monkeypatch: pytest.MonkeyPatch, variant, save_call
-):
+def test_validate_knowledge_datasets_rejects_missing_or_out_of_scope_datasets(monkeypatch: pytest.MonkeyPatch):
    captured = {}
    missing_dataset_id = "550e8400-e29b-41d4-a716-446655440000"

@@ -4360,20 +4355,70 @@ def test_composer_save_rejects_missing_or_out_of_scope_knowledge_datasets(

    monkeypatch.setattr(dataset_service_module.DatasetService, "get_datasets_by_ids", fake_get_datasets_by_ids)

+    agent_soul = AgentSoulConfig.model_validate(
+        {
+            "knowledge": {
+                "sets": [
+                    {
+                        "id": "support",
+                        "name": "Support KB",
+                        "datasets": [{"id": missing_dataset_id}],
+                        "query": {"mode": "generated_query"},
+                        "retrieval": {"mode": "multiple", "top_k": 4},
+                    }
+                ]
+            },
+        }
+    )
+
+    with pytest.raises(InvalidComposerConfigError, match=missing_dataset_id):
+        AgentComposerService.validate_knowledge_datasets(tenant_id="tenant-1", agent_soul=agent_soul)
+
+    assert captured == {"ids": [missing_dataset_id], "tenant_id": "tenant-1"}
+
+
+def test_save_agent_composer_allows_incomplete_knowledge_draft(monkeypatch: pytest.MonkeyPatch):
+    agent = SimpleNamespace(
+        id="agent-1",
+        source=AgentSource.AGENT_APP,
+        active_config_snapshot_id="version-1",
+        active_config_is_published=True,
+        updated_by=None,
+    )
+    active_version = SimpleNamespace(config_snapshot_dict=AgentSoulConfig().model_dump(mode="json"))
+    fake_session = FakeSession(scalar=[agent])
+    saved = {}
+
+    import services.dataset_service as dataset_service_module
+
+    monkeypatch.setattr(composer_service.db, "session", fake_session)
+    monkeypatch.setattr(
+        dataset_service_module.DatasetService,
+        "get_datasets_by_ids",
+        lambda *_args, **_kwargs: (_ for _ in ()).throw(AssertionError("draft save must skip dataset lookup")),
+    )
+    monkeypatch.setattr(
+        AgentComposerService,
+        "_save_agent_draft",
+        lambda **kwargs: saved.update(kwargs) or SimpleNamespace(id="draft-1"),
+    )
+    monkeypatch.setattr(AgentComposerService, "_get_version_if_present", lambda **_kwargs: active_version)
+    monkeypatch.setattr(AgentComposerService, "load_agent_composer", lambda **_kwargs: {"loaded": True})
+
    payload = ComposerSavePayload.model_validate(
        {
-            "variant": variant.value,
+            "variant": ComposerVariant.AGENT_APP.value,
            "save_strategy": ComposerSaveStrategy.SAVE_TO_CURRENT_VERSION.value,
-            "soul_lock": {"locked": False},
            "agent_soul": {
                "knowledge": {
                    "sets": [
                        {
                            "id": "support",
                            "name": "Support KB",
-                            "datasets": [{"id": missing_dataset_id}],
+                            "datasets": [{"id": "not-a-uuid"}],
                            "query": {"mode": "generated_query"},
-                            "retrieval": {"mode": "multiple", "top_k": 4},
+                            "retrieval": {"mode": "single"},
+                            "metadata_filtering": {"mode": "automatic"},
                        }
                    ]
                }
@@ -4381,10 +4426,20 @@ def test_composer_save_rejects_missing_or_out_of_scope_knowledge_datasets(
        }
    )

-    with pytest.raises(InvalidComposerConfigError, match=missing_dataset_id):
-        save_call(payload)
+    result = AgentComposerService.save_agent_composer(
+        tenant_id="tenant-1",
+        agent_id="agent-1",
+        account_id="account-1",
+        payload=payload,
+    )

-    assert captured == {"ids": [missing_dataset_id], "tenant_id": "tenant-1"}
+    assert result["loaded"] is True
+    assert saved["draft_type"] == AgentConfigDraftType.DRAFT
+    assert saved["agent_soul"].knowledge.sets[0].retrieval.mode == "single"
+    assert saved["agent_soul"].knowledge.sets[0].retrieval.model is None
+    assert saved["agent_soul"].knowledge.sets[0].metadata_filtering.mode == "automatic"
+    assert saved["agent_soul"].knowledge.sets[0].metadata_filtering.metadata_model_config is None
+    assert fake_session.commits == 1


 def test_workspace_dify_tools_returns_provider_and_tool_granularities(monkeypatch: pytest.MonkeyPatch):
@@ -0,0 +1,42 @@
+from types import SimpleNamespace
+from typing import cast
+from unittest.mock import Mock
+
+import pytest
+
+from graphon.enums import BuiltinNodeTypes
+from models import App
+from services import app_dsl_service
+from services.app_dsl_service import AppDslService
+from services.errors.app import WorkflowAgentNodeDslExportUnsupportedError
+
+
+def test_append_workflow_export_data_rejects_agent_v2_nodes(monkeypatch: pytest.MonkeyPatch) -> None:
+    workflow = SimpleNamespace(
+        to_dict=lambda *, include_secret: {
+            "graph": {
+                "nodes": [
+                    {
+                        "id": "agent-node",
+                        "data": {
+                            "type": BuiltinNodeTypes.AGENT,
+                            "version": "2",
+                        },
+                    }
+                ]
+            }
+        }
+    )
+    workflow_service = Mock()
+    workflow_service.get_draft_workflow.return_value = workflow
+    monkeypatch.setattr(app_dsl_service, "WorkflowService", lambda: workflow_service)
+
+    with pytest.raises(
+        WorkflowAgentNodeDslExportUnsupportedError,
+        match="Workflow DSL export does not support Agent nodes yet.",
+    ):
+        AppDslService._append_workflow_export_data(
+            export_data={},
+            app_model=cast(App, SimpleNamespace(tenant_id="tenant-1")),
+            include_secret=False,
+        )
@@ -37,10 +37,7 @@ from pydantic_ai.exceptions import UnexpectedModelBehavior
 from pydantic_ai.messages import (
    AudioUrl,
    BinaryContent,
-    BuiltinToolCallPart,
-    BuiltinToolReturnPart,
    CachePoint,
-    CompactionPart,
    DocumentUrl,
    FilePart,
    FinishReason,
@@ -357,10 +354,8 @@ def _map_model_response_to_prompt_message(
                    ),
                )
            )
-        elif isinstance(part, BuiltinToolCallPart | BuiltinToolReturnPart | CompactionPart):
-            raise UnexpectedModelBehavior(f"Unsupported response part for daemon adapter: {type(part).__name__}")
        else:
-            assert_never(part)
+            raise UnexpectedModelBehavior(f"Unsupported response part for daemon adapter: {type(part).__name__}")

    content = _normalize_prompt_content(content_parts)
    if content is None and not tool_calls:
@@ -487,10 +482,16 @@ def _map_binary_content_to_prompt_content(
 def _normalize_prompt_content(
    content: list[PromptMessageContentUnionTypes],
 ) -> str | list[PromptMessageContentUnionTypes] | None:
+    """Collapse text-only daemon message content to the string form.
+
+    The daemon protocol supports content-part lists for multimodal messages, but
+    text-only history is safer as plain text because provider plugins commonly
+    JSON-encode text payloads without Graphon model encoders.
+    """
    if not content:
        return None
-    if len(content) == 1 and isinstance(content[0], TextPromptMessageContent):
-        return content[0].data
+    if all(isinstance(item, TextPromptMessageContent) for item in content):
+        return "".join(item.data for item in content)
    return content


@@ -242,6 +242,45 @@ class DifyLLMAdapterModelTests(unittest.IsolatedAsyncioTestCase):
        self.assertEqual(response.parts[0].part_kind, "text")
        self.assertEqual(cast(TextPart, response.parts[0]).content, "adapter response")

+    async def test_request_collapses_text_only_assistant_history_parts_to_string_content(self) -> None:
+        messages = [
+            ModelRequest(parts=[UserPromptPart("initial request")]),
+            ModelResponse(
+                parts=[
+                    ThinkingPart(content="plan"),
+                    TextPart(content="answer"),
+                ]
+            ),
+            ModelRequest(parts=[UserPromptPart("follow up")]),
+        ]
+
+        def handler(request: httpx.Request) -> httpx.Response:
+            payload = json.loads(request.content.decode("utf-8"))
+            prompt_messages = payload["data"]["prompt_messages"]
+
+            self.assertEqual([message["role"] for message in prompt_messages], ["user", "assistant", "user"])
+            self.assertEqual(prompt_messages[1]["content"], "<think>\nplan\n</think>answer")
+
+            return build_stream_response(*single_text_chunk("adapter response", prompt_tokens=11, completion_tokens=7))
+
+        async with self.mock_daemon_stream(httpx.MockTransport(handler)):
+            adapter = DifyLLMAdapterModel(
+                "demo-model",
+                self.make_provider(),
+                model_provider="openai",
+                credentials={"api_key": "secret"},
+            )
+
+            response = await adapter.request(
+                messages,
+                model_settings=None,
+                model_request_parameters=ModelRequestParameters(),
+            )
+
+        self.assertEqual(response.model_name, "demo-model")
+        self.assertEqual(response.parts[0].part_kind, "text")
+        self.assertEqual(cast(TextPart, response.parts[0]).content, "adapter response")
+
    async def test_request_omits_empty_assistant_history_when_response_has_no_content_or_tool_calls(self) -> None:
        messages = [
            ModelRequest(parts=[SystemPromptPart("request system"), UserPromptPart("hello")]),
@@ -1,5 +1,15 @@
@agent-v2 @authenticated @build
 Feature: Agent v2 build draft
+  @core
+  Scenario: Build chat is blocked until a model is configured
+    Given I am signed in as the default E2E admin
+    And an Agent v2 test agent has been created via API
+    And the Agent v2 composer draft uses the normal E2E prompt
+    When I open the Agent v2 configure page
+    And I try to generate an Agent v2 Build draft without a model
+    Then Agent v2 Build chat should be blocked until a model is configured
+    And the Agent v2 Build draft should not be checked out
+
  @external-model @agent-backend-runtime @stable-model
  Scenario: Generating a Build draft leaves the normal Agent configuration unchanged
    Given I am signed in as the default E2E admin
@@ -4,7 +4,7 @@ Feature: Agent Builder preseeded environment
  Scenario: Agent lifecycle permissions are available
    Given I am signed in as the default E2E admin
    And an Agent v2 test agent has been created via API
-    And the Agent v2 composer draft uses the normal E2E prompt
+    And the Agent v2 composer draft is publishable
    When I open the Agent v2 configure page
    And I publish the Agent v2 draft
    Then the Agent v2 draft should be published and up to date
@@ -1,5 +1,15 @@
@agent-v2 @authenticated @publish
 Feature: Agent v2 publish
+  @core
+  Scenario: Publish is blocked until a model is configured
+    Given I am signed in as the default E2E admin
+    And an Agent v2 test agent has been created via API
+    And the Agent v2 composer draft uses the normal E2E prompt
+    When I open the Agent v2 configure page
+    And I try to publish the Agent v2 draft without a model
+    Then Agent v2 publish should be blocked until a model is configured
+    And the Agent v2 draft should remain unpublished
+
  @core @stable-model
  Scenario: Publish a configured Agent v2 draft
    Given I am signed in as the default E2E admin
@@ -39,6 +39,21 @@ export async function saveAgentBuildDraft(
  }
 }

+export async function agentBuildDraftExists(agentId: string): Promise<boolean> {
+  const ctx = await createApiContext()
+  try {
+    const response = await ctx.get(`/console/api/agent/${agentId}/build-draft`)
+    if (response.status() === 404)
+      return false
+
+    await expectApiResponseOK(response, `Get Agent v2 build draft for ${agentId}`)
+    return true
+  }
+  finally {
+    await ctx.dispose()
+  }
+}
+
 export async function applyAgentBuildDraft(agentId: string): Promise<void> {
  const ctx = await createApiContext()
  try {
@@ -36,6 +36,11 @@ export const normalAgentSoulConfig: AgentSoulConfig = {
  },
 }

+export const publishOnlyAgentModel: AgentModelSelection = {
+  name: 'gpt-5-nano',
+  provider: 'openai',
+}
+
 export const updatedAgentSoulConfig: AgentSoulConfig = {
  prompt: {
    system_prompt: updatedAgentPrompt,
@@ -81,6 +86,13 @@ export function createAgentSoulConfigWithModel(
  }
 }

+export function createPublishableAgentSoulConfig(agentSoul: AgentSoulConfig): AgentSoulConfig {
+  if (agentSoul.model)
+    return agentSoul
+
+  return createAgentSoulConfigWithModel(agentSoul, publishOnlyAgentModel)
+}
+
 export function createAgentSoulConfigWithKnowledgeDataset(
  agentSoul: AgentSoulConfig,
  dataset: AgentKnowledgeDatasetConfig,
@@ -8,7 +8,7 @@ import type {
 } from '@dify/contracts/api/console/agent/types.gen'
 import { createApiContext, expectApiResponseOK } from '../../../support/api'
 import { assertE2EResourceName, createE2EResourceName } from '../../../support/naming'
-import { defaultAgentSoulConfig, normalAgentSoulConfig } from './agent-soul'
+import { createPublishableAgentSoulConfig, defaultAgentSoulConfig, normalAgentSoulConfig } from './agent-soul'

 export type AgentSeed = Pick<
  AgentAppDetailWithSite,
@@ -156,6 +156,12 @@ export async function getAgentComposerDraft(agentId: string): Promise<AgentAppCo
  }
 }

+export async function ensureAgentComposerDraftIsPublishable(agentId: string): Promise<void> {
+  const composer = await getAgentComposerDraft(agentId)
+  if (!composer.agent_soul?.model)
+    await saveAgentComposerDraft(agentId, createPublishableAgentSoulConfig(composer.agent_soul ?? defaultAgentSoulConfig))
+}
+
 export async function publishAgent(agentId: string, versionNote = 'E2E publish'): Promise<void> {
  const ctx = await createApiContext()
  try {
@@ -168,3 +174,11 @@ export async function publishAgent(agentId: string, versionNote = 'E2E publish')
    await ctx.dispose()
  }
 }
+
+export async function publishAgentWithPublishableDraft(
+  agentId: string,
+  versionNote = 'E2E publish',
+): Promise<void> {
+  await ensureAgentComposerDraftIsPublishable(agentId)
+  await publishAgent(agentId, versionNote)
+}
@@ -6,7 +6,7 @@ import {
  setAgentApiAccess,
  setAgentSiteAccessAndGetURL,
 } from '../../agent-v2/support/access-point'
-import { getAgentAccessPath, publishAgent } from '../../agent-v2/support/agent'
+import { getAgentAccessPath, publishAgentWithPublishableDraft } from '../../agent-v2/support/agent'
 import {
  getAccessRegion,
  getAccessSurfaceCard,
@@ -15,7 +15,7 @@ import {
 } from './access-point-helpers'

 Given('the Agent v2 draft has been published via API', async function (this: DifyWorld) {
-  await publishAgent(getCurrentAgentId(this))
+  await publishAgentWithPublishableDraft(getCurrentAgentId(this))
 })

 Given(
@@ -8,6 +8,7 @@ import {
  saveAgentComposerDraft,
 } from '../../agent-v2/support/agent'
 import {
+  agentBuildDraftExists,
  applyAgentBuildDraft,
  saveAgentBuildDraft,
 } from '../../agent-v2/support/agent-build-draft'
@@ -25,6 +26,7 @@ import { hasToolEntry } from '../../agent-v2/support/preflight/tools'
 import { agentBuilderTestMaterials, getAgentBuilderTestMaterialPath } from '../../agent-v2/support/test-materials'
 import { getPreseededToolContract } from '../../agent-v2/support/tools'
 import {
+  expectAgentModelRequiredFeedback,
  getAgentEnvVariableValue,
  getCurrentAgentId,
  uploadSummaryConfigSkillForBuildDraft,
@@ -142,6 +144,17 @@ When(
  },
 )

+When(
+  'I try to generate an Agent v2 Build draft without a model',
+  async function (this: DifyWorld) {
+    const page = this.getPage()
+
+    await page.getByRole('button', { exact: true, name: 'Build' }).click()
+    await page.getByPlaceholder('Describe what your agent should do').fill('Update the agent instructions for E2E.')
+    await page.getByRole('button', { name: 'Start build' }).click()
+  },
+)
+
 const expectPageResponseOK = async (response: Response, action: string) => {
  if (response.ok())
    return
@@ -246,6 +259,17 @@ Then('I should see the Agent v2 Build mode confirmation state', async function (
  ).toBeVisible()
 })

+Then('Agent v2 Build chat should be blocked until a model is configured', async function (this: DifyWorld) {
+  await expectAgentModelRequiredFeedback(this.getPage())
+})
+
+Then('the Agent v2 Build draft should not be checked out', async function (this: DifyWorld) {
+  await expect.poll(
+    async () => agentBuildDraftExists(getCurrentAgentId(this)),
+    { timeout: 30_000 },
+  ).toBe(false)
+})
+
 Then('I should see the e2e-summary-skill Skill in the Skills section', async function (this: DifyWorld) {
  const skillsSection = this.getPage().getByRole('region', { name: 'Skills' })

@@ -138,6 +138,10 @@ export const expectAgentConfigFileSaved = async (
    })
 }

+export const expectAgentModelRequiredFeedback = async (page: ReturnType<DifyWorld['getPage']>) => {
+  await expect(page.getByText('Select your model')).toBeVisible({ timeout: 10_000 })
+}
+
 export const uploadSummaryConfigSkillForBuildDraft = async (world: DifyWorld) => {
  const agentId = getCurrentAgentId(world)
  const skill = await uploadAgentConfigSkillToDraft({
@@ -15,6 +15,7 @@ import {
  concurrentFirstAgentPrompt,
  concurrentSecondAgentPrompt,
  createAgentSoulConfigWithModel,
+  createPublishableAgentSoulConfig,
  normalAgentPrompt,
  normalAgentSoulConfig,
  updatedAgentPrompt,
@@ -131,6 +132,16 @@ Given('the Agent v2 composer draft uses the normal E2E prompt', async function (
  await saveAgentComposerDraft(getCurrentAgentId(this), normalAgentSoulConfig)
 })

+Given(
+  'the Agent v2 composer draft is publishable',
+  async function (this: DifyWorld) {
+    await saveAgentComposerDraft(
+      getCurrentAgentId(this),
+      createPublishableAgentSoulConfig(normalAgentSoulConfig),
+    )
+  },
+)
+
 Given('the e2e-summary-skill Skill is available to the Agent v2 test agent', async function (this: DifyWorld) {
  const agentId = getCurrentAgentId(this)
  const upload = await uploadAgentDriveSkill({
@@ -4,7 +4,7 @@ import { expect } from '@playwright/test'
 import { waitForAgentConfigureAutosaved } from '../../../support/agent-configure'
 import { getAgentVersionDetail, getTestAgent } from '../../agent-v2/support/agent'
 import { normalAgentPrompt } from '../../agent-v2/support/agent-soul'
-import { getCurrentAgentId } from './configure-helpers'
+import { expectAgentModelRequiredFeedback, getCurrentAgentId } from './configure-helpers'

 When('I publish the Agent v2 draft', async function (this: DifyWorld) {
  const page = this.getPage()
@@ -14,6 +14,25 @@ When('I publish the Agent v2 draft', async function (this: DifyWorld) {
  await publishButton.click()
 })

+When('I try to publish the Agent v2 draft without a model', async function (this: DifyWorld) {
+  const page = this.getPage()
+  const publishButton = page.getByRole('button', { name: /^Publish(?: update)?$/ })
+
+  await expect(publishButton).toBeEnabled({ timeout: 30_000 })
+  await publishButton.click()
+})
+
+Then('Agent v2 publish should be blocked until a model is configured', async function (this: DifyWorld) {
+  await expectAgentModelRequiredFeedback(this.getPage())
+})
+
+Then('the Agent v2 draft should remain unpublished', async function (this: DifyWorld) {
+  await expect.poll(
+    async () => (await getTestAgent(getCurrentAgentId(this))).active_config_is_published,
+    { timeout: 30_000 },
+  ).toBe(false)
+})
+
 Then('the Agent v2 configuration should be saved automatically', async function (this: DifyWorld) {
  await waitForAgentConfigureAutosaved(this.getPage())
 })
@@ -1116,6 +1116,7 @@ export type AgentSource = 'agent_app' | 'imported' | 'roster' | 'system' | 'work
 export type AgentStatus = 'active' | 'archived'

 export type AgentSoulAppFeaturesConfig = {
+  file_upload?: AgentFileUploadFeatureConfig
  opening_statement?: string | null
  retriever_resource?: AgentFeatureToggleConfig | null
  sensitive_word_avoidance?: AgentSensitiveWordAvoidanceFeatureConfig | null
@@ -1428,6 +1429,16 @@ export type AgentConfigRevisionOperation
    | 'save_new_version'
    | 'save_to_roster'

+export type AgentFileUploadFeatureConfig = {
+  allowed_file_extensions?: Array<string>
+  allowed_file_types?: Array<FileType>
+  allowed_file_upload_methods?: Array<FileTransferMethod>
+  enabled?: boolean
+  image?: AgentFileUploadImageFeatureConfig
+  number_limits?: number
+  [key: string]: unknown
+}
+
 export type AgentSecretRefConfig = {
  credential_id?: string | null
  env_name?: string | null
@@ -1679,6 +1690,15 @@ export type FormInputConfig

 export type JsonValue2 = unknown

+export type FileType = 'audio' | 'custom' | 'document' | 'image' | 'video'
+
+export type FileTransferMethod = 'datasource_file' | 'local_file' | 'remote_url' | 'tool_file'
+
+export type AgentFileUploadImageFeatureConfig = {
+  enabled?: boolean
+  [key: string]: unknown
+}
+
 export type AgentKnowledgeDatasetConfig = {
  description?: string | null
  id?: string | null
@@ -1801,10 +1821,6 @@ export type StringListSource = {
  value?: Array<string>
 }

-export type FileType = 'audio' | 'custom' | 'document' | 'image' | 'video'
-
-export type FileTransferMethod = 'datasource_file' | 'local_file' | 'remote_url' | 'tool_file'
-
 export type AgentKnowledgeMetadataCondition = {
  comparison_operator:
    | '<'
@@ -1926,19 +1926,6 @@ export const zAgentAppFeaturesPayload = z.object({
  text_to_speech: zAgentTextToSpeechFeatureConfig.nullish(),
 })

-/**
- * AgentSoulAppFeaturesConfig
- */
-export const zAgentSoulAppFeaturesConfig = z.object({
-  opening_statement: z.string().nullish(),
-  retriever_resource: zAgentFeatureToggleConfig.nullish(),
-  sensitive_word_avoidance: zAgentSensitiveWordAvoidanceFeatureConfig.nullish(),
-  speech_to_text: zAgentFeatureToggleConfig.nullish(),
-  suggested_questions: z.array(z.string()).nullish(),
-  suggested_questions_after_answer: zAgentSuggestedQuestionsAfterAnswerFeatureConfig.nullish(),
-  text_to_speech: zAgentTextToSpeechFeatureConfig.nullish(),
-})
-
 export const zJsonValue2 = z.unknown()

 /**
@@ -1953,6 +1940,54 @@ export const zHumanInputFormSubmissionData = z.object({
  submitted_data: z.record(z.string(), zJsonValue2).nullish(),
 })

+/**
+ * FileType
+ */
+export const zFileType = z.enum(['audio', 'custom', 'document', 'image', 'video'])
+
+/**
+ * FileTransferMethod
+ */
+export const zFileTransferMethod = z.enum([
+  'datasource_file',
+  'local_file',
+  'remote_url',
+  'tool_file',
+])
+
+/**
+ * AgentFileUploadImageFeatureConfig
+ */
+export const zAgentFileUploadImageFeatureConfig = z.object({
+  enabled: z.boolean().optional().default(true),
+})
+
+/**
+ * AgentFileUploadFeatureConfig
+ */
+export const zAgentFileUploadFeatureConfig = z.object({
+  allowed_file_extensions: z.array(z.string()).optional(),
+  allowed_file_types: z.array(zFileType).optional(),
+  allowed_file_upload_methods: z.array(zFileTransferMethod).optional(),
+  enabled: z.boolean().optional().default(true),
+  image: zAgentFileUploadImageFeatureConfig.optional(),
+  number_limits: z.int().optional().default(3),
+})
+
+/**
+ * AgentSoulAppFeaturesConfig
+ */
+export const zAgentSoulAppFeaturesConfig = z.object({
+  file_upload: zAgentFileUploadFeatureConfig.optional(),
+  opening_statement: z.string().nullish(),
+  retriever_resource: zAgentFeatureToggleConfig.nullish(),
+  sensitive_word_avoidance: zAgentSensitiveWordAvoidanceFeatureConfig.nullish(),
+  speech_to_text: zAgentFeatureToggleConfig.nullish(),
+  suggested_questions: z.array(z.string()).nullish(),
+  suggested_questions_after_answer: zAgentSuggestedQuestionsAfterAnswerFeatureConfig.nullish(),
+  text_to_speech: zAgentTextToSpeechFeatureConfig.nullish(),
+})
+
 /**
 * AgentKnowledgeDatasetConfig
 */
@@ -2182,6 +2217,29 @@ export const zUserActionConfig = z.object({
  title: z.string().max(100),
 })

+/**
+ * FileInputConfig
+ */
+export const zFileInputConfig = z.object({
+  allowed_file_extensions: z.array(z.string()).optional(),
+  allowed_file_types: z.array(zFileType).optional(),
+  allowed_file_upload_methods: z.array(zFileTransferMethod).optional(),
+  output_variable_name: z.string(),
+  type: z.literal('file').optional().default('file'),
+})
+
+/**
+ * FileListInputConfig
+ */
+export const zFileListInputConfig = z.object({
+  allowed_file_extensions: z.array(z.string()).optional(),
+  allowed_file_types: z.array(zFileType).optional(),
+  allowed_file_upload_methods: z.array(zFileTransferMethod).optional(),
+  number_limits: z.int().gte(0).optional().default(0),
+  output_variable_name: z.string(),
+  type: z.literal('file-list').optional().default('file-list'),
+})
+
 /**
 * AgentKnowledgeModelConfig
 */
@@ -2204,8 +2262,9 @@ export const zAgentKnowledgeQueryMode = z.enum(['generated_query', 'user_query']
 *
 * Agent v2 stores knowledge as explicit ``knowledge.sets`` rather than the
 * legacy flat ``datasets`` / ``query_mode`` / ``query_config`` shape. Each
- * set owns its own query policy, so ``user_query`` must carry an explicit
- * ``value`` while ``generated_query`` leaves that value empty.
+ * set owns its own query policy. Mode-dependent completeness, such as
+ * requiring ``value`` for ``user_query``, is enforced by composer publish
+ * validation so draft saves can persist partially configured knowledge sets.
 */
 export const zAgentKnowledgeQueryConfig = z.object({
  mode: zAgentKnowledgeQueryMode,
@@ -2235,8 +2294,9 @@ export const zAgentKnowledgeWeightedScoreConfig = z.object({
 * Per-set retrieval policy for Agent v2 knowledge retrieval.
 *
 * Retrieval settings now live on each knowledge set instead of one shared
- * flat config. A set may use either ``multiple`` retrieval with ``top_k`` or
- * ``single`` retrieval with a required model config.
+ * flat config. Mode-dependent completeness, such as requiring ``top_k`` for
+ * ``multiple`` or a model for ``single``, is enforced by composer publish
+ * validation so draft saves can persist partially configured knowledge sets.
 */
 export const zAgentKnowledgeRetrievalConfig = z.object({
  mode: z.enum(['multiple', 'single']),
@@ -2249,44 +2309,6 @@ export const zAgentKnowledgeRetrievalConfig = z.object({
  weights: zAgentKnowledgeWeightedScoreConfig.nullish(),
 })

-/**
- * FileType
- */
-export const zFileType = z.enum(['audio', 'custom', 'document', 'image', 'video'])
-
-/**
- * FileTransferMethod
- */
-export const zFileTransferMethod = z.enum([
-  'datasource_file',
-  'local_file',
-  'remote_url',
-  'tool_file',
-])
-
-/**
- * FileInputConfig
- */
-export const zFileInputConfig = z.object({
-  allowed_file_extensions: z.array(z.string()).optional(),
-  allowed_file_types: z.array(zFileType).optional(),
-  allowed_file_upload_methods: z.array(zFileTransferMethod).optional(),
-  output_variable_name: z.string(),
-  type: z.literal('file').optional().default('file'),
-})
-
-/**
- * FileListInputConfig
- */
-export const zFileListInputConfig = z.object({
-  allowed_file_extensions: z.array(z.string()).optional(),
-  allowed_file_types: z.array(zFileType).optional(),
-  allowed_file_upload_methods: z.array(zFileTransferMethod).optional(),
-  number_limits: z.int().gte(0).optional().default(0),
-  output_variable_name: z.string(),
-  type: z.literal('file-list').optional().default('file-list'),
-})
-
 /**
 * AgentKnowledgeMetadataCondition
 */
@@ -2331,6 +2353,8 @@ export const zAgentKnowledgeMetadataConditions = z.object({
 * The Python attribute uses ``metadata_model_config`` for clarity because the
 * model belongs to metadata filtering specifically, while the external API and
 * generated schema keep the historical ``model_config`` field name via alias.
+ * Mode-dependent completeness is enforced by composer publish validation so
+ * draft saves can persist partially configured metadata filters.
 */
 export const zAgentKnowledgeMetadataFilteringConfig = z.object({
  conditions: zAgentKnowledgeMetadataConditions.nullish(),
@@ -2368,6 +2368,7 @@ export type AgentSource = 'agent_app' | 'imported' | 'roster' | 'system' | 'work
 export type AgentStatus = 'active' | 'archived'

 export type AgentSoulAppFeaturesConfig = {
+  file_upload?: AgentFileUploadFeatureConfig
  opening_statement?: string | null
  retriever_resource?: AgentFeatureToggleConfig | null
  sensitive_word_avoidance?: AgentSensitiveWordAvoidanceFeatureConfig | null
@@ -2627,6 +2628,16 @@ export type WorkflowFileUploadPreviewConfigPayload = {
  mode?: string | null
 }

+export type AgentFileUploadFeatureConfig = {
+  allowed_file_extensions?: Array<string>
+  allowed_file_types?: Array<FileType>
+  allowed_file_upload_methods?: Array<FileTransferMethod>
+  enabled?: boolean
+  image?: AgentFileUploadImageFeatureConfig
+  number_limits?: number
+  [key: string]: unknown
+}
+
 export type AgentFeatureToggleConfig = {
  enabled?: boolean
  [key: string]: unknown
@@ -2873,6 +2884,15 @@ export type FileListInputConfig = {
  type?: 'file-list'
 }

+export type FileType = 'audio' | 'custom' | 'document' | 'image' | 'video'
+
+export type FileTransferMethod = 'datasource_file' | 'local_file' | 'remote_url' | 'tool_file'
+
+export type AgentFileUploadImageFeatureConfig = {
+  enabled?: boolean
+  [key: string]: unknown
+}
+
 export type AgentModerationProviderConfig = {
  api_based_extension_id?: string | null
  inputs_config?: AgentModerationIoConfig | null
@@ -2942,10 +2962,6 @@ export type StringListSource = {
  value?: Array<string>
 }

-export type FileType = 'audio' | 'custom' | 'document' | 'image' | 'video'
-
-export type FileTransferMethod = 'datasource_file' | 'local_file' | 'remote_url' | 'tool_file'
-
 export type AgentModerationIoConfig = {
  enabled?: boolean
  preset_response?: string | null
@@ -3466,6 +3466,63 @@ export const zUserActionConfig = z.object({
  title: z.string().max(100),
 })

+/**
+ * FileType
+ */
+export const zFileType = z.enum(['audio', 'custom', 'document', 'image', 'video'])
+
+/**
+ * FileTransferMethod
+ */
+export const zFileTransferMethod = z.enum([
+  'datasource_file',
+  'local_file',
+  'remote_url',
+  'tool_file',
+])
+
+/**
+ * FileInputConfig
+ */
+export const zFileInputConfig = z.object({
+  allowed_file_extensions: z.array(z.string()).optional(),
+  allowed_file_types: z.array(zFileType).optional(),
+  allowed_file_upload_methods: z.array(zFileTransferMethod).optional(),
+  output_variable_name: z.string(),
+  type: z.literal('file').optional().default('file'),
+})
+
+/**
+ * FileListInputConfig
+ */
+export const zFileListInputConfig = z.object({
+  allowed_file_extensions: z.array(z.string()).optional(),
+  allowed_file_types: z.array(zFileType).optional(),
+  allowed_file_upload_methods: z.array(zFileTransferMethod).optional(),
+  number_limits: z.int().gte(0).optional().default(0),
+  output_variable_name: z.string(),
+  type: z.literal('file-list').optional().default('file-list'),
+})
+
+/**
+ * AgentFileUploadImageFeatureConfig
+ */
+export const zAgentFileUploadImageFeatureConfig = z.object({
+  enabled: z.boolean().optional().default(true),
+})
+
+/**
+ * AgentFileUploadFeatureConfig
+ */
+export const zAgentFileUploadFeatureConfig = z.object({
+  allowed_file_extensions: z.array(z.string()).optional(),
+  allowed_file_types: z.array(zFileType).optional(),
+  allowed_file_upload_methods: z.array(zFileTransferMethod).optional(),
+  enabled: z.boolean().optional().default(true),
+  image: zAgentFileUploadImageFeatureConfig.optional(),
+  number_limits: z.int().optional().default(3),
+})
+
 /**
 * AgentSuggestedQuestionsAfterAnswerModelConfig
 *
@@ -3662,44 +3719,6 @@ export const zAgentSoulToolsConfig = z.object({
  dify_tools: z.array(zAgentSoulDifyToolConfig).optional(),
 })

-/**
- * FileType
- */
-export const zFileType = z.enum(['audio', 'custom', 'document', 'image', 'video'])
-
-/**
- * FileTransferMethod
- */
-export const zFileTransferMethod = z.enum([
-  'datasource_file',
-  'local_file',
-  'remote_url',
-  'tool_file',
-])
-
-/**
- * FileInputConfig
- */
-export const zFileInputConfig = z.object({
-  allowed_file_extensions: z.array(z.string()).optional(),
-  allowed_file_types: z.array(zFileType).optional(),
-  allowed_file_upload_methods: z.array(zFileTransferMethod).optional(),
-  output_variable_name: z.string(),
-  type: z.literal('file').optional().default('file'),
-})
-
-/**
- * FileListInputConfig
- */
-export const zFileListInputConfig = z.object({
-  allowed_file_extensions: z.array(z.string()).optional(),
-  allowed_file_types: z.array(zFileType).optional(),
-  allowed_file_upload_methods: z.array(zFileTransferMethod).optional(),
-  number_limits: z.int().gte(0).optional().default(0),
-  output_variable_name: z.string(),
-  type: z.literal('file-list').optional().default('file-list'),
-})
-
 /**
 * AgentModerationIOConfig
 */
@@ -3731,6 +3750,7 @@ export const zAgentSensitiveWordAvoidanceFeatureConfig = z.object({
 * AgentSoulAppFeaturesConfig
 */
 export const zAgentSoulAppFeaturesConfig = z.object({
+  file_upload: zAgentFileUploadFeatureConfig.optional(),
  opening_statement: z.string().nullish(),
  retriever_resource: zAgentFeatureToggleConfig.nullish(),
  sensitive_word_avoidance: zAgentSensitiveWordAvoidanceFeatureConfig.nullish(),
@@ -3762,8 +3782,9 @@ export const zAgentKnowledgeQueryMode = z.enum(['generated_query', 'user_query']
 *
 * Agent v2 stores knowledge as explicit ``knowledge.sets`` rather than the
 * legacy flat ``datasets`` / ``query_mode`` / ``query_config`` shape. Each
- * set owns its own query policy, so ``user_query`` must carry an explicit
- * ``value`` while ``generated_query`` leaves that value empty.
+ * set owns its own query policy. Mode-dependent completeness, such as
+ * requiring ``value`` for ``user_query``, is enforced by composer publish
+ * validation so draft saves can persist partially configured knowledge sets.
 */
 export const zAgentKnowledgeQueryConfig = z.object({
  mode: zAgentKnowledgeQueryMode,
@@ -3793,8 +3814,9 @@ export const zAgentKnowledgeWeightedScoreConfig = z.object({
 * Per-set retrieval policy for Agent v2 knowledge retrieval.
 *
 * Retrieval settings now live on each knowledge set instead of one shared
- * flat config. A set may use either ``multiple`` retrieval with ``top_k`` or
- * ``single`` retrieval with a required model config.
+ * flat config. Mode-dependent completeness, such as requiring ``top_k`` for
+ * ``multiple`` or a model for ``single``, is enforced by composer publish
+ * validation so draft saves can persist partially configured knowledge sets.
 */
 export const zAgentKnowledgeRetrievalConfig = z.object({
  mode: z.enum(['multiple', 'single']),
@@ -3972,6 +3994,8 @@ export const zAgentKnowledgeMetadataConditions = z.object({
 * The Python attribute uses ``metadata_model_config`` for clarity because the
 * model belongs to metadata filtering specifically, while the external API and
 * generated schema keep the historical ``model_config`` field name via alias.
+ * Mode-dependent completeness is enforced by composer publish validation so
+ * draft saves can persist partially configured metadata filters.
 */
 export const zAgentKnowledgeMetadataFilteringConfig = z.object({
  conditions: zAgentKnowledgeMetadataConditions.nullish(),
@@ -8,7 +8,7 @@ import {
  FieldRoot,
 } from '../field'
 import { FieldsetLegend, FieldsetRoot } from '../fieldset'
-import { Radio, RadioControl, RadioRoot } from '../radio'
+import { Radio, RadioControl, RadioIndicator, RadioRoot } from '../radio'

 const meta = {
  title: 'Base/Form/RadioGroup',
@@ -17,7 +17,7 @@ const meta = {
    layout: 'centered',
    docs: {
      description: {
-        component: 'RadioGroup primitive built on Base UI. For normal form rows, compose FieldRoot, FieldsetRoot, FieldLabel, RadioGroup, and Radio. For option cards, wrap each option in FieldItem and make the card itself a RadioRoot with variant="unstyled".',
+        component: '`RadioGroup` owns single-selection state. Use `Radio` for plain form rows, `RadioRoot` when an entire row or card is the radio item, `RadioControl` for the standard visual dot inside custom roots, and `RadioIndicator` only when the design owns a custom control shell.',
      },
    },
  },
@@ -60,7 +60,7 @@ export const StandardFormRows: Story = {
  parameters: {
    docs: {
      description: {
-        story: 'Default form composition. Most product code should use this shape: RadioGroup owns value, FieldsetLegend names the group, and FieldLabel makes each row clickable.',
+        story: 'Plain form-row composition. `RadioGroup` owns value, `FieldsetLegend` names the group, `FieldLabel` makes each option label clickable, and `Radio` renders the default dot.',
      },
    },
  },
@@ -107,31 +107,39 @@ export const BooleanInline: Story = {
  },
 }

+type PromptMode = 'default' | 'custom'
+
 function OptionCardsDemo() {
-  const [value, setValue] = React.useState('default')
+  const [value, setValue] = React.useState<PromptMode>('default')
+
+  const options = [
+    {
+      value: 'default',
+      title: 'Default prompt',
+      description: 'Use the built-in prompt for consistent output.',
+    },
+    {
+      value: 'custom',
+      title: 'Custom prompt',
+      description: 'Write a prompt for this app and keep full control.',
+    },
+  ] satisfies Array<{
+    value: PromptMode
+    title: string
+    description: string
+  }>

  return (
    <FieldRoot name="promptMode" className="w-100">
      <FieldsetRoot
        render={(
-          <RadioGroup value={value} onValueChange={setValue} className="flex-col items-stretch gap-3" />
+          <RadioGroup<PromptMode> value={value} onValueChange={setValue} className="flex-col items-stretch gap-3" />
        )}
      >
        <FieldsetLegend>Prompt mode</FieldsetLegend>
-        {[
-          {
-            value: 'default',
-            title: 'Default prompt',
-            description: 'Use the built-in prompt for consistent output.',
-          },
-          {
-            value: 'custom',
-            title: 'Custom prompt',
-            description: 'Write a prompt for this app and keep full control.',
-          },
-        ].map(option => (
+        {options.map(option => (
          <FieldItem key={option.value}>
-            <RadioRoot
+            <RadioRoot<PromptMode>
              value={option.value}
              variant="unstyled"
              nativeButton
@@ -162,7 +170,76 @@ export const OptionCards: Story = {
  parameters: {
    docs: {
      description: {
-        story: 'Wrap each option card in FieldItem, then use RadioRoot with variant="unstyled" when the entire card is the radio. RadioControl renders the visual dot inside the card.',
+        story: 'Product option cards should make the whole card the radio item with `RadioRoot variant="unstyled"`. `RadioControl` renders the standard visual dot inside the custom root.',
+      },
+    },
+  },
+}
+
+type ApprovalMode = 'automatic' | 'manual'
+
+function CustomIndicatorPartDemo() {
+  const [value, setValue] = React.useState<ApprovalMode>('automatic')
+
+  const options = [
+    {
+      value: 'automatic',
+      title: 'Automatic approval',
+      description: 'Approve requests that match policy.',
+    },
+    {
+      value: 'manual',
+      title: 'Manual review',
+      description: 'Ask an admin to review each request.',
+    },
+  ] satisfies Array<{
+    value: ApprovalMode
+    title: string
+    description: string
+  }>
+
+  return (
+    <FieldRoot name="approvalMode" className="w-100">
+      <FieldsetRoot
+        render={(
+          <RadioGroup<ApprovalMode> value={value} onValueChange={setValue} className="flex-col items-stretch gap-2" />
+        )}
+      >
+        <FieldsetLegend>Approval mode</FieldsetLegend>
+        {options.map(option => (
+          <FieldItem key={option.value}>
+            <RadioRoot<ApprovalMode>
+              value={option.value}
+              variant="unstyled"
+              nativeButton
+              render={<button type="button" />}
+              className="flex w-full items-center gap-3 rounded-lg border border-components-option-card-option-border bg-components-option-card-option-bg px-3 py-2 text-left outline-hidden transition-colors hover:bg-state-base-hover focus-visible:ring-2 focus-visible:ring-state-accent-solid data-checked:border-components-option-card-option-selected-border data-checked:bg-components-option-card-option-selected-bg"
+            >
+              <span className="flex size-4 shrink-0 items-center justify-center rounded-full border border-components-radio-border bg-components-radio-bg">
+                <RadioIndicator className="text-components-radio-border-checked" />
+              </span>
+              <span className="min-w-0 grow">
+                <span className="block truncate system-sm-semibold text-text-primary">
+                  {option.title}
+                </span>
+                <span className="block truncate system-xs-regular text-text-tertiary">
+                  {option.description}
+                </span>
+              </span>
+            </RadioRoot>
+          </FieldItem>
+        ))}
+      </FieldsetRoot>
+    </FieldRoot>
+  )
+}
+
+export const CustomIndicatorPart: Story = {
+  render: () => <CustomIndicatorPartDemo />,
+  parameters: {
+    docs: {
+      description: {
+        story: '`RadioIndicator` is the low-level indicator part. Use it only when a custom root owns the outer control shell; otherwise prefer `Radio` for form rows or `RadioControl` inside option cards.',
      },
    },
  },
@@ -15,7 +15,7 @@ const meta = {
    layout: 'centered',
    docs: {
      description: {
-        component: 'Radio primitive built on Base UI. It preserves RadioGroup selection, hidden input, disabled, and form semantics while applying the Dify 16px radio design from Figma. Import from `@langgenius/dify-ui/radio` and place radios inside `RadioGroup` from `@langgenius/dify-ui/radio-group`.',
+        component: '`Radio` is the default Dify 16px radio control, intended for plain form rows inside `RadioGroup`. It does not accept children. For option cards or rich rows, use `RadioRoot` as the item root and place `RadioControl` inside it; see the `RadioGroup` stories.',
      },
    },
  },
@@ -68,6 +68,13 @@ export const Default: Story = {
  args: {
    disabled: false,
  },
+  parameters: {
+    docs: {
+      description: {
+        story: '`Radio` renders the standard visual control. `FieldLabel` owns the clickable text label, and `RadioGroup` owns the selected value.',
+      },
+    },
+  },
 }

 export const Disabled: Story = {
@@ -80,17 +80,11 @@ export function RadioControl({
 export type RadioProps<Value = string>
  = Omit<RadioRootProps<Value>, 'children'>

-export function Radio<Value = string>({
-  ...props
-}: RadioProps<Value>) {
+export function Radio<Value = string>(props: RadioProps<Value>) {
  return <RadioRoot<Value> {...props} />
 }

-export type RadioSkeletonProps
-  = Omit<React.ComponentProps<'div'>, 'className'>
-    & {
-      className?: string
-    }
+export type RadioSkeletonProps = React.ComponentProps<'div'>

 export function RadioSkeleton({
  className,
@@ -211,16 +211,18 @@ const ChatInputArea = ({ readonly, botName, customPlaceholder, showFeatureBar, s
      {shouldShowFooterNotice && (
        <div className="m-1 mt-0 -translate-y-2 rounded-b-[10px] border-r border-b border-l border-components-panel-border-subtle bg-util-colors-indigo-indigo-50 px-2.5 py-2 pt-4">
          <div className="flex items-center gap-1">
+            <span aria-hidden className="i-ri-information-line size-3.5 shrink-0 text-text-accent" />
+            <div className="body-xs-medium text-text-accent">{footerNotice}</div>
            {shouldShowFooterNoticeTooltip && (
              <Tooltip>
                <TooltipTrigger
                  render={(
                    <button
                      type="button"
-                      className="flex size-5 items-center justify-center rounded-md text-text-accent hover:bg-state-base-hover focus-visible:outline focus-visible:outline-2 focus-visible:outline-offset-2 focus-visible:outline-components-button-primary-bg"
+                      className="ml-auto flex size-5 items-center justify-center rounded-md system-xs-medium text-text-accent hover:bg-state-base-hover focus-visible:outline focus-visible:outline-2 focus-visible:outline-offset-2 focus-visible:outline-components-button-primary-bg"
                      aria-label={typeof footerNoticeTooltip === 'string' ? footerNoticeTooltip : undefined}
                    >
-                      <span aria-hidden className="i-ri-information-line size-3.5 shrink-0" />
+                      <span aria-hidden className="i-ri-question-line size-3.5 shrink-0" />
                    </button>
                  )}
                />
@@ -229,7 +231,6 @@ const ChatInputArea = ({ readonly, botName, customPlaceholder, showFeatureBar, s
                </TooltipContent>
              </Tooltip>
            )}
-            <div className="body-xs-medium text-text-accent">{footerNotice}</div>
          </div>
        </div>
      )}
@@ -14,7 +14,7 @@ export function SnippetPlaceholderIcon({
  return (
    <div
      className={cn(
-        'flex size-10 shrink-0 items-center justify-center rounded-[10px] border border-divider-subtle bg-state-accent-hover text-text-accent shadow-xs',
+        'flex size-10 shrink-0 items-center justify-center rounded-[10px] border border-divider-subtle bg-state-accent-hover text-[#5D6FBB] shadow-xs',
        className,
      )}
      aria-hidden="true"
@@ -1,3 +1,7 @@
+import type {
+  AgentInviteOptionResponse,
+  AgentInviteOptionsResponse,
+} from '@dify/contracts/api/console/agent/types.gen'
 import type { NodeDefault } from '../../types'
 import { QueryClient, QueryClientProvider } from '@tanstack/react-query'
 import { render, screen, waitFor } from '@testing-library/react'
@@ -15,7 +19,7 @@ const runtimeState = vi.hoisted(() => ({
 }))

 const queryMocks = vi.hoisted(() => ({
-  inviteOptionsQueryFn: vi.fn(),
+  request: vi.fn(),
  toastError: vi.fn(),
 }))

@@ -35,19 +39,8 @@ vi.mock('@/app/components/app/store', () => ({
  }),
 }))

-vi.mock('@/service/client', () => ({
-  consoleQuery: {
-    agent: {
-      inviteOptions: {
-        get: {
-          queryOptions: (options: unknown) => ({
-            queryKey: ['agents', 'invite-options', options],
-            queryFn: () => queryMocks.inviteOptionsQueryFn(options),
-          }),
-        },
-      },
-    },
-  },
+vi.mock('@/service/base', () => ({
+  request: (...args: unknown[]) => queryMocks.request(...args),
 }))

 vi.mock('@langgenius/dify-ui/toast', () => ({
@@ -74,6 +67,58 @@ const createBlock = (
  checkValid: () => ({ isValid: true }),
 })

+const createInviteOption = (
+  overrides: Partial<AgentInviteOptionResponse> & Pick<AgentInviteOptionResponse, 'id' | 'name'>,
+): AgentInviteOptionResponse => {
+  const { id, name, ...rest } = overrides
+
+  return {
+    id,
+    name,
+    description: rest.description ?? 'Clarification Drafter',
+    active_config_snapshot_id: rest.active_config_snapshot_id ?? 'version-1',
+    role: rest.role ?? 'Researcher',
+    agent_kind: rest.agent_kind ?? 'dify_agent',
+    icon: rest.icon ?? 'A',
+    icon_background: rest.icon_background ?? '#E9D7FE',
+    icon_type: rest.icon_type ?? 'emoji',
+    scope: rest.scope ?? 'roster',
+    source: rest.source ?? 'workflow',
+    status: rest.status ?? 'active',
+    ...rest,
+  }
+}
+
+const createInviteOptionsResponse = (
+  agents: AgentInviteOptionResponse[],
+): AgentInviteOptionsResponse => ({
+  data: agents,
+  has_more: false,
+  limit: 8,
+  page: 1,
+  total: agents.length,
+})
+
+const createJsonResponse = (body: unknown) =>
+  new Response(JSON.stringify(body), {
+    status: 200,
+    headers: {
+      'Content-Type': 'application/json',
+    },
+  })
+
+const mockInviteOptionsResponse = (agents: AgentInviteOptionResponse[]) => {
+  queryMocks.request.mockImplementation(() => Promise.resolve(createJsonResponse(createInviteOptionsResponse(agents))))
+}
+
+const expectLastInviteOptionsRequest = () => {
+  const [url] = queryMocks.request.mock.calls.at(-1) ?? []
+  const requestURL = new URL(String(url), window.location.origin)
+
+  expect(requestURL.pathname).toBe('/console/api/agent/invite-options')
+  return requestURL
+}
+
 describe('Blocks', () => {
  beforeEach(() => {
    vi.clearAllMocks()
@@ -125,13 +170,7 @@ describe('Blocks', () => {

  it('opens the agent selector on Agent block hover', async () => {
    const user = userEvent.setup()
-    queryMocks.inviteOptionsQueryFn.mockResolvedValue({
-      data: [],
-      has_more: false,
-      limit: 8,
-      page: 1,
-      total: 0,
-    })
+    mockInviteOptionsResponse([])
    const queryClient = new QueryClient({
      defaultOptions: {
        queries: {
@@ -171,28 +210,12 @@ describe('Blocks', () => {
  it('opens the agent selector from the Agent block and selects an agent', async () => {
    const user = userEvent.setup()
    const onSelect = vi.fn()
-    queryMocks.inviteOptionsQueryFn.mockResolvedValue({
-      data: [
-        {
-          id: 'agent-1',
-          name: 'Nadia',
-          description: 'Clarification Drafter',
-          active_config_snapshot_id: 'version-1',
-          role: 'Researcher',
-          agent_kind: 'dify_agent',
-          icon: 'A',
-          icon_background: '#E9D7FE',
-          icon_type: 'emoji',
-          scope: 'roster',
-          source: 'workflow',
-          status: 'active',
-        },
-      ],
-      has_more: false,
-      limit: 8,
-      page: 1,
-      total: 1,
-    })
+    mockInviteOptionsResponse([
+      createInviteOption({
+        id: 'agent-1',
+        name: 'Nadia',
+      }),
+    ])

    const queryClient = new QueryClient({
      defaultOptions: {
@@ -246,42 +269,84 @@ describe('Blocks', () => {
      agent_node_kind: 'dify_agent',
      version: '2',
    })
-    expect(queryMocks.inviteOptionsQueryFn).toHaveBeenCalledWith({
-      input: {
-        query: {
-          app_id: 'app-1',
-          limit: 8,
-          page: 1,
+    const requestURL = expectLastInviteOptionsRequest()
+    expect(requestURL.searchParams.get('app_id')).toBe('app-1')
+    expect(requestURL.searchParams.get('limit')).toBe('8')
+    expect(requestURL.searchParams.get('page')).toBe('1')
+  })
+
+  it('should refresh Agent v2 roster options when the selector is reopened', async () => {
+    const user = userEvent.setup()
+    queryMocks.request
+      .mockImplementationOnce(() => Promise.resolve(createJsonResponse(createInviteOptionsResponse([
+        createInviteOption({
+          id: 'agent-1',
+          name: 'Nadia',
+        }),
+      ]))))
+      .mockImplementation(() => Promise.resolve(createJsonResponse(createInviteOptionsResponse([
+        createInviteOption({
+          id: 'agent-2',
+          name: 'Bruno',
+          role: 'Planner',
+        }),
+      ]))))
+    const queryClient = new QueryClient({
+      defaultOptions: {
+        queries: {
+          retry: false,
+          staleTime: 5 * 60 * 1000,
        },
      },
    })
+    const hooksStore = createHooksStore({
+      configsMap: {
+        flowId: 'app-1',
+        flowType: FlowType.appFlow,
+        fileSettings: {} as never,
+      },
+    })
+
+    render(
+      <QueryClientProvider client={queryClient}>
+        <HooksStoreContext value={hooksStore}>
+          <Blocks
+            searchText=""
+            onSelect={vi.fn()}
+            availableBlocksTypes={[BlockEnum.AgentV2]}
+            blocks={[createBlock(BlockEnum.AgentV2, 'Agent', BlockClassificationEnum.Default, 3)]}
+          />
+        </HooksStoreContext>
+      </QueryClientProvider>,
+    )
+
+    await user.click(screen.getByRole('button', { name: /Agent/ }))
+    expect(await screen.findByText('Nadia')).toBeInTheDocument()
+
+    await user.click(screen.getByRole('combobox', { name: 'agentV2.roster.searchLabel' }))
+    await user.keyboard('{Escape}')
+    await waitFor(() => {
+      expect(screen.queryByRole('dialog', { name: 'agentV2.roster.nodeSelector.dialogLabel' })).not.toBeInTheDocument()
+    })
+
+    await user.click(screen.getByRole('button', { name: /Agent/ }))
+
+    expect(await screen.findByText('Bruno')).toBeInTheDocument()
+    expect(screen.getByText('Planner')).toBeInTheDocument()
+    await waitFor(() => expect(queryMocks.request).toHaveBeenCalledTimes(2))
+    expect(screen.queryByText('Nadia')).not.toBeInTheDocument()
  })

  it('does not select an Agent v2 roster agent without active config snapshot', async () => {
    const user = userEvent.setup()
    const onSelect = vi.fn()
-    queryMocks.inviteOptionsQueryFn.mockResolvedValue({
-      data: [
-        {
-          id: 'agent-1',
-          name: 'Nadia',
-          description: 'Clarification Drafter',
-          active_config_snapshot_id: null,
-          role: 'Researcher',
-          agent_kind: 'dify_agent',
-          icon: 'A',
-          icon_background: '#E9D7FE',
-          icon_type: 'emoji',
-          scope: 'roster',
-          source: 'workflow',
-          status: 'active',
-        },
-      ],
-      has_more: false,
-      limit: 8,
-      page: 1,
-      total: 1,
-    })
+    mockInviteOptionsResponse([
+      createInviteOption({
+        id: 'agent-1',
+        name: 'Nadia',
+        active_config_snapshot_id: null,
+      }),
+    ])

    const queryClient = new QueryClient({
      defaultOptions: {
@@ -323,13 +388,7 @@ describe('Blocks', () => {
  it('inserts an inline Agent v2 node from the selector start action', async () => {
    const user = userEvent.setup()
    const onSelect = vi.fn()
-    queryMocks.inviteOptionsQueryFn.mockResolvedValue({
-      data: [],
-      has_more: false,
-      limit: 8,
-      page: 1,
-      total: 0,
-    })
+    mockInviteOptionsResponse([])
    const queryClient = new QueryClient({
      defaultOptions: {
        queries: {
@@ -376,13 +435,7 @@ describe('Blocks', () => {

  it('closes the agent selector when Escape closes the combobox', async () => {
    const user = userEvent.setup()
-    queryMocks.inviteOptionsQueryFn.mockResolvedValue({
-      data: [],
-      has_more: false,
-      limit: 8,
-      page: 1,
-      total: 0,
-    })
+    mockInviteOptionsResponse([])
    const queryClient = new QueryClient({
      defaultOptions: {
        queries: {
@@ -66,6 +66,7 @@ export function AgentSelectorContent({
        },
      },
    }),
+    staleTime: 0,
  })
  const agents = agentsQuery.data?.data ?? []
  const actionOptions: AgentSelectorActionOption[] = onStartFromScratch
@@ -101,6 +101,10 @@ describe('agent/default', () => {
    })
  })

+  it('reuses the legacy agent node help document', () => {
+    expect(nodeDefault.metaData.helpLinkUri).toBe('agent')
+  })
+
  it('identifies version 2 agent data as Agent v2', () => {
    expect(isAgentV2NodeData(createPayload({ type: BlockEnum.Agent }))).toBe(true)
    expect(isAgentV2NodeData({
@@ -7,6 +7,7 @@ import { hasValidAgentBinding } from './types'
 const metaData = genNodeMetaData({
  sort: 3,
  type: BlockEnum.AgentV2,
+  helpLinkUri: 'agent',
 })

 const nodeDefault: NodeDefault<AgentV2NodeType> = {
@@ -1,18 +1,33 @@
-import { render, screen } from '@testing-library/react'
+import { render, screen, waitFor } from '@testing-library/react'
 import { AgentDetailLayout } from '../layout'

+const mockReplace = vi.hoisted(() => vi.fn())
+const mockAgentQuery = vi.hoisted(() => ({
+  data: {
+    name: 'Agent',
+  } as { name: string } | undefined,
+  error: null as unknown,
+}))
+
 vi.mock('@tanstack/react-query', async (importOriginal) => {
  const actual = await importOriginal<typeof import('@tanstack/react-query')>()
  return {
    ...actual,
-    useQuery: vi.fn(() => ({
-      data: {
-        name: 'Agent',
-      },
-    })),
+    useQuery: vi.fn(() => mockAgentQuery),
  }
 })

+vi.mock('@/next/navigation', () => ({
+  useRouter: () => ({
+    back: vi.fn(),
+    forward: vi.fn(),
+    refresh: vi.fn(),
+    push: vi.fn(),
+    replace: mockReplace,
+    prefetch: vi.fn(),
+  }),
+}))
+
 vi.mock('@/hooks/use-document-title', () => ({
  default: vi.fn(),
 }))
@@ -32,6 +47,10 @@ vi.mock('@/service/client', () => ({
 describe('AgentDetailLayout', () => {
  beforeEach(() => {
    vi.clearAllMocks()
+    mockAgentQuery.data = {
+      name: 'Agent',
+    }
+    mockAgentQuery.error = null
  })

  it('should render detail content without owning navigation landmarks', () => {
@@ -45,4 +64,20 @@ describe('AgentDetailLayout', () => {
    expect(screen.queryByRole('main')).not.toBeInTheDocument()
    expect(screen.queryByRole('complementary', { name: 'Detail sidebar' })).not.toBeInTheDocument()
  })
+
+  it('should redirect to roster when agent detail returns 404', async () => {
+    mockAgentQuery.data = undefined
+    mockAgentQuery.error = new Response(null, { status: 404 })
+
+    render(
+      <AgentDetailLayout agentId="missing-agent">
+        <div>Agent detail content</div>
+      </AgentDetailLayout>,
+    )
+
+    await waitFor(() => {
+      expect(mockReplace).toHaveBeenCalledWith('/roster')
+    })
+    expect(screen.queryByText('Agent detail content')).not.toBeInTheDocument()
+  })
 })
@@ -5,9 +5,10 @@ import { WorkflowReferencesTable } from '../workflow-references-table'

 const mocks = vi.hoisted(() => ({
  queryFn: vi.fn(),
-  queryOptions: vi.fn((input: unknown) => ({
+  queryOptions: vi.fn(({ enabled = true, input }: { enabled?: boolean, input: unknown }) => ({
    queryKey: ['agent-referencing-workflows', input],
    queryFn: () => mocks.queryFn(input),
+    enabled,
  })),
 }))

@@ -31,7 +32,7 @@ vi.mock('@/hooks/use-timestamp', () => ({
  }),
 }))

-const renderTable = () => {
+const renderTable = ({ enabled }: { enabled?: boolean } = {}) => {
  const queryClient = new QueryClient({
    defaultOptions: {
      queries: {
@@ -42,7 +43,7 @@ const renderTable = () => {

  render(
    <QueryClientProvider client={queryClient}>
-      <WorkflowReferencesTable agentId="agent-1" />
+      <WorkflowReferencesTable agentId="agent-1" enabled={enabled} />
    </QueryClientProvider>,
  )

@@ -67,9 +68,27 @@ describe('WorkflowReferencesTable', () => {
              agent_id: 'agent-1',
            },
          },
+          enabled: true,
        })
      })
    })
+
+    it('should not fetch workflow references when disabled', async () => {
+      renderTable({ enabled: false })
+
+      await waitFor(() => {
+        expect(mocks.queryOptions).toHaveBeenCalledWith({
+          input: {
+            params: {
+              agent_id: 'agent-1',
+            },
+          },
+          enabled: false,
+        })
+      })
+      expect(mocks.queryFn).not.toHaveBeenCalled()
+      expect(screen.queryByText('agentV2.agentDetail.access.workflow.loading')).not.toBeInTheDocument()
+    })
  })

  describe('Rendering', () => {
@@ -12,6 +12,7 @@ import { consoleQuery } from '@/service/client'

 type WorkflowReferencesTableProps = {
  agentId: string
+  enabled?: boolean
 }

 const workflowTableColSpan = 5
@@ -20,6 +21,7 @@ const getWorkflowReferenceHref = (reference: AgentReferencingWorkflowResponse) =

 export function WorkflowReferencesTable({
  agentId,
+  enabled = true,
 }: WorkflowReferencesTableProps) {
  const { t } = useTranslation('agentV2')
  const { t: tCommon } = useTranslation('common')
@@ -29,6 +31,7 @@ export function WorkflowReferencesTable({
        agent_id: agentId,
      },
    },
+    enabled,
  }))
  const workflowReferences = workflowReferencesQuery.data?.data ?? []

@@ -62,12 +65,12 @@ export function WorkflowReferencesTable({
          </tr>
        </thead>
        <tbody className="system-sm-regular text-text-secondary">
-          {workflowReferencesQuery.isPending && (
+          {enabled && workflowReferencesQuery.isPending && (
            <WorkflowAccessStateRow>
              {t('agentDetail.access.workflow.loading')}
            </WorkflowAccessStateRow>
          )}
-          {workflowReferencesQuery.isError && (
+          {enabled && workflowReferencesQuery.isError && (
            <WorkflowAccessStateRow>
              <div className="flex items-center justify-center gap-2">
                <span>{t('agentDetail.access.workflow.loadFailed')}</span>
@@ -83,12 +86,12 @@ export function WorkflowReferencesTable({
              </div>
            </WorkflowAccessStateRow>
          )}
-          {workflowReferencesQuery.isSuccess && workflowReferences.length === 0 && (
+          {enabled && workflowReferencesQuery.isSuccess && workflowReferences.length === 0 && (
            <WorkflowAccessStateRow>
              {t('agentDetail.access.workflow.empty')}
            </WorkflowAccessStateRow>
          )}
-          {workflowReferencesQuery.isSuccess && workflowReferences.map(reference => (
+          {enabled && workflowReferencesQuery.isSuccess && workflowReferences.map(reference => (
            <WorkflowAccessRow
              key={`${reference.app_id}:${reference.workflow_id}`}
              reference={reference}
@@ -71,7 +71,7 @@ export function AgentAccessPage({
              </p>
            </div>

-            <WorkflowReferencesTable agentId={agentId} />
+            <WorkflowReferencesTable agentId={agentId} enabled={agentQuery.isSuccess} />
          </section>
        </div>
      </ScrollArea>
@@ -54,6 +54,19 @@ const mocks = vi.hoisted(() => ({
  },
 }))

+const toastMock = vi.hoisted(() => ({
+  error: vi.fn(),
+}))
+
+const modelHooksState = vi.hoisted(() => ({
+  defaultTextGenerationModel: {
+    provider: {
+      provider: 'langgenius/openai/openai',
+    },
+    model: 'gpt-4o-mini',
+  } as { provider: { provider: string }, model: string } | undefined,
+}))
+
 function createDeferredPromise<T>() {
  let resolve!: (value: T) => void
  const promise = new Promise<T>((promiseResolve) => {
@@ -101,6 +114,10 @@ vi.mock('@tanstack/react-query', async (importOriginal) => {
  }
 })

+vi.mock('@langgenius/dify-ui/toast', () => ({
+  toast: toastMock,
+}))
+
 vi.mock('@/service/client', () => ({
  consoleQuery: {
    agent: {
@@ -194,7 +211,7 @@ vi.mock('@/service/client', () => ({
 }))

 vi.mock('@/app/components/header/account-setting/model-provider-page/hooks', () => ({
-  useDefaultModel: () => ({ data: undefined }),
+  useDefaultModel: () => ({ data: modelHooksState.defaultTextGenerationModel }),
  useTextGenerationCurrentProviderAndModelAndModelList: () => ({
    textGenerationModelList: [],
  }),
@@ -271,7 +288,7 @@ vi.mock('../components/preview/build-chat', async () => {
              void props.onSaveDraftBeforeRun?.().then(() => {
                setMessageSent(true)
                props.onConversationIdChange?.('build-conversation-new')
-              })
+              }).catch(() => undefined)
            }}
          >
            send build message
@@ -359,6 +376,12 @@ vi.mock('../components/preview/versions-panel', () => ({
 describe('AgentConfigurePage', () => {
  beforeEach(() => {
    vi.clearAllMocks()
+    modelHooksState.defaultTextGenerationModel = {
+      provider: {
+        provider: 'langgenius/openai/openai',
+      },
+      model: 'gpt-4o-mini',
+    }
    mocks.refreshDebugConversation.mockResolvedValue({
      debug_conversation_has_messages: false,
      debug_conversation_id: 'debug-conversation-new',
@@ -1036,6 +1059,50 @@ describe('AgentConfigurePage', () => {
      expect(screen.getByRole('button', { name: 'discard build draft' })).toBeDisabled()
    })

+    it('should block build chat checkout when no model is configured', async () => {
+      const queryClient = new QueryClient()
+      modelHooksState.defaultTextGenerationModel = undefined
+      mocks.queryState.composer = {
+        data: {
+          agent_soul: {
+            prompt: {
+              system_prompt: 'draft prompt',
+            },
+          },
+        },
+        isFetching: false,
+        isError: false,
+        isPending: false,
+        isSuccess: true,
+        refetch: vi.fn(),
+      }
+      mocks.queryState.buildDraft = {
+        data: undefined as unknown,
+        dataUpdatedAt: 0,
+        error: new Response(null, { status: 404 }),
+        isFetching: false,
+        isError: true,
+        isPending: false,
+        isSuccess: false,
+        refetch: vi.fn(),
+      }
+
+      render(
+        <QueryClientProvider client={queryClient}>
+          <AgentConfigurePage agentId="agent-1" />
+        </QueryClientProvider>,
+      )
+
+      fireEvent.click(screen.getByRole('button', { name: 'send build message' }))
+
+      await waitFor(() => {
+        expect(toastMock.error).toHaveBeenCalledWith('common.modelProvider.selectModel')
+      })
+      expect(mocks.checkoutBuildDraft).not.toHaveBeenCalled()
+      expect(screen.getByRole('region', { name: 'build-chat' })).toHaveTextContent('sent:no')
+      expect(screen.getByRole('region', { name: 'orchestrate-panel' })).toHaveTextContent('buildDraft:no')
+    })
+
    it('should keep the build draft bar disabled while a build conversation is responding', async () => {
      vi.useFakeTimers()
      const queryClient = new QueryClient()
@@ -91,6 +91,11 @@ function setDocumentVisibilityState(visibilityState: DocumentVisibilityState) {
  })
 }

+const configuredModel = {
+  provider: 'langgenius/openai/openai',
+  model: 'gpt-4o-mini',
+}
+
 vi.mock('@langgenius/dify-ui/toast', () => ({
  toast: toastMock,
 }))
@@ -607,7 +612,9 @@ describe('useAgentConfigureSync', () => {
  })

  it('should publish only when publishDraft is called explicitly', async () => {
-    const { queryClient, result, store } = renderUseAgentConfigureSync()
+    const { queryClient, result, store } = renderUseAgentConfigureSync({
+      currentModel: configuredModel,
+    })
    const invalidateQueries = vi.spyOn(queryClient, 'invalidateQueries')
    queryClient.setQueryData(['agent-detail', 'agent-1'], {
      active_config_is_published: false,
@@ -654,12 +661,28 @@ describe('useAgentConfigureSync', () => {
    expect(toastMock.success).toHaveBeenCalledWith('common.api.actionSuccess')
  })

+  it('should toast and skip publish when no model is configured', async () => {
+    const { result, store } = renderUseAgentConfigureSync()
+
+    act(() => {
+      store.set(agentComposerDraftAtom, {
+        ...defaultAgentSoulConfigFormState,
+        prompt: 'Published prompt',
+      })
+    })
+
+    await act(async () => {
+      await result.current.publishDraft()
+    })
+
+    expect(composerPutMutationFn).not.toHaveBeenCalled()
+    expect(publishAgentMutationFn).not.toHaveBeenCalled()
+    expect(toastMock.error).toHaveBeenCalledWith('common.modelProvider.selectModel')
+  })
+
  it('should keep default model fallback from creating unpublished changes after publish', async () => {
    const { result, store } = renderUseAgentConfigureSync({
-      currentModel: {
-        provider: 'langgenius/openai/openai',
-        model: 'gpt-4o-mini',
-      },
+      currentModel: configuredModel,
    })
    act(() => {
      store.set(agentComposerDraftAtom, {
@@ -681,6 +704,7 @@ describe('useAgentConfigureSync', () => {

  it('should keep base config fallback fields from creating unpublished changes after publish', async () => {
    const { result, store } = renderUseAgentConfigureSync({
+      currentModel: configuredModel,
      baseConfig: {
        app_features: {
          file_upload: {
@@ -708,7 +732,9 @@ describe('useAgentConfigureSync', () => {
  })

  it('should publish the current draft snapshot instead of a stale caller payload', async () => {
-    const { result, store } = renderUseAgentConfigureSync()
+    const { result, store } = renderUseAgentConfigureSync({
+      currentModel: configuredModel,
+    })

    act(() => {
      store.set(agentComposerDraftAtom, {
@@ -736,7 +762,9 @@ describe('useAgentConfigureSync', () => {

  it('should reject publish and keep the publish mutation untouched when saving the draft fails', async () => {
    composerPutMutationFn.mockRejectedValueOnce(new Error('save failed'))
-    const { queryClient, result, store } = renderUseAgentConfigureSync()
+    const { queryClient, result, store } = renderUseAgentConfigureSync({
+      currentModel: configuredModel,
+    })
    queryClient.setQueryData(['agent-detail', 'agent-1'], {
      active_config_is_published: false,
      name: 'Agent',
@@ -760,7 +788,9 @@ describe('useAgentConfigureSync', () => {
  })

  it('should toast and skip publish when knowledge retrieval validation fails', async () => {
-    const { result, store } = renderUseAgentConfigureSync()
+    const { result, store } = renderUseAgentConfigureSync({
+      currentModel: configuredModel,
+    })

    act(() => {
      store.set(agentComposerDraftAtom, {
@@ -785,7 +815,9 @@ describe('useAgentConfigureSync', () => {
  })

  it('should toast metadata filtering model error when publishing with automatic metadata filtering and no model', async () => {
-    const { result, store } = renderUseAgentConfigureSync()
+    const { result, store } = renderUseAgentConfigureSync({
+      currentModel: configuredModel,
+    })

    act(() => {
      store.set(agentComposerDraftAtom, {
@@ -813,7 +845,9 @@ describe('useAgentConfigureSync', () => {
  it('should expose publishing status from the publish mutation while publish is pending', async () => {
    const publishDeferred = createDeferredPromise<PublishAgentResponse>()
    publishAgentMutationFn.mockReturnValueOnce(publishDeferred.promise)
-    const { result } = renderUseAgentConfigureSync()
+    const { result } = renderUseAgentConfigureSync({
+      currentModel: configuredModel,
+    })
    let publishPromise!: Promise<void>
    act(() => {
      publishPromise = result.current.publishDraft()
@@ -2,6 +2,7 @@

 import type { AgentAppDetailWithSite, AgentIconType, AgentSoulConfig } from '@dify/contracts/api/console/agent/types.gen'
 import type { useAgentConfigureData } from '../hooks'
+import { toast } from '@langgenius/dify-ui/toast'
 import { useMutation, useQueryClient } from '@tanstack/react-query'
 import { useAtomValue, useSetAtom } from 'jotai'
 import { ScopeProvider } from 'jotai-scope'
@@ -214,6 +215,7 @@ function AgentConfigurePageComposerContent({
    activeConfigSnapshot,
    agentSoulConfig,
  } = configureData
+  const { t: tCommon } = useTranslation('common')
  const [buildDraftActionsDisabled, setBuildDraftActionsDisabled] = useState(false)
  const [clearPreviewChat, setClearPreviewChat] = useState(false)
  const [completedBuildConversationId, setCompletedBuildConversationId] = useState<string | null>(null)
@@ -333,6 +335,7 @@ function AgentConfigurePageComposerContent({
          isBuildDraftActive={buildDraft.isActive}
          buildDraftChangedKeys={buildDraft.changedKeys}
          showPublishBar={!buildDraft.isActive}
+          workflowReferencesEnabled={agentQuery.isSuccess}
          bottomAction={showBuildDraftBar
            ? (
                <AgentBuildDraftBar
@@ -407,6 +410,11 @@ function AgentConfigurePageComposerContent({
              }}
              onSaveDraftBeforeRun={rightPanelChatMode === 'build'
                ? async () => {
+                  if (!currentModel?.provider || !currentModel.model) {
+                    toast.error(tCommon('modelProvider.selectModel'))
+                    throw new Error('Agent model is required.')
+                  }
+
                  setBuildDraftActionsDisabled(true)
                  try {
                    return await buildDraftActions.prepareBuildDraftBeforeRun()
@@ -74,8 +74,9 @@ vi.mock('@/service/client', () => ({
        },
        referencingWorkflows: {
          get: {
-            queryOptions: ({ input }: { input: { params: { agent_id: string } } }) => ({
+            queryOptions: ({ enabled = true, input }: { enabled?: boolean, input: { params: { agent_id: string } } }) => ({
              queryKey: ['agent-referencing-workflows', input],
+              enabled,
              queryFn: async () => ({
                data: (workflowReferences.fetchCount++, workflowReferences.data),
              }),
@@ -162,6 +163,7 @@ function renderPublishBar({
  selectedVersionSnapshot,
  setupStore,
  usedByAppReferences = [],
+  workflowReferencesEnabled,
 }: {
  activeConfigIsPublished?: boolean
  activeConfigSnapshot?: AgentConfigSnapshotSummaryResponse | null
@@ -173,6 +175,7 @@ function renderPublishBar({
  selectedVersionSnapshot?: AgentConfigSnapshotSummaryResponse | null
  setupStore?: (store: ReturnType<typeof createStore>) => void
  usedByAppReferences?: AgentReferencingWorkflowResponse[]
+  workflowReferencesEnabled?: boolean
 } = {}) {
  workflowReferences.data = usedByAppReferences
  const queryClient = new QueryClient({
@@ -200,6 +203,7 @@ function renderPublishBar({
          agentName="Iris"
          isPublishing={nextProps?.isPublishing ?? isPublishing}
          selectedVersionSnapshot={selectedVersionSnapshot}
+          workflowReferencesEnabled={workflowReferencesEnabled}
          onPublish={onPublish}
          onExitVersions={onExitVersions}
          onOpenVersions={vi.fn()}
@@ -407,6 +411,28 @@ describe('AgentConfigurePublishBar', () => {
    })
  })

+  it('should publish without loading workflow references when references are disabled', async () => {
+    const { onPublish } = renderPublishBar({
+      activeConfigSnapshot,
+      prompt: 'Updated system prompt',
+      usedByAppReferences: publishedReferences,
+      workflowReferencesEnabled: false,
+    })
+
+    await waitFor(() => {
+      expect(workflowReferences.fetchCount).toBe(0)
+    })
+    fireEvent.click(screen.getByRole('button', { name: /agentV2\.agentDetail\.configure\.publishBar\.publishUpdate/ }))
+
+    await waitFor(() => {
+      expect(onPublish).toHaveBeenCalledTimes(1)
+    })
+    expect(workflowReferences.fetchCount).toBe(0)
+    expect(screen.queryByRole('region', {
+      name: /agentV2\.agentDetail\.configure\.publishImpact\.title/,
+    })).not.toBeInTheDocument()
+  })
+
  it('should mark non-prompt draft changes as unpublished', () => {
    renderPublishBar({
      activeConfigSnapshot,
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Yansong Zhang	48e95bc1c2	fix: block workflow DSL export with agent nodes	2026-07-06 17:48:26 +08:00
autofix-ci[bot]andGitHub	4f777ca6e1	[autofix.ci] apply automated fixes	2026-07-06 09:33:21 +00:00
Joel	d22232d577	chore: fs download show	2026-07-06 17:27:38 +08:00
Joel	ac158193e2	chore: download and preview space	2026-07-06 16:42:59 +08:00
yyh	a7139e90e4	fix(agent-v2): allow file tree scroll chaining	2026-07-06 16:37:26 +08:00
Joel	1c3f5b10d9	fix: / portal place	2026-07-06 16:35:57 +08:00
zyssyz123andGitHub	4f039c218c	fix(agent): avoid null agent thought stream fields (#38469 )	2026-07-06 16:32:20 +08:00
yyh	7acf2a61df	fix(agent-v2): open tool picker directly from add action	2026-07-06 16:28:16 +08:00
Joel	9e92e804aa	chore: remove agent translate in roster	2026-07-06 16:13:21 +08:00
Joel	5eddb77fb2	chore: fix generated tooptip not show when hover	2026-07-06 16:03:35 +08:00
Joel	b1c4067e71	chore: files change	2026-07-06 15:52:47 +08:00
Joel	460e380fee	Improve Agent V2 configure CE notices	2026-07-06 15:52:24 +08:00
Joel	0adf95a4df	feat(agent-v2): support file and skill downloads	2026-07-06 15:51:50 +08:00
Joel	6171810ba5	fix(agent-v2): keep prompt focus after slash insert	2026-07-06 15:50:58 +08:00
林玮 (Jade Lin)andGitHub	c1d98f64bb	fix(agent-v2): agent text history serialization (#38464 )	2026-07-06 15:26:42 +08:00
yyh	afdefad6bc	Merge remote-tracking branch 'origin/main' into feat/agent-v2	2026-07-06 14:57:00 +08:00
autofix-ci[bot]andGitHub	b2160b4f56	[autofix.ci] apply automated fixes	2026-07-06 06:42:44 +00:00
yyhandGitHub	8208b786ee	docs(dify-ui): clarify radio composition stories (#38456 )	2026-07-06 06:40:17 +00:00
Joel	3f0bb9ecc1	fix: not found agent to agent list and fix ts problem	2026-07-06 14:36:17 +08:00
Joel	4328231bd5	fix(agent-v2): align build note badge	2026-07-06 14:36:17 +08:00
zyssyz123andGitHub	a9cd1aaaa2	fix: return friendly error for unpublished agent app (#38455 )	2026-07-06 14:33:51 +08:00
Joel	b62a349ab5	chore: enchance agent 404 flow	2026-07-06 14:24:43 +08:00
Joel	98b9d8feac	fix(agent-v2): insert prompt references at slash cursor	2026-07-06 14:24:43 +08:00
Joel	531095e195	chroe: get agent id 404 retry	2026-07-06 14:24:43 +08:00
Joel	c954b27d0a	feat(agent-v2): add file row downloads	2026-07-06 14:24:42 +08:00
yyhandGitHub	4779144dc2	Merge branch 'main' into feat/agent-v2	2026-07-06 13:45:49 +08:00
yyh	5e75ffa03f	test: cover agent selector option refresh	2026-07-06 13:45:13 +08:00
yyh	d8bea4f889	fix: refresh agent selector options on mount	2026-07-06 13:31:29 +08:00
Pranav Agarwal GitHub FFXN	d9c99daf29	fix(api): isolate side-effect session writes in multimodal and RAG handlers (#38210 ) Co-authored-by: FFXN <31929997+FFXN@users.noreply.github.com>	2026-07-06 05:17:12 +00:00
yyh	8e1ea9d336	test(e2e): align agent v2 publish fixtures with model guard	2026-07-06 13:16:43 +08:00
KVOJJJinandGitHub	93eb6d32b5	fix(web): update snippet placeholder icon color (#38445 )	2026-07-06 05:10:36 +00:00
zyssyz123 GitHub yyh	f8b1d9ab95	fix(agent): pass workflow files to agent node runtime (#38453 ) Co-authored-by: yyh <yuanyouhuilyz@gmail.com>	2026-07-06 13:06:45 +08:00
yyh	3f314aeeaa	fix(workflow): add agent v2 help link Refs DIFY-2705	2026-07-06 13:02:26 +08:00
dependabot[bot]GitHubdependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>	de5af6c2da	build(deps): bump the github-actions-dependencies group across 1 directory with 12 updates (#38430 ) Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>	2026-07-06 04:39:10 +00:00
i晟andGitHub	586c8de1a0	refactor: drop redundant len(tag_ids)==0 check in get_target_ids_by_tag_ids (#38447 )	2026-07-06 04:35:58 +00:00
zyssyz123andGitHub	0313c55638	fix(agent): require model before publishing roster agent (#38451 )	2026-07-06 12:02:50 +08:00
autofix-ci[bot]andGitHub	5cf2bf7044	[autofix.ci] apply automated fixes	2026-07-06 03:43:15 +00:00
zyssyz123andGitHub	e3457e013d	fix(agent): allow incomplete composer draft knowledge (#38446 )	2026-07-06 11:37:51 +08:00
autofix-ci[bot]andGitHub	59bc1a7fb2	[autofix.ci] apply automated fixes	2026-07-06 03:24:09 +00:00
zyssyz123andGitHub	e96eb2e2d5	fix(agent): enable default file upload feature (#38441 )	2026-07-06 11:17:46 +08:00
yyhandGitHub	708c962724	Merge branch 'main' into feat/agent-v2	2026-07-06 11:05:37 +08:00
yyh	0ed3a1f866	test(e2e): tighten agent v2 model guard assertion	2026-07-06 11:01:47 +08:00
yyh	e0e33feabb	fix(agent-v2): block missing model configure actions	2026-07-06 11:01:33 +08:00
yyh	bf6eecfd75	test(e2e): cover agent v2 missing model guards	2026-07-06 10:43:58 +08:00
zyssyz123andGitHub	855e653486	fix(agent): normalize empty streaming response fields (#38439 )	2026-07-06 10:29:59 +08:00
Joel	01d39e0ed4	chore: content in prompt not breakline	2026-07-06 09:32:48 +08:00