fix(ci): actions/cache@v4 加 continue-on-error——缓存故障不阻断 job

所有 5 个 actions/cache@v4 步骤加 continue-on-error: true： - validate-style pip cache - validate-security pip cache - validate-python pip cache - unit-tests pip cache - frontend-unit-test npm cache 当 Gitea Actions 缓存基础设施出现瞬态故障（git cache EOF / 归档损坏）时，cache restore 步骤会标黄但不中断 job，后续的 pip install 重试逻辑自动 fresh download 兜底。
fix(ci): Web 镜像 buildx 幽灵缓存防护——Git Tree Hash Cache Bust
2026-08-31 22:42:01 +08:00 · 2026-08-31 22:40:57 +08:00 · 2026-08-31 22:26:03 +08:00 · 2026-08-31 21:43:42 +08:00 · 2026-08-31 13:26:58 +00:00 · 2026-08-31 21:07:43 +08:00
48 changed files with 2185 additions and 402 deletions
@@ -58,10 +58,11 @@ jobs:
          echo "::warning::PR #${PR_NUMBER} 已合并，测试由合并后 push 流水线承接，PR 侧测试类 job 跳过"
          exit 0
        fi
-        # 情形2：同一 head_sha 已有在跑/排队的 push 流水线（rebase/ff 合并竞态）
+        # 情形2：同一 head_sha 已有在跑/排队/已成功的 push 流水线（rebase/ff 合并竞态；
+        #        已成功也去重——竞态窗口内两边都通过判断时，PR侧再跑全量测试纯属重复）
        DUP=$(curl -sfH "Authorization: token $GITHUB_TOKEN" \
          "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/actions/runs?head_sha=${HEAD_SHA}&per_page=30" \
-          | python3 -c "import json,sys; d=json.load(sys.stdin); runs=d if isinstance(d,list) else d.get('workflow_runs',d.get('runs',[])); hit=[r for r in runs if r.get('event')=='push' and r.get('status') in ('in_progress','queued','waiting','pending')]; print('true' if hit else 'false')" || echo false)
+          | python3 -c "import json,sys; d=json.load(sys.stdin); runs=d if isinstance(d,list) else d.get('workflow_runs',d.get('runs',[])); hit=[r for r in runs if r.get('event')=='push' and (r.get('status') in ('in_progress','queued','waiting','pending') or r.get('conclusion')=='success')]; print('true' if hit else 'false')" || echo false)
        if [ "$DUP" = "true" ]; then
          echo "skip_tests=true" >> $GITHUB_OUTPUT
          echo "reason=duplicate-push-run-active" >> $GITHUB_OUTPUT
@@ -80,12 +81,6 @@ jobs:
      skip_backend: ${{ steps.check.outputs.skip_backend }}
      skip_frontend: ${{ steps.check.outputs.skip_frontend }}
    steps:
-    - name: Checkout code
-      shell: sh
-      env:
-        GITHUB_TOKEN: ${{ github.token }}
-      run: |
-        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
    - name: Check changed files
      id: check
      shell: bash
@@ -124,14 +119,14 @@ jobs:
        [ ${{ job.status }} = "success" ] || STATUS="error"
        START_TIME=""
        [ -f /tmp/ci_job_start_time ] && START_TIME=$(cat /tmp/ci_job_start_time)
-        python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
+        curl -sfH "Authorization: token ${GITHUB_TOKEN:-$GITEA_TOKEN}" -o /tmp/_ci_trace.py "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/ci_trace_report.py?ref=${GITHUB_SHA}" 2>/dev/null && python3 /tmp/_ci_trace.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true

-  validate-code-quality:
+  validate-style:
    needs: dedupe-check
    if: always() && needs.dedupe-check.outputs.skip_tests != 'true'
-    name: Validate - Code Quality
+    name: Validate - Style
    runs-on: ci-l2
-    timeout-minutes: 8
+    timeout-minutes: 6
    env:
      PIP_CACHE_DIR: /root/.cache/pip
      PIP_NO_CACHE_DIR: ''
@@ -144,16 +139,23 @@ jobs:
        GITHUB_TOKEN: ${{ github.token }}
      run: |
        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Zombie run selfcheck
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+        GITHUB_RUN_ID: ${{ github.run_id }}
+      run: bash scripts/ci/ci_run_selfcheck.sh
    - name: Record job start time
      shell: sh
      run: bash scripts/ci/step_timer_start.sh
    - name: Cache pip dependencies
      uses: actions/cache@v4
+      continue-on-error: true
      with:
        path: /root/.cache/pip
-        key: ${{ runner.os }}-pip-codequality-${{ hashFiles('requirements*.txt') }}
+        key: ${{ runner.os }}-pip-style-${{ hashFiles('requirements*.txt') }}
        restore-keys: |
-          ${{ runner.os }}-pip-codequality-
+          ${{ runner.os }}-pip-style-
          ${{ runner.os }}-pip-
    - name: Install dependencies
      shell: sh
@@ -177,17 +179,9 @@ jobs:
          [ $i -eq 3 ] && exit 1
          sleep 5
        done
-        for i in 1 2 3; do
-          python3 -m pip install --no-binary :all: black==26.5.1 isort==8.0.1 && break
-          echo "pip install black/isort 失败，重试 $i/3..."
-          [ $i -eq 3 ] && exit 1
-          sleep 5
-        done
-    - name: Run code quality and security checks
+    - name: Run style checks
      shell: bash
-      env:
-        GITHUB_TOKEN: ${{ github.token }}
-      run: bash scripts/ci/validate_code_quality.sh
+      run: bash scripts/ci/validate_style.sh
    - name: Auto-fix formatting (black + isort)
      if: failure()
      shell: sh
@@ -203,7 +197,7 @@ jobs:
        CI_WEBHOOK_URL: ${{ secrets.CI_WEBHOOK_URL }}
      run: |
        set +e
-        FAILED_JOB="Validate - Code Quality" python3 scripts/ci_notify_failure.py
+        FAILED_JOB="Validate - Style" python3 scripts/ci_notify_failure.py
    - name: Job duration summary
      if: always()
      shell: sh
@@ -216,7 +210,7 @@ jobs:
        CI_NOTIFY_WEBHOOK: ${{ secrets.CI_NOTIFY_WEBHOOK }}
      run: |
        set +e
-        NOTIFY_MODE=failure JOB_NAME="Validate - Code Quality" python3 scripts/ci_notify.py
+        NOTIFY_MODE=failure JOB_NAME="Validate - Style" python3 scripts/ci_notify.py
    - name: Report CI trace
      if: always()
      shell: sh
@@ -229,12 +223,16 @@ jobs:
        [ -f /tmp/ci_job_start_time ] && START_TIME=$(cat /tmp/ci_job_start_time)
        python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true

-  validate-type-check:
+
+  validate-security:
    needs: dedupe-check
    if: always() && needs.dedupe-check.outputs.skip_tests != 'true'
-    name: Validate - Type Check (mypy)
+    name: Validate - Security
    runs-on: ci-l2
    timeout-minutes: 8
+    env:
+      PIP_CACHE_DIR: /root/.cache/pip
+      PIP_NO_CACHE_DIR: ''
    permissions:
      contents: read
    steps:
@@ -244,9 +242,125 @@ jobs:
        GITHUB_TOKEN: ${{ github.token }}
      run: |
        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Zombie run selfcheck
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+        GITHUB_RUN_ID: ${{ github.run_id }}
+      run: bash scripts/ci/ci_run_selfcheck.sh
    - name: Record job start time
      shell: sh
      run: bash scripts/ci/step_timer_start.sh
+    - name: Cache pip dependencies
+      uses: actions/cache@v4
+      continue-on-error: true
+      with:
+        path: /root/.cache/pip
+        key: ${{ runner.os }}-pip-security-${{ hashFiles('requirements*.txt') }}
+        restore-keys: |
+          ${{ runner.os }}-pip-security-
+          ${{ runner.os }}-pip-
+    - name: Install dependencies
+      shell: sh
+      run: |
+        set -eu
+        for i in 1 2 3; do
+          python3 -m pip install -q -r requirements-base.txt && break
+          echo "pip install requirements-base.txt 失败，重试 $i/3..."
+          [ $i -eq 3 ] && exit 1
+          sleep 5
+        done
+        for i in 1 2 3; do
+          python3 -m pip install -q -r requirements.txt && break
+          echo "pip install requirements.txt 失败，重试 $i/3..."
+          [ $i -eq 3 ] && exit 1
+          sleep 5
+        done
+        for i in 1 2 3; do
+          python3 -m pip install -q -r requirements-dev.txt && break
+          echo "pip install requirements-dev.txt 失败，重试 $i/3..."
+          [ $i -eq 3 ] && exit 1
+          sleep 5
+        done
+    - name: Run security checks
+      shell: bash
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+      run: bash scripts/ci/validate_security.sh
+    - name: CI failure notification
+      if: failure()
+      shell: sh
+      env:
+        GITEA_TOKEN: ${{ secrets.GITEA_TOKEN }}
+        CI_WEBHOOK_URL: ${{ secrets.CI_WEBHOOK_URL }}
+      run: |
+        set +e
+        FAILED_JOB="Validate - Security" python3 scripts/ci_notify_failure.py
+    - name: Job duration summary
+      if: always()
+      shell: sh
+      run: bash scripts/ci/step_timer_end.sh
+    - name: Notify on failure
+      continue-on-error: true
+      if: failure()
+      shell: sh
+      env:
+        CI_NOTIFY_WEBHOOK: ${{ secrets.CI_NOTIFY_WEBHOOK }}
+      run: |
+        set +e
+        NOTIFY_MODE=failure JOB_NAME="Validate - Security" python3 scripts/ci_notify.py
+    - name: Report CI trace
+      if: always()
+      shell: sh
+      env:
+        AGENTLOOP_LICENSE_KEY: ${{ secrets.AGENTLOOP_LICENSE_KEY }}
+      run: |
+        STATUS="ok"
+        [ ${{ job.status }} = "success" ] || STATUS="error"
+        START_TIME=""
+        [ -f /tmp/ci_job_start_time ] && START_TIME=$(cat /tmp/ci_job_start_time)
+        python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
+
+
+  validate-python:
+    needs: dedupe-check
+    if: always() && needs.dedupe-check.outputs.skip_tests != 'true'
+    name: Validate - Python (mypy + alembic)
+    runs-on: ci-l2
+    timeout-minutes: 10
+    env:
+      PIP_CACHE_DIR: /root/.cache/pip
+      PIP_NO_CACHE_DIR: ''
+      DATABASE_URL: postgresql+psycopg://postgres:postgres@host.docker.internal:5432/xiaoxia_saas
+      USE_IN_MEMORY_DB: 'false'
+      CI_USE_SHARED_PG: 'true'
+    permissions:
+      contents: read
+    steps:
+    - name: Checkout code
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+      run: |
+        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Zombie run selfcheck
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+        GITHUB_RUN_ID: ${{ github.run_id }}
+      run: bash scripts/ci/ci_run_selfcheck.sh
+    - name: Record job start time
+      shell: sh
+      run: bash scripts/ci/step_timer_start.sh
+    - name: Cache pip dependencies
+      uses: actions/cache@v4
+      continue-on-error: true
+      with:
+        path: /root/.cache/pip
+        key: ${{ runner.os }}-pip-python-${{ hashFiles('requirements*.txt') }}
+        restore-keys: |
+          ${{ runner.os }}-pip-python-
+          ${{ runner.os }}-pip-
    - name: Install dependencies
      shell: sh
      run: |
@@ -272,6 +386,9 @@ jobs:
    - name: Run mypy type check
      shell: bash
      run: bash scripts/ci/validate_mypy.sh
+    - name: Run alembic migration validation
+      shell: bash
+      run: bash scripts/ci/validate_migration.sh
    - name: CI failure notification
      if: failure()
      shell: sh
@@ -280,7 +397,7 @@ jobs:
        CI_WEBHOOK_URL: ${{ secrets.CI_WEBHOOK_URL }}
      run: |
        set +e
-        FAILED_JOB="Validate - Type Check (mypy)" python3 scripts/ci_notify_failure.py
+        FAILED_JOB="Validate - Python (mypy + alembic)" python3 scripts/ci_notify_failure.py
    - name: Job duration summary
      if: always()
      shell: sh
@@ -293,7 +410,7 @@ jobs:
        CI_NOTIFY_WEBHOOK: ${{ secrets.CI_NOTIFY_WEBHOOK }}
      run: |
        set +e
-        NOTIFY_MODE=failure JOB_NAME="Validate - Type Check (mypy)" python3 scripts/ci_notify.py
+        NOTIFY_MODE=failure JOB_NAME="Validate - Python (mypy + alembic)" python3 scripts/ci_notify.py
    - name: Report CI trace
      if: always()
      shell: sh
@@ -306,86 +423,6 @@ jobs:
        [ -f /tmp/ci_job_start_time ] && START_TIME=$(cat /tmp/ci_job_start_time)
        python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true

-  validate-migration:
-    needs: dedupe-check
-    if: always() && needs.dedupe-check.outputs.skip_tests != 'true'
-    name: Validate - Migration (alembic)
-    runs-on: ci-l2
-    timeout-minutes: 8
-    permissions:
-      contents: read
-    env:
-      DATABASE_URL: postgresql+psycopg://postgres:postgres@host.docker.internal:5432/xiaoxia_saas
-      USE_IN_MEMORY_DB: 'false'
-      CI_USE_SHARED_PG: 'true'
-    steps:
-    - name: Checkout code
-      shell: sh
-      env:
-        GITHUB_TOKEN: ${{ github.token }}
-      run: |
-        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
-    - name: Record job start time
-      shell: sh
-      run: bash scripts/ci/step_timer_start.sh
-    - name: Install dependencies
-      shell: sh
-      run: |
-        set -eu
-        for i in 1 2 3; do
-          python3 -m pip install -q -r requirements-base.txt && break
-          echo "pip install requirements-base.txt 失败，重试 $i/3..."
-          [ $i -eq 3 ] && exit 1
-          sleep 5
-        done
-        for i in 1 2 3; do
-          python3 -m pip install -q -r requirements.txt && break
-          echo "pip install requirements.txt 失败，重试 $i/3..."
-          [ $i -eq 3 ] && exit 1
-          sleep 5
-        done
-        for i in 1 2 3; do
-          python3 -m pip install -q -r requirements-dev.txt && break
-          echo "pip install requirements-dev.txt 失败，重试 $i/3..."
-          [ $i -eq 3 ] && exit 1
-          sleep 5
-        done
-    - name: Run alembic migration validation
-      shell: bash
-      run: bash scripts/ci/validate_migration.sh
-    - name: CI failure notification
-      if: failure()
-      shell: sh
-      env:
-        GITEA_TOKEN: ${{ secrets.GITEA_TOKEN }}
-        CI_WEBHOOK_URL: ${{ secrets.CI_WEBHOOK_URL }}
-      run: |
-        set +e
-        FAILED_JOB="Validate - Migration (alembic)" python3 scripts/ci_notify_failure.py
-    - name: Job duration summary
-      if: always()
-      shell: sh
-      run: bash scripts/ci/step_timer_end.sh
-    - name: Notify on failure
-      continue-on-error: true
-      if: failure()
-      shell: sh
-      env:
-        CI_NOTIFY_WEBHOOK: ${{ secrets.CI_NOTIFY_WEBHOOK }}
-      run: |
-        set +e
-        NOTIFY_MODE=failure JOB_NAME="Validate - Migration (alembic)" python3 scripts/ci_notify.py
-    - name: Report CI trace
-      if: always()
-      shell: sh
-      env:
-        AGENTLOOP_LICENSE_KEY: ${{ secrets.AGENTLOOP_LICENSE_KEY }}
-      run: |
-        STATUS="ok"
-        [ ${{ job.status }} = "success" ] || STATUS="error"
-        START_TIME=""
-        [ -f /tmp/ci_job_start_time ] && START_TIME=$(cat /tmp/ci_job_start_time)
-        python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true

  unit-tests:
    needs: [check-frontend-only, dedupe-check]
@@ -408,6 +445,12 @@ jobs:
        GITHUB_TOKEN: ${{ github.token }}
      run: |
        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Zombie run selfcheck
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+        GITHUB_RUN_ID: ${{ github.run_id }}
+      run: bash scripts/ci/ci_run_selfcheck.sh
    - name: Record job start time
      shell: sh
      run: bash scripts/ci/step_timer_start.sh
@@ -416,6 +459,7 @@ jobs:
      run: bash scripts/ci/step_install_ffmpeg.sh
    - name: Cache pip dependencies
      uses: actions/cache@v4
+      continue-on-error: true
      with:
        path: /root/.cache/pip
        key: ${{ runner.os }}-pip-unittests-${{ hashFiles('requirements*.txt') }}
@@ -470,9 +514,6 @@ jobs:
    needs:
      - check-frontend-only
      - dedupe-check
-      - validate-code-quality
-      - validate-type-check
-      - validate-migration
    env:
      DATABASE_URL: postgresql+psycopg://postgres:postgres@host.docker.internal:5432/xiaoxia_saas
      USE_IN_MEMORY_DB: 'false'
@@ -488,6 +529,12 @@ jobs:
        GITHUB_TOKEN: ${{ github.token }}
      run: |
        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Zombie run selfcheck
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+        GITHUB_RUN_ID: ${{ github.run_id }}
+      run: bash scripts/ci/ci_run_selfcheck.sh
    - name: Record job start time
      shell: sh
      run: bash scripts/ci/step_timer_start.sh
@@ -542,6 +589,12 @@ jobs:
        GITHUB_TOKEN: ${{ github.token }}
      run: |
        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Zombie run selfcheck
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+        GITHUB_RUN_ID: ${{ github.run_id }}
+      run: bash scripts/ci/ci_run_selfcheck.sh
    - name: Record job start time
      shell: sh
      run: bash scripts/ci/step_timer_start.sh
@@ -604,11 +657,18 @@ jobs:
        GITHUB_TOKEN: ${{ github.token }}
      run: |
        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Zombie run selfcheck
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+        GITHUB_RUN_ID: ${{ github.run_id }}
+      run: bash scripts/ci/ci_run_selfcheck.sh
    - name: Record job start time
      shell: sh
      run: bash scripts/ci/step_timer_start.sh
    - name: Cache npm dependencies
      uses: actions/cache@v4
+      continue-on-error: true
      with:
        path: /root/.npm
        key: ${{ runner.os }}-npm-vitest-${{ hashFiles('apps/web/package-lock.json') }}
@@ -695,6 +755,12 @@ jobs:
        GITHUB_TOKEN: ${{ github.token }}
      run: |
        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Zombie run selfcheck
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+        GITHUB_RUN_ID: ${{ github.run_id }}
+      run: bash scripts/ci/ci_run_selfcheck.sh
    - name: Record job start time
      shell: sh
      run: bash scripts/ci/step_timer_start.sh
@@ -809,19 +875,13 @@ jobs:
      skip_backend: ${{ steps.check.outputs.skip_backend }}
      skip_frontend: ${{ steps.check.outputs.skip_frontend }}
    steps:
-    - name: Checkout code
-      shell: sh
-      env:
-        GITHUB_TOKEN: ${{ github.token }}
-      run: |
-        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
    - name: Check changed paths
      id: check
      shell: bash
      env:
        GITHUB_TOKEN: ${{ github.token }}
      run: |
-        bash scripts/ci/ci_push_paths.sh
+        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_push_paths.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/ci_push_paths.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_push_paths.sh
    - name: Report CI trace
      if: always()
      shell: sh
@@ -832,7 +892,7 @@ jobs:
        [ ${{ job.status }} = "success" ] || STATUS="error"
        START_TIME=""
        [ -f /tmp/ci_job_start_time ] && START_TIME=$(cat /tmp/ci_job_start_time)
-        python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
+        curl -sfH "Authorization: token ${GITHUB_TOKEN:-$GITEA_TOKEN}" -o /tmp/_ci_trace.py "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/ci_trace_report.py?ref=${GITHUB_SHA}" 2>/dev/null && python3 /tmp/_ci_trace.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true

  build-staging:
    name: Build Staging ${{ matrix.service_display }} Image
@@ -873,6 +933,12 @@ jobs:
        GITHUB_TOKEN: ${{ github.token }}
      run: |
        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Zombie run selfcheck
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+        GITHUB_RUN_ID: ${{ github.run_id }}
+      run: bash scripts/ci/ci_run_selfcheck.sh
    - name: Record job start time
      shell: sh
      run: bash scripts/ci/step_timer_start.sh
@@ -1084,9 +1150,7 @@ jobs:
    - check-push-paths
    - build-staging
    - retag-staging-skipped
-    # 显式 success() 状态检查：上游 build/retag 被路径过滤 if 跳过(skipped)时不阻塞本 job；
-    # 上游真正失败时仍然阻断（act_runner 对无状态函数的 if 隐式包 success()，纯 skipped 也会连带跳过）
-    if: success() && github.event_name == 'push' && (github.ref_name == 'main' || github.ref_name == 'develop')
+    if: (!cancelled()) && github.event_name == 'push' && (github.ref_name == 'main' || github.ref_name == 'develop')
    steps:
    - name: Checkout code
      shell: sh
@@ -1246,6 +1310,12 @@ jobs:
        GITHUB_TOKEN: ${{ github.token }}
      run: |
        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Zombie run selfcheck
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+        GITHUB_RUN_ID: ${{ github.run_id }}
+      run: bash scripts/ci/ci_run_selfcheck.sh
    - name: Record job start time
      shell: sh
      run: bash scripts/ci/step_timer_start.sh
@@ -1293,6 +1363,12 @@ jobs:
        GITHUB_TOKEN: ${{ github.token }}
      run: |
        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Zombie run selfcheck
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+        GITHUB_RUN_ID: ${{ github.run_id }}
+      run: bash scripts/ci/ci_run_selfcheck.sh
    - name: Record job start time
      shell: sh
      run: bash scripts/ci/step_timer_start.sh
@@ -1332,8 +1408,9 @@ jobs:
    runs-on: runtime-builder
    timeout-minutes: ${{ matrix.timeout }}
    needs:
-      - validate-code-quality
-      - validate-type-check
+      - validate-style
+      - validate-security
+      - validate-python
      - unit-tests
      - frontend-lint
      - frontend-unit-test
@@ -1367,6 +1444,12 @@ jobs:
        GITHUB_TOKEN: ${{ github.token }}
      run: |
        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Zombie run selfcheck
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+        GITHUB_RUN_ID: ${{ github.run_id }}
+      run: bash scripts/ci/ci_run_selfcheck.sh
    - name: Record job start time
      shell: sh
      run: bash scripts/ci/step_timer_start.sh
@@ -1813,9 +1896,9 @@ jobs:
    if: always() && github.event_name == 'pull_request'
    needs:
      - check-frontend-only
-      - validate-code-quality
-      - validate-type-check
-      - validate-migration
+      - validate-style
+      - validate-security
+      - validate-python
      - unit-tests
      - integration-tests
      - frontend-lint
@@ -1823,14 +1906,6 @@ jobs:
      - build-pr
    timeout-minutes: 3
    steps:
-      - name: Checkout code
-        shell: sh
-        env:
-          GITHUB_TOKEN: ${{ github.token }}
-        run: |
-          curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh \
-            "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
-
      - name: Evaluate CI Gate
        id: gate
        shell: bash
@@ -1839,9 +1914,9 @@ jobs:
          PR_NUMBER: ${{ github.event.pull_request.number }}
          PR_HEAD_SHA: ${{ github.event.pull_request.head.sha }}
          RESULT_CHECK_FRONTEND: ${{ needs.check-frontend-only.result }}
-          RESULT_CODE_QUALITY: ${{ needs.validate-code-quality.result }}
-          RESULT_TYPE_CHECK: ${{ needs.validate-type-check.result }}
-          RESULT_MIGRATION: ${{ needs.validate-migration.result }}
+          RESULT_STYLE: ${{ needs.validate-style.result }}
+          RESULT_SECURITY: ${{ needs.validate-security.result }}
+          RESULT_PYTHON: ${{ needs.validate-python.result }}
          RESULT_UNIT_TESTS: ${{ needs.unit-tests.result }}
          RESULT_INTEGRATION: ${{ needs.integration-tests.result }}
          RESULT_FRONTEND_LINT: ${{ needs.frontend-lint.result }}
@@ -1853,9 +1928,9 @@ jobs:
          echo ""
          echo "各job结果:"
          echo "  check-frontend-only: $RESULT_CHECK_FRONTEND"
-          echo "  validate-code-quality: $RESULT_CODE_QUALITY"
-          echo "  validate-type-check: $RESULT_TYPE_CHECK"
-          echo "  validate-migration: $RESULT_MIGRATION"
+          echo "  validate-style: $RESULT_STYLE"
+          echo "  validate-security: $RESULT_SECURITY"
+          echo "  validate-python: $RESULT_PYTHON"
          echo "  unit-tests: $RESULT_UNIT_TESTS"
          echo "  integration-tests: $RESULT_INTEGRATION"
          echo "  frontend-lint: $RESULT_FRONTEND_LINT"
@@ -1895,9 +1970,9 @@ jobs:
          # 必填检查项（根据PR类型决定）
          # 通用检查（所有PR都必须过）
          REQUIRED_GENERAL=(
-            "validate-code-quality:$RESULT_CODE_QUALITY"
-            "validate-type-check:$RESULT_TYPE_CHECK"
-            "validate-migration:$RESULT_MIGRATION"
+            "validate-style:$RESULT_STYLE"
+            "validate-security:$RESULT_SECURITY"
+            "validate-python:$RESULT_PYTHON"
            "frontend-lint:$RESULT_FRONTEND_LINT"
            "build-pr:$RESULT_BUILD_PR"
            "ai-code-review:$AI_REVIEW_STATUS"
@@ -1992,4 +2067,4 @@ jobs:
          [ "${{ steps.gate.outputs.gate_result }}" = "success" ] || STATUS="error"
          START_TIME=""
          [ -f /tmp/ci_job_start_time ] && START_TIME=$(cat /tmp/ci_job_start_time)
-          python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
+          curl -sfH "Authorization: token ${GITHUB_TOKEN:-$GITEA_TOKEN}" -o /tmp/_ci_trace.py "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/ci_trace_report.py?ref=${GITHUB_SHA}" 2>/dev/null && python3 /tmp/_ci_trace.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
@@ -2,13 +2,13 @@ name: CI Trigger Monitor

 on:
  schedule:
-    - cron: '*/5 * * * *'  # 每5分钟检查一次
+    - cron: '*/10 * * * *'  # 每10分钟检查一次（与pr-auto-scan同步降频）
  workflow_dispatch:
    inputs:
      stale_threshold:
        description: 'CI未触发告警阈值（分钟）'
        required: false
-        default: '5'
+        default: '10'

 permissions:
  contents: read
@@ -3,7 +3,7 @@ name: PR Auto Scan
 # 作为短作业模式的兜底，防止事件驱动遗漏
 on:
  schedule:
-    - cron: "*/5 * * * *"  # 每5分钟扫描一次
+    - cron: "*/10 * * * *"  # 每10分钟扫描一次（脚本自带240s墙钟上限，降频减负）
  workflow_dispatch:

 permissions:
@@ -0,0 +1,23 @@
+"""add duplicate_rate to generated_videos
+
+Revision ID: 059_duplicate_rate
+Revises: 058_uq_asset_lib_project_kind
+Create Date: 2026-08-31
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "059_duplicate_rate"
+down_revision = "058_uq_asset_lib_project_kind"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column("generated_videos", sa.Column("duplicate_rate", sa.Float(), nullable=True))
+
+
+def downgrade() -> None:
+    op.drop_column("generated_videos", "duplicate_rate")
@@ -0,0 +1,57 @@
+"""migrate template_segments data to template_clip_configs
+
+Revision ID: 060_migrate_segments
+Revises: 059_duplicate_rate
+Create Date: 2026-08-31
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "060_migrate_segments"
+down_revision = "059_duplicate_rate"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    dialect = op.get_bind().dialect.name
+
+    if dialect == "postgresql":
+        config_expr = (
+            "CASE WHEN s.material_type IS NOT NULL AND s.material_type != '' "
+            "THEN json_build_object('material_type', s.material_type)::jsonb "
+            "ELSE '{}'::jsonb END"
+        )
+        empty_json = "'{}'::jsonb"
+    else:
+        config_expr = (
+            "CASE WHEN s.material_type IS NOT NULL AND s.material_type != '' "
+            "THEN JSON_OBJECT('material_type', s.material_type) "
+            "ELSE '{}' END"
+        )
+        empty_json = "'{}'"
+
+    sql_str = (
+        "INSERT INTO template_clip_configs "
+        '(id, template_id, clip_type, "order", min_duration, max_duration, '
+        "text_template, material_requirements, transition_effect, config, "
+        "created_at, updated_at) "
+        "SELECT "
+        "s.id, s.template_id, 'main', s.segment_order, "
+        "s.duration_min, s.duration_max, "
+        "'', " + empty_json + ", "
+        "'cut', " + config_expr + ", "
+        "s.created_at, s.updated_at "
+        "FROM template_segments s "
+        "WHERE NOT EXISTS ("
+        "  SELECT 1 FROM template_clip_configs c "
+        "  WHERE c.template_id = s.template_id"
+        ")"
+    )
+    op.execute(sa.text(sql_str))
+
+
+def downgrade() -> None:
+    pass
@@ -26,7 +26,7 @@ from app.schemas.asset import (
    UpdateAssetReviewRequest,
 )
 from app.schemas.tag import TagAssetsRequest
-from app.services.asset_segment_tracker import compute_asset_availability
+from app.services.asset_segment_tracker import compute_asset_availability, get_asset_recent_use_counts
 from fastapi import APIRouter, Depends, HTTPException, Query, Response

 from packages.domain.smart_match import smart_select_assets
@@ -576,9 +576,7 @@ def smart_match_assets(
            request.library_id, request.kind, status=["ready"], limit=10000
        )
    else:
-        filtered_assets = asset_repository.find_by_library(
-            request.library_id, status=["ready"], limit=10000
-        )
+        filtered_assets = asset_repository.find_by_library(request.library_id, status=["ready"], limit=10000)
    total_candidates = len(filtered_assets)

    # 调用统一智能选素材算法（kind 已在 DB 层过滤，无需重复过滤）
@@ -610,9 +608,43 @@ def smart_match_assets(
            continue
        filtered_results.append(r)

+    # 高频使用排除：同一素材在最近 5 个视频中出现超过 3 次则排除
+    MAX_RECENT_USE_COUNT = 3
+    if filtered_results:
+        asset_ids = [getattr(r.asset, "id", "") for r in filtered_results if getattr(r.asset, "id", "")]
+        if asset_ids:
+            try:
+                use_counts = get_asset_recent_use_counts(
+                    db=asset_repository.session,
+                    asset_ids=asset_ids,
+                    recent_video_count=5,
+                )
+                high_use_excluded = set()
+                for r in filtered_results:
+                    aid = getattr(r.asset, "id", "")
+                    count = use_counts.get(aid, 0)
+                    if count > MAX_RECENT_USE_COUNT:
+                        logger.info(
+                            "smart-match 排除高频使用素材: asset_id=%s use_count=%d limit=%d",
+                            aid, count, MAX_RECENT_USE_COUNT,
+                        )
+                        high_use_excluded.add(id(r))
+                    else:
+                        pass
+                # 如果排除后不够 limit，放宽到不限制
+                remaining = [r for r in filtered_results if id(r) not in high_use_excluded]
+                if len(remaining) >= request.limit:
+                    filtered_results = remaining
+                else:
+                    logger.info("smart-match 高频排除后素材不足(%d<%d)，保留全部", len(remaining), request.limit)
+            except Exception:
+                logger.warning("smart-match 高频使用查询失败，跳过排除", exc_info=True)
+
+    # 扁平结构：SmartMatchItem 继承 AssetResponse，素材字段直接在条目顶层，
+    # 前端无需解析 item.asset 包装层，item.id / item.usable / 余量字段直接可读
    items = [
        SmartMatchItem(
-            asset=_to_asset_response(r.asset),
+            **_to_asset_response(r.asset).model_dump(),
            score=r.score,
            breakdown=r.breakdown,
        )
@@ -367,7 +367,6 @@ def batch_delete_editor_clips(
    return ClipBatchDeleteResponse(deleted_count=deleted, plan_id=plan_id)


-
 def _safe_segment_duration(value, default: float) -> float:
    """安全地将数据库中的时长值转换为正浮点数.

@@ -403,9 +402,7 @@ def _get_template_segments(
        if clip_configs:
            result = []
            for cc in clip_configs:
-                dur_min = _safe_segment_duration(
-                    cc.min_duration, _DEFAULT_EDITOR_CLIP_DURATION
-                )
+                dur_min = _safe_segment_duration(cc.min_duration, _DEFAULT_EDITOR_CLIP_DURATION)
                dur_max = _safe_segment_duration(
                    cc.max_duration or cc.min_duration,
                    _DEFAULT_EDITOR_CLIP_DURATION,
@@ -492,7 +489,7 @@ def _get_mediakit_recommendations(
        prompt = (
            "请分析每段视频，找出最精彩的5秒片段应该从哪个时间点开始。"
            "考虑因素：画面清晰度、主体是否明确、是否有明显的动作或场景变化。"
-            '请严格以JSON数组格式返回，不要包含其他文字：'
+            "请严格以JSON数组格式返回，不要包含其他文字："
            '[{"asset_id": "素材ID", "recommended_start_time": 12.5, "reason": "原因"}]'
        )

@@ -545,9 +542,7 @@ def _get_mediakit_recommendations(

            # 尝试正则提取
            if not parsed:
-                time_match = re.search(
-                    r'recommended_start_time["\s:]+([\d.]+)', content_text
-                )
+                time_match = re.search(r'recommended_start_time["\s:]+([\d.]+)', content_text)
                if time_match:
                    try:
                        recommendations[asset_id] = float(time_match.group(1))
@@ -598,14 +593,17 @@ def create_clips_from_assets_editor(
            detail="模板没有片段配置，无法创建片段",
        )

-    if not body.asset_ids:
+    # 防御：schema validator 已过滤 null/空串，这里再归一化一次，
+    # 避免异常入参（undefined → null）导致后续 /assets/{id} 404 / 422
+    asset_ids = [str(aid).strip() for aid in (body.asset_ids or []) if isinstance(aid, str) and aid.strip()]
+    if not asset_ids:
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail="素材列表为空，无法创建片段",
        )

    # 2. 获取素材实际时长（去重查询）
-    unique_asset_ids = list(dict.fromkeys(body.asset_ids))
+    unique_asset_ids = list(dict.fromkeys(asset_ids))
    asset_durations: dict[str, float] = {}
    for asset_id in unique_asset_ids:
        asset = asset_repo.get(asset_id)
@@ -615,9 +613,7 @@ def create_clips_from_assets_editor(
    # 3. 在内存中计算所有片段数据（使用随机起始时间，不调用MediaKit）
    # 读取素材 metadata 中持久化的历史已用区间（跨任务/跨调用去重），
    # 格式与 _calc_random_start_time 的 used_segments 参数一致
-    used_segments: dict[str, list[tuple[float, float]]] = get_used_segments(
-        db, unique_asset_ids
-    )
+    used_segments: dict[str, list[tuple[float, float]]] = get_used_segments(db, unique_asset_ids)
    # 受控复用回调：可用区间耗尽时复用最久未用且未达复用上限（3次）的历史区间，
    # 复用片段时长累加到 reused_durations 供 15% 占比控制
    reused_durations: dict[str, float] = {}
@@ -655,9 +651,9 @@ def create_clips_from_assets_editor(
        asset_id = ""
        clip_duration = 0.0
        start_time: float | None = None
-        n_assets = len(body.asset_ids)
+        n_assets = len(asset_ids)
        for offset in range(n_assets):
-            candidate = body.asset_ids[(i + offset) % n_assets]
+            candidate = asset_ids[(i + offset) % n_assets]
            candidate_total = asset_durations.get(candidate, 0.0)
            if candidate_total <= 0:
                continue
@@ -701,18 +697,12 @@ def create_clips_from_assets_editor(
            )

        # 记录已使用时间段（内存，供本次后续片段避开）
-        used_segments.setdefault(asset_id, []).append(
-            (start_time, start_time + clip_duration)
-        )
-        asset_assigned_durations[asset_id] = (
-            asset_assigned_durations.get(asset_id, 0.0) + clip_duration
-        )
+        used_segments.setdefault(asset_id, []).append((start_time, start_time + clip_duration))
+        asset_assigned_durations[asset_id] = asset_assigned_durations.get(asset_id, 0.0) + clip_duration
        # 同步写入素材 metadata（不 commit，与下方 replace_all_clips_transactional
        # 处于同一事务，任一步失败整体回滚，不留脏数据）；
        # 复用区间与历史记录高度重叠时 record 内部自动累加 use_count
-        record_used_segments(
-            db, asset_id, start_time, start_time + clip_duration, plan_id
-        )
+        record_used_segments(db, asset_id, start_time, start_time + clip_duration, plan_id)

        clips_data.append(
            {
@@ -803,18 +793,14 @@ def _update_mediakit_recommendations_async(  # pragma: no cover

        # 批量预加载所有涉及的素材（消除 N+1 查询）
        unique_asset_ids = list({getattr(c, "asset_id", "") or "" for c in clips} - {""})
-        assets_map: dict[str, object] = {
-            a.id: a for a in asset_repo.find_by_ids(unique_asset_ids)
-        }
+        assets_map: dict[str, object] = {a.id: a for a in asset_repo.find_by_ids(unique_asset_ids)}

        # 按 asset_id 预分组片段时间段（消除 O(N^2) 嵌套循环）
        clips_by_asset: dict[str, list[tuple[str, float, float]]] = defaultdict(list)
        for clip in clips:
            aid = getattr(clip, "asset_id", "") or ""
            if aid and clip.start_time is not None:
-                clips_by_asset[aid].append(
-                    (clip.id, clip.start_time, clip.start_time + clip.duration)
-                )
+                clips_by_asset[aid].append((clip.id, clip.start_time, clip.start_time + clip.duration))

        # 读取素材全部历史已用区间（跨任务/跨 plan 持久化记录）：
        # MediaKit 挪点必须与随机选片一样避让历史区间，否则会把片段挪回已用过的画面
@@ -861,6 +847,7 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
                if cid != clip.id and cid not in updated_clip_ids
            ]
            other_segments.extend(updated_segments.get(asset_id, []))
+
            # 并入该素材全部历史已用区间（含其他 plan/其他任务），set 去重：
            # 本 plan 片段创建时已写入历史记录
            # 并入该素材全部历史已用区间（含其他 plan/其他任务）。
@@ -869,9 +856,7 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
            def _norm(segs):
                return {(round(float(a), 3), round(float(b), 3)) for a, b in segs}

-            other_segments = list(
-                _norm(other_segments) | _norm(historical_segments.get(asset_id, []))
-            )
+            other_segments = list(_norm(other_segments) | _norm(historical_segments.get(asset_id, [])))

            # 检查推荐时间是否与同 plan 片段或历史已用区间冲突（含 0.3s 边缘间隙）：
            # 冲突时放弃该推荐、保留原随机起点（不硬挪到已用过的画面）
@@ -893,9 +878,7 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
                # 保证 clip.start_time 与 metadata.used_time_ranges 不出现不一致。
                plan_svc.update_clip(clip.id, start_time=recommended_start)
                try:
-                    if remove_used_segment(
-                        db, asset_id, old_start, old_end, plan_id=plan_id
-                    ):
+                    if remove_used_segment(db, asset_id, old_start, old_end, plan_id=plan_id):
                        record_used_segments(
                            db,
                            asset_id,
@@ -915,18 +898,14 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
                updated_count += 1
                updated_clip_ids.add(clip.id)
            except Exception as ue:
-                logger.warning(
-                    "后台任务: 单个片段更新失败: clip_id=%s error=%s", clip.id, ue
-                )
+                logger.warning("后台任务: 单个片段更新失败: clip_id=%s error=%s", clip.id, ue)
                try:
                    db.rollback()
                except Exception:
                    pass
                continue

-            updated_segments.setdefault(asset_id, []).append(
-                (recommended_start, recommended_start + clip_duration)
-            )
+            updated_segments.setdefault(asset_id, []).append((recommended_start, recommended_start + clip_duration))
            logger.info(
                "后台任务: 更新片段起始时间: clip_id=%s asset_id=%s start_time=%.2f",
                clip.id,
@@ -950,4 +929,3 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
                db.close()
            except Exception:
                pass
-
@@ -167,7 +167,18 @@ class ClipsFromAssetsRequest(BaseModel):

    asset_ids: List[str] = Field(..., min_length=1, max_length=200, description="素材 ID 列表，按顺序追加到时间线末尾")
    clip_type: str = Field(default="main", description="片段类型，默认 main")
-    required_clips_count: Optional[int] = Field(default=None, ge=1, le=200, description="要求创建的片段数量；不传则等于素材数量")
+    required_clips_count: Optional[int] = Field(
+        default=None, ge=1, le=200, description="要求创建的片段数量；不传则等于素材数量"
+    )
+
+    @validator("asset_ids", pre=True)
+    def _drop_invalid_asset_ids(cls, v):  # noqa: N805
+        """容错过滤：前端异常情况下可能把 undefined 序列化成 null 或空串混入
+        asset_ids（会直接 422 或导致后续 /assets/{id} 404），这里统一剔除。
+        过滤后为空时由 Field(min_length=1) / 路由层 400 兜底。"""
+        if not isinstance(v, list):
+            return v
+        return [x for x in v if isinstance(x, str) and x.strip()]


 class ClipsFromAssetsResponse(BaseModel):
@@ -52,6 +52,7 @@ def _to_video_response(item, storage: OSSStorageService | None = None) -> VideoI
        generation_params=item.generation_params,
        download_url=download_url,
        generated_at=format_utc_datetime(item.generated_at) if hasattr(item, "generated_at") else "",
+        duplicate_rate=getattr(item, "duplicate_rate", None),
    )


@@ -129,10 +129,13 @@ class SmartMatchRequest(BaseModel):
    )


-class SmartMatchItem(BaseModel):
-    """智能选素材结果条目。"""
+class SmartMatchItem(AssetResponse):
+    """智能选素材结果条目（扁平结构）。
+
+    素材字段（id/usable/余量等）直接挂在条目顶层，前端拿到 item 即可读 item.id，
+    与 AssetResponse 字段完全一致；score/breakdown 为智能匹配附加的评分字段。
+    """

-    asset: AssetResponse
    score: float = Field(..., ge=0, le=100, description="综合得分 0-100")
    breakdown: dict[str, float] = Field(default_factory=dict, description="各维度得分明细")

@@ -22,6 +22,7 @@ class VideoItemResponse(BaseModel):
    generation_params: dict = Field(default_factory=dict)
    download_url: str | None = None
    generated_at: str = ""
+    duplicate_rate: float | None = None


 class ListVideosResponse(BaseModel):
@@ -438,3 +438,57 @@ def make_reuse_callback(
        return result

    return _reuse
+
+
+def get_asset_recent_use_counts(
+    db: Session,
+    asset_ids: list[str],
+    recent_video_count: int = 5,
+) -> dict[str, int]:
+    """统计每个素材在最近 N 个不同 plan_id 中的使用次数。
+
+    遍历素材 metadata 中的 used_time_ranges，统计有多少个不同的 plan_id（去重），
+    返回 {asset_id: count}。只统计最近 recent_video_count 个不同 plan_id 的使用次数。
+
+    Args:
+        db: 数据库会话
+        asset_ids: 素材 ID 列表
+        recent_video_count: 统计最近多少个不同 plan_id
+
+    Returns:
+        {asset_id: 在最近 recent_video_count 个 plan 中的使用次数}
+    """
+    if not asset_ids:
+        return {}
+
+    result: dict[str, int] = {}
+    models = db.query(AssetModel).filter(AssetModel.id.in_(asset_ids)).all()
+    for model in models:
+        meta = _read_meta(model)
+        ranges = meta.get(USED_RANGES_KEY) or []
+        if not ranges:
+            result[model.id] = 0
+            continue
+
+        # 按 created_at 倒序收集不同 plan_id
+        sorted_ranges = sorted(
+            ranges,
+            key=lambda r: r.get("created_at") or "",
+            reverse=True,
+        )
+        recent_plan_ids: set[str] = set()
+        for r in sorted_ranges:
+            plan_id = r.get("plan_id")
+            if plan_id:
+                recent_plan_ids.add(plan_id)
+            if len(recent_plan_ids) >= recent_video_count:
+                break
+
+        result[model.id] = len(recent_plan_ids)
+
+    # 未找到的素材计为 0
+    for aid in asset_ids:
+        if aid not in result:
+            result[aid] = 0
+
+    return result
@@ -69,10 +69,13 @@ interface SmartMatchWrappedItem {
  breakdown?: unknown
 }

-export const smartMatchAssets = async (libraryId: string): Promise<SmartMatchResult> => {
-  const response = await apiClient.post("/assets/smart-match", {
-    library_id: libraryId,
-  })
+export const smartMatchAssets = async (
+  libraryId: string,
+  limit?: number,
+): Promise<SmartMatchResult> => {
+  const payload: Record<string, unknown> = { library_id: libraryId }
+  if (limit && limit > 0) payload.limit = limit
+  const response = await apiClient.post("/assets/smart-match", payload)
  const rawItems: SmartMatchWrappedItem[] = response.data?.items ?? []
  const items = rawItems
    .map((it) =>
@@ -2,12 +2,17 @@
 * 素材诊断 API
 */
 import apiClient from "../client"
+import { getOrCreateDefaultProject } from "../projects"
 import type { AssetDiagnosis } from "./types"

 /** 获取素材诊断信息（可选 asset_id 查单素材，否则全局诊断） */
-export const getAssetDiagnosis = async (assetId?: string): Promise<AssetDiagnosis> => {
+export const getAssetDiagnosis = async (
+  assetId?: string,
+  projectId?: string,
+): Promise<AssetDiagnosis> => {
+  const pid = projectId ?? (await getOrCreateDefaultProject()).id
  const params: Record<string, string> = {}
  if (assetId) params.asset_id = assetId
-  const response = await apiClient.get("/asset-diagnosis", { params })
+  const response = await apiClient.get(`/projects/${pid}/asset-diagnosis`, { params })
  return response.data
 }
@@ -98,8 +98,8 @@ export function useTemplateSave(options: UseTemplateSaveOptions) {
        estimated_duration: totalDuration,
        segments: clips.map((c, i) => ({
          segment_order: i,
-          duration_min: Math.max(1, c.duration - 2),
-          duration_max: c.duration + 2,
+          duration_min: c.duration,
+          duration_max: c.duration,
          material_type: c.type === "voice" ? "voiceover" : "video",
          transition: c.transition
            ? { type: c.transition.type, duration: c.transition.duration }
@@ -1,12 +1,33 @@
 import { useState, useCallback } from "react"
 import { message } from "antd"
 import type { AssetItem } from "@/api/assets"
+import type { TemplateSegment } from "@/api/templates/types"
 import { smartMatchAssets, isAssetUsable } from "@/api/assets"

 interface UseSmartMatchOptions {
  libraryId: string
  materials: { items: AssetItem[]; total: number }
  onSmartSelectedIdsChange: (ids: string[]) => void
+  /** 当前模板的 segments，用于根据总时长计算 limit */
+  templateSegments?: TemplateSegment[]
+}
+
+/** 默认 limit（拿不到目标时长时的兜底上限） */
+const DEFAULT_LIMIT = 10
+/** 每个素材切片按 15 秒估算所需素材数 */
+const SECONDS_PER_ASSET = 15
+
+/**
+ * 根据模板 segments 计算所需素材数量上限。
+ * 取每个 segment 的 duration_min 之和作为目标视频总时长，
+ * 再按 15 秒/素材估算需要多少个素材；结果钳制到 [1, 200] 区间（后端 limit 上限 200）。
+ */
+function computeLimitFromSegments(segments?: TemplateSegment[]): number {
+  if (!segments || segments.length === 0) return DEFAULT_LIMIT
+  const totalSeconds = segments.reduce((sum, seg) => sum + (seg.duration_min || 0), 0)
+  if (totalSeconds <= 0) return DEFAULT_LIMIT
+  const limit = Math.ceil(totalSeconds / SECONDS_PER_ASSET)
+  return Math.max(1, Math.min(limit, 200))
 }

 /**
@@ -18,6 +39,7 @@ export function useSmartMatch({
  libraryId,
  materials,
  onSmartSelectedIdsChange,
+  templateSegments,
 }: UseSmartMatchOptions) {
  const [smartMatching, setSmartMatching] = useState(false)
  const [hasMatched, setHasMatched] = useState(false)
@@ -42,12 +64,14 @@ export function useSmartMatch({
    setSmartMatching(true)

    try {
+      // 根据目标视频时长计算合理的素材数量上限，避免"有几个选几个"
+      const limit = computeLimitFromSegments(templateSegments)
+
      // 调用后端智能匹配 API（后端也会排除已用尽素材，这里前端兜底过滤）
-      // items 已在 API 层归一化（兼容后端 {asset, score} 包装结构）；
-      // 这里再过滤一遍无 id/已用尽项，杜绝 undefined id 流入预览链路
-      const result = await smartMatchAssets(libraryId)
+      const result = await smartMatchAssets(libraryId, limit)
+      // 兜底过滤：id 为空或不可用的素材不参与匹配（smartMatchAssets 已做归一化，这里双保险）
      const matched = (result.items ?? []).filter((a) => !!a?.id && isAssetUsable(a))
-      const matchedIds = matched.map((a) => a.id)
+      const matchedIds = matched.map((a: AssetItem) => a.id)

      if (matchedIds.length > 0) {
        onSmartSelectedIdsChange(matchedIds)
@@ -67,7 +91,7 @@ export function useSmartMatch({
    } finally {
      setSmartMatching(false)
    }
-  }, [libraryId, materials.items, onSmartSelectedIdsChange])
+  }, [libraryId, materials.items, onSmartSelectedIdsChange, templateSegments])

  /* ── 换一批 = 重新触发智能匹配 ── */
  const handleRefreshMatch = useCallback(async () => {
@@ -51,6 +51,7 @@ export function useStep2Materials({
    libraryId: selectedLibraryId,
    materials: selectableMaterials,
    onSmartSelectedIdsChange,
+    templateSegments,
  })

  /* ── 自动触发智能匹配：选择视频库后自动调用 ── */
@@ -1,6 +1,6 @@
 /**
 * 成片库页面 — V21 设计系统
- * 卡片网格布局，支持视频播放/下载/分享、批量操作、筛选
+ * 卡片网格布局，支持视频内联播放/下载/分享、批量操作、筛选
 *
 * 主组件仅保留 Hook 组装与整体布局
 * 列表查询 → hooks/useProductList
@@ -8,15 +8,12 @@
 * 筛选栏 → components/ProductFilterBar
 * 批量操作栏 → components/ProductBatchBar
 * 空状态 → components/ProductEmptyState
- * 产品卡片 → components/ProductCard
- * 视频播放 → components/VideoPlayer
+ * 产品卡片 → components/ProductCard（内联视频播放）
 */
-import React, { useState } from "react"
+import React from "react"
 import { VideoCameraOutlined, DownloadOutlined } from "@ant-design/icons"
 import { Button } from "@/components/ui"
-import type { ProductItem } from "./types"
 import { ProductCard } from "./components/ProductCard"
-import { VideoPlayer } from "./components/VideoPlayer"
 import { ProductFilterBar } from "./components/ProductFilterBar"
 import { ProductBatchBar } from "./components/ProductBatchBar"
 import { ProductEmptyState } from "./components/ProductEmptyState"
@@ -53,13 +50,9 @@ const ProductLibrary: React.FC = () => {
    clearSelection,
  } = useProductList()

-  /* 播放器 */
-  const [playingProduct, setPlayingProduct] = useState<ProductItem | null>(null)
-
  const {
    handleDownload,
    handleShare,
-    handleViewDetail,
    handleDelete,
    handlePublish,
    handleReviewStatusChange,
@@ -71,7 +64,7 @@ const ProductLibrary: React.FC = () => {
    selectedIds,
    clearSelection,
    products,
-    setPlayingProduct,
+    setPlayingProduct: () => {}, // 不再使用弹窗播放
  })

  // ── Loading 状态 ──
@@ -146,7 +139,6 @@ const ProductLibrary: React.FC = () => {
              isSelected={selectedIds.has(product.id)}
              batchMode={batchMode}
              onToggleSelect={handleToggleSelect}
-              onPlay={setPlayingProduct}
              onDownload={handleDownload}
              onShare={handleShare}
              onDelete={handleDelete}
@@ -158,17 +150,6 @@ const ProductLibrary: React.FC = () => {
      ) : (
        <ProductEmptyState type="empty" />
      )}
-
-      {/* 视频播放弹窗 */}
-      {playingProduct && (
-        <VideoPlayer
-          product={playingProduct}
-          onClose={() => setPlayingProduct(null)}
-          onDownload={handleDownload}
-          onShare={handleShare}
-          onViewDetail={handleViewDetail}
-        />
-      )}
    </div>
  )
 }
@@ -1,4 +1,4 @@
-import React from "react"
+import React, { useState, useRef, useCallback, useEffect } from "react"
 import { Popconfirm } from "antd"
 import {
  CheckOutlined,
@@ -7,6 +7,7 @@ import {
  ShareAltOutlined,
  DeleteOutlined,
  CloudUploadOutlined,
+  VideoCameraOutlined,
 } from "@ant-design/icons"
 import type { ProductItem } from "../types"
 import { statusConfig, reviewStatusConfig } from "../constants"
@@ -17,7 +18,6 @@ interface ProductCardProps {
  isSelected: boolean
  batchMode: boolean
  onToggleSelect: (id: string) => void
-  onPlay: (product: ProductItem) => void
  onDownload: (product: ProductItem) => void
  onShare: (product: ProductItem) => void
  onDelete: (id: string) => void
@@ -30,7 +30,6 @@ export const ProductCard: React.FC<ProductCardProps> = ({
  isSelected,
  batchMode,
  onToggleSelect,
-  onPlay,
  onDownload,
  onShare,
  onDelete,
@@ -38,17 +37,39 @@ export const ProductCard: React.FC<ProductCardProps> = ({
  onReviewStatusChange,
 }) => {
  const st = statusConfig[product.status]
+  const videoRef = useRef<HTMLVideoElement>(null)
+  const [isPlaying, setIsPlaying] = useState(false)
+  const [aspectRatio, setAspectRatio] = useState<string | null>(null)
+  const [hasVideo, setHasVideo] = useState(!!product.videoUrl)

  /** 查重率样式 */
  const dupClass =
    product.duplicateRate <= 5 ? "good" : product.duplicateRate <= 15 ? "warn" : "bad"

-  /** 点击卡片 */
-  const handleCardClick = () => {
-    if (batchMode) {
-      onToggleSelect(product.id)
+  /** 视频元数据加载后获取实际比例 */
+  const handleVideoLoaded = useCallback(() => {
+    const v = videoRef.current
+    if (v && v.videoWidth > 0 && v.videoHeight > 0) {
+      setAspectRatio(`${v.videoWidth} / ${v.videoHeight}`)
+    }
+  }, [])
+
+  /** 视频播放/暂停结束事件 */
+  const handlePlay = useCallback(() => setIsPlaying(true), [])
+  const handlePause = useCallback(() => setIsPlaying(false), [])
+
+  /** 视频出错时降级为缩略图 */
+  const handleVideoError = useCallback(() => setHasVideo(false), [])
+
+  /** 点击缩略图区域：有视频则内联播放，否则不响应 */
+  const handleThumbClick = () => {
+    if (batchMode) return
+    if (!hasVideo || !videoRef.current) return
+    const v = videoRef.current
+    if (v.paused) {
+      v.play()
    } else {
-      onPlay(product)
+      v.pause()
    }
  }

@@ -58,6 +79,24 @@ export const ProductCard: React.FC<ProductCardProps> = ({
    onToggleSelect(product.id)
  }

+  /** 卡片容器点击（非批量模式下不再触发弹窗播放） */
+  const handleCardClick = () => {
+    if (batchMode) {
+      onToggleSelect(product.id)
+    }
+  }
+
+  /** 组件卸载时暂停视频 */
+  useEffect(() => {
+    const v = videoRef.current
+    return () => {
+      v?.pause()
+    }
+  }, [])
+
+  /** 缩略图区域 style：有实际比例则用实际比例，否则默认 16:9 */
+  const thumbStyle: React.CSSProperties = aspectRatio ? { aspectRatio } : { aspectRatio: "16 / 9" }
+
  return (
    <div
      className={`xx-product-card${isSelected ? " selected" : ""}${product.isPublished ? " published" : ""}`}
@@ -102,21 +141,61 @@ export const ProductCard: React.FC<ProductCardProps> = ({
        </div>
      )}

-      {/* 缩略图 */}
-      <div className="xx-product-thumb">
-        {product.thumbnailUrl ? (
-          <img
-            className="xx-product-thumb-bg"
-            src={product.thumbnailUrl}
-            alt={product.name}
-            style={{ width: "100%", height: "100%", objectFit: "cover" }}
-          />
+      {/* 视频/缩略图区域 */}
+      <div className="xx-product-thumb" style={thumbStyle}>
+        {hasVideo ? (
+          <>
+            <video
+              ref={videoRef}
+              className="xx-product-thumb-video"
+              src={product.videoUrl}
+              preload="metadata"
+              onLoadedMetadata={handleVideoLoaded}
+              onPlay={handlePlay}
+              onPause={handlePause}
+              onEnded={handlePause}
+              onError={handleVideoError}
+              controls={isPlaying}
+              onClick={(e) => {
+                e.stopPropagation()
+                handleThumbClick()
+              }}
+            />
+            {/* 未播放时显示播放按钮覆盖层 */}
+            {!isPlaying && (
+              <div className="xx-product-play" onClick={handleThumbClick}>
+                <PlayCircleOutlined />
+              </div>
+            )}
+          </>
        ) : (
-          <div className="xx-product-thumb-bg" style={{ background: product.gradient }} />
+          <>
+            {product.thumbnailUrl ? (
+              <img
+                className="xx-product-thumb-bg"
+                src={product.thumbnailUrl}
+                alt={product.name}
+                style={{ width: "100%", height: "100%", objectFit: "cover" }}
+              />
+            ) : (
+              <div
+                className="xx-product-thumb-bg"
+                style={{
+                  background: product.gradient,
+                  display: "grid",
+                  placeItems: "center",
+                  color: "rgba(255,255,255,0.3)",
+                  fontSize: "48px",
+                }}
+              >
+                <VideoCameraOutlined />
+              </div>
+            )}
+            <div className="xx-product-play">
+              <PlayCircleOutlined />
+            </div>
+          </>
        )}
-        <div className="xx-product-play">
-          <PlayCircleOutlined />
-        </div>
        {product.duration > 0 && (
          <span className="xx-product-duration">{formatTime(product.duration)}</span>
        )}
@@ -260,13 +260,13 @@

 /* 缩略图区域 */
 .xx-product-thumb {
-  aspect-ratio: 9 / 16;
-  max-height: 220px;
  position: relative;
  overflow: hidden;
  display: grid;
  place-items: center;
  color: var(--text-inverse);
+  background: var(--color-gray-950);
+  max-height: 320px;
 }

 .xx-product-thumb-bg {
@@ -276,6 +276,18 @@
  background-position: center;
 }

+/* 内联视频播放器 */
+.xx-product-thumb-video {
+  width: 100%;
+  height: 100%;
+  object-fit: contain;
+  display: block;
+}
+
+.xx-product-thumb-video[controls] {
+  cursor: pointer;
+}
+
 .xx-product-play {
  position: relative;
  z-index: 1;
@@ -308,6 +308,71 @@ class VideoDeduplicator:

        return sum(similarities) / len(similarities) if similarities else 0.0

+    def compute_duplicate_rate(
+        self,
+        fingerprint: VideoFingerprint,
+        project_id: str,
+        current_video_id: str | None,
+        session: Session,
+    ) -> float:
+        """计算当前视频与项目内已有视频的最高相似度百分比。
+
+        遍历项目内所有其他有指纹的视频，对每个计算相似度：
+        - MD5 精确匹配 → 100%
+        - pHash 相似度 → (1.0 - avg_distance / 64) * 100
+        取最高值作为 duplicate_rate（0~100）。
+        如果没有其他视频可比较，返回 0.0。
+
+        Args:
+            fingerprint: 当前视频的指纹
+            project_id: 项目 ID
+            current_video_id: 当前视频 ID（排除自身，可为 None）
+            session: 数据库会话
+
+        Returns:
+            duplicate_rate: 0~100 的浮点数
+        """
+        # 限制查询最近 100 个视频，避免大项目内存溢出
+        from packages.adapters.sqlalchemy_impl.models import GeneratedVideoModel
+
+        recent_models = (
+            session.query(GeneratedVideoModel)
+            .filter(GeneratedVideoModel.project_id == project_id)
+            .order_by(GeneratedVideoModel.generated_at.desc())
+            .limit(100)
+            .all()
+        )
+        video_repo = SQLAlchemyGeneratedVideoRepository(session)
+        existing_videos = [video_repo._to_domain(m) for m in recent_models]
+
+        max_similarity = 0.0
+        for existing in existing_videos:
+            if current_video_id and existing.id == current_video_id:
+                continue
+            if not existing.video_fingerprint:
+                continue
+
+            ef = existing.video_fingerprint
+
+            # MD5 精确匹配 → 100%
+            if fingerprint.md5 == ef.get("md5"):
+                return 100.0
+
+            # pHash 相似度
+            existing_phashes = ef.get("keyframe_phashes", [])
+            if not existing_phashes or not fingerprint.keyframe_phashes:
+                continue
+
+            min_distances = []
+            for phash in fingerprint.keyframe_phashes:
+                distances = [hamming_distance(phash, ep) for ep in existing_phashes]
+                min_distances.append(min(distances))
+            avg_distance = sum(min_distances) / len(min_distances) if min_distances else 64
+            similarity = (1.0 - avg_distance / 64) * 100
+            max_similarity = max(max_similarity, similarity)
+
+        return round(max(max_similarity, 0.0), 2)
+

@celery_app.task(bind=True, max_retries=3, name="worker.check_duplicate")
 def check_duplicate_task(self: Task, generated_video_id: str) -> dict:
@@ -121,6 +121,15 @@ def create_video_record_and_dedup(
            generated_video.is_duplicate = False
            generated_video.duplicate_of = None

+        # 计算重复率百分比（与项目内所有已有视频对比取最高相似度）
+        try:
+            dup_rate = deduplicator.compute_duplicate_rate(fingerprint, project_id, video_id, session)
+            generated_video.duplicate_rate = dup_rate
+            logger.info("Duplicate rate for %s: %.2f%%", video_id, dup_rate)
+        except Exception as rate_err:
+            logger.warning("Failed to compute duplicate_rate for %s: %s", video_id, rate_err)
+            generated_video.duplicate_rate = None
+
        video_repo.update(generated_video)
        session.commit()
        logger.info(
@@ -1,5 +1,6 @@
 # Build stage
 FROM git.xiaoxiajianji.com/xiaoxia/base/node:20 AS builder
+ARG SOURCE_HASH=""
 WORKDIR /app
 ARG VITE_API_URL=https://saas-api.xiaoxiajianji.com
 ENV VITE_API_URL=$VITE_API_URL
@@ -18,8 +19,10 @@ COPY apps/web/ ./

 # 构建：TS增量编译 + Vite构建，tsbuildinfo用cache mount持久化
 # node_modules直接使用镜像中已安装的（layer缓存保证完整性）
+# SOURCE_HASH 变化时强制重新执行（防止 buildkit 幽灵缓存命中）
 RUN --mount=type=cache,target=/app/apps/web/.tscache,sharing=locked \
    mkdir -p .tscache \
+    && echo "SOURCE_HASH=${SOURCE_HASH}" > .cache_bust \
    && ./node_modules/.bin/tsc --incremental --tsBuildInfoFile .tscache/tsconfig.tsbuildinfo \
    && ./node_modules/.bin/vite build

@@ -30,6 +30,7 @@ class SQLAlchemyGeneratedVideoRepository:
            video_fingerprint=json.dumps(video.video_fingerprint) if video.video_fingerprint else None,
            is_duplicate=video.is_duplicate,
            duplicate_of=video.duplicate_of,
+            duplicate_rate=video.duplicate_rate,
            generated_at=video.generated_at,
            created_at=video.created_at,
        )
@@ -60,6 +61,7 @@ class SQLAlchemyGeneratedVideoRepository:
            video_fingerprint=json.loads(getattr(model, "video_fingerprint", "null") or "null"),
            is_duplicate=getattr(model, "is_duplicate", False),
            duplicate_of=getattr(model, "duplicate_of", None),
+            duplicate_rate=getattr(model, "duplicate_rate", None),
            generated_at=model.generated_at,
            created_at=model.created_at,
        )
@@ -74,6 +76,7 @@ class SQLAlchemyGeneratedVideoRepository:
        model.video_fingerprint = json.dumps(video.video_fingerprint) if video.video_fingerprint else None
        model.is_duplicate = video.is_duplicate
        model.duplicate_of = video.duplicate_of
+        model.duplicate_rate = video.duplicate_rate
        self.session.add(model)
        self.session.commit()
        return video
@@ -204,6 +207,7 @@ class SQLAlchemyGeneratedVideoRepository:
            video_fingerprint=json.loads(getattr(model, "video_fingerprint", "null") or "null"),
            is_duplicate=getattr(model, "is_duplicate", False),
            duplicate_of=getattr(model, "duplicate_of", None),
+            duplicate_rate=getattr(model, "duplicate_rate", None),
            generated_at=model.generated_at,
            created_at=model.created_at,
        )
@@ -338,6 +338,7 @@ class GeneratedVideoModel(Base):
    video_fingerprint = Column(Text, nullable=True)
    is_duplicate = Column(Boolean, nullable=False, default=False)
    duplicate_of = Column(String(36), nullable=True)
+    duplicate_rate = Column(Float, nullable=True)


 class TitleLibraryModel(Base):
@@ -1,4 +1,9 @@
-"""SQLAlchemy implementation of TemplateRepository."""
+"""SQLAlchemy implementation of TemplateRepository.
+
+模板 segments 数据源已统一为 template_clip_configs 表。
+读取时优先 template_clip_configs，回退 template_segments（兼容历史数据）。
+写入全部走 template_clip_configs。
+"""

 from __future__ import annotations

@@ -10,6 +15,7 @@ from sqlalchemy.orm import Session
 from packages.adapters.sqlalchemy_impl.models import (
    EditPlanModel,
    TemplateCategoryModel,
+    TemplateClipConfigModel,
    TemplateModel,
    TemplateSegmentModel,
 )
@@ -47,27 +53,38 @@ class SQLAlchemyTemplateRepository:
            like_pattern = f"%{keyword}%"
            query = query.filter(TemplateModel.name.like(like_pattern))
        if tag:
-            # JSON 数组包含指定标签（MySQL JSON_CONTAINS / SQLite json_each 兼容写法用 LIKE）
-            query = query.filter(TemplateModel.tags.like(f'%"{tag}"%'))
+            query = query.filter(TemplateModel.tags.like(f'"%{tag}"%'))
        models = query.order_by(TemplateModel.created_at.desc()).offset(skip).limit(limit).all()
        templates = [self._model_to_entity(m) for m in models]
-        # 批量加载所有 segments，避免 N+1 查询
+        # 批量加载 segments —— 优先 template_clip_configs
        if templates:
            template_ids = [t.id for t in templates]
-            seg_models = (
-                self.session.query(TemplateSegmentModel)
-                .filter(TemplateSegmentModel.template_id.in_(template_ids))
-                .order_by(TemplateSegmentModel.segment_order)
+            clip_models = (
+                self.session.query(TemplateClipConfigModel)
+                .filter(TemplateClipConfigModel.template_id.in_(template_ids))
+                .order_by(TemplateClipConfigModel.order)
                .all()
            )
-            # 按 template_id 分组
-            seg_map: dict[str, list] = {}
-            for sm in seg_models:
-                seg_map.setdefault(sm.template_id, []).append(
-                    self._segment_model_to_entity(sm),
+            clip_map: dict[str, list] = {}
+            for cm in clip_models:
+                clip_map.setdefault(cm.template_id, []).append(
+                    self._clip_config_to_segment(cm),
                )
+            # 对没有 clip_configs 的模板，回退读 template_segments
+            missing_ids = [t.id for t in templates if t.id not in clip_map]
+            if missing_ids:
+                old_models = (
+                    self.session.query(TemplateSegmentModel)
+                    .filter(TemplateSegmentModel.template_id.in_(missing_ids))
+                    .order_by(TemplateSegmentModel.segment_order)
+                    .all()
+                )
+                for om in old_models:
+                    clip_map.setdefault(om.template_id, []).append(
+                        self._segment_model_to_entity(om),
+                    )
            for t in templates:
-                t.segments = seg_map.get(t.id, [])
+                t.segments = clip_map.get(t.id, [])
        return templates

    def get(self, template_id: str, user_id: str) -> Optional[Template]:
@@ -100,7 +117,6 @@ class SQLAlchemyTemplateRepository:
            is_active=template.is_active,
        )
        self.session.add(model)
-        # flush 而非 commit，让 create + create_segments 在同一事务中提交
        self.session.flush()
        self.session.refresh(model)
        result = self._model_to_entity(model)
@@ -145,11 +161,8 @@ class SQLAlchemyTemplateRepository:
        if model is None:
            return False
        model.is_active = False
-        # 级联清理关联的 segments，避免孤儿数据
-        self.session.query(TemplateSegmentModel).filter(
-            TemplateSegmentModel.template_id == template_id,
-        ).delete(synchronize_session=False)
-        self.session.commit()
+        # 复用 delete_segments_by_template 清理两张表的关联数据
+        self.delete_segments_by_template(template_id)
        return True

    def count_by_user(
@@ -172,7 +185,7 @@ class SQLAlchemyTemplateRepository:
        if keyword:
            query = query.filter(TemplateModel.name.like(f"%{keyword}%"))
        if tag:
-            query = query.filter(TemplateModel.tags.like(f'%"{tag}"%'))
+            query = query.filter(TemplateModel.tags.like(f'"%{tag}"%'))
        return query.count()

    def copy_template(self, template_id: str, user_id: str, new_name: str) -> Template:
@@ -181,9 +194,8 @@ class SQLAlchemyTemplateRepository:
        if source is None:
            raise ValueError(f"Template {template_id} not found")

-        new_id = str(uuid.uuid4())
        new_template = Template(
-            id=new_id,
+            id=str(uuid.uuid4()),
            user_id=user_id,
            name=new_name,
            mode=source.mode,
@@ -197,28 +209,22 @@ class SQLAlchemyTemplateRepository:
        )
        created = self.create(new_template)

-        # 复制 segments
+        # 复用 create_segments 写入 template_clip_configs
        new_segments: List[TemplateSegment] = []
        for seg in source.segments:
-            new_seg = TemplateSegment(
-                id=str(uuid.uuid4()),
-                template_id=new_id,
-                segment_order=seg.segment_order,
-                duration_min=seg.duration_min,
-                duration_max=seg.duration_max,
-                material_type=seg.material_type,
+            new_segments.append(
+                TemplateSegment(
+                    id=str(uuid.uuid4()),
+                    template_id=created.id,
+                    segment_order=seg.segment_order,
+                    duration_min=seg.duration_min,
+                    duration_max=seg.duration_max,
+                    material_type=seg.material_type,
+                )
            )
-            new_segments.append(new_seg)
-            model = TemplateSegmentModel(
-                id=new_seg.id,
-                template_id=new_seg.template_id,
-                segment_order=new_seg.segment_order,
-                duration_min=new_seg.duration_min,
-                duration_max=new_seg.duration_max,
-                material_type=new_seg.material_type,
-            )
-            self.session.add(model)
        if new_segments:
+            self.create_segments(new_segments)
+        else:
            self.session.commit()

        created.segments = new_segments
@@ -227,34 +233,58 @@ class SQLAlchemyTemplateRepository:
    # ── Segments ──

    def list_segments(self, template_id: str) -> List[TemplateSegment]:
-        models = (
+        """优先从 template_clip_configs 读取，回退读 template_segments。"""
+        clips = (
+            self.session.query(TemplateClipConfigModel)
+            .filter(TemplateClipConfigModel.template_id == template_id)
+            .order_by(TemplateClipConfigModel.order)
+            .all()
+        )
+        if clips:
+            return [self._clip_config_to_segment(m) for m in clips]
+        # 回退：旧表
+        old = (
            self.session.query(TemplateSegmentModel)
            .filter(TemplateSegmentModel.template_id == template_id)
            .order_by(TemplateSegmentModel.segment_order)
            .all()
        )
-        return [self._segment_model_to_entity(m) for m in models]
+        return [self._segment_model_to_entity(m) for m in old]

    def create_segments(self, segments: List[TemplateSegment]) -> List[TemplateSegment]:
+        """写入 template_clip_configs 表。material_type 存入 config JSON。"""
        for seg in segments:
-            model = TemplateSegmentModel(
+            config = {"material_type": seg.material_type} if seg.material_type else {}
+            model = TemplateClipConfigModel(
                id=seg.id,
                template_id=seg.template_id,
-                segment_order=seg.segment_order,
-                duration_min=seg.duration_min,
-                duration_max=seg.duration_max,
-                material_type=seg.material_type,
+                clip_type="main",
+                order=seg.segment_order,
+                min_duration=seg.duration_min,
+                max_duration=seg.duration_max,
+                text_template="",
+                material_requirements={},
+                transition_effect="cut",
+                config=config,
            )
            self.session.add(model)
        self.session.commit()
        return segments

    def delete_segments_by_template(self, template_id: str) -> int:
-        count = (
-            self.session.query(TemplateSegmentModel).filter(TemplateSegmentModel.template_id == template_id).delete()
+        """删除两张表中的 segments 数据，返回删除总数。"""
+        c1 = (
+            self.session.query(TemplateClipConfigModel)
+            .filter(TemplateClipConfigModel.template_id == template_id)
+            .delete(synchronize_session=False)
+        )
+        c2 = (
+            self.session.query(TemplateSegmentModel)
+            .filter(TemplateSegmentModel.template_id == template_id)
+            .delete(synchronize_session=False)
        )
        self.session.commit()
-        return count
+        return c1 + c2

    # ── Categories ──

@@ -366,6 +396,23 @@ class SQLAlchemyTemplateRepository:
            updated_at=model.updated_at,
        )

+    @staticmethod
+    def _clip_config_to_segment(model: TemplateClipConfigModel) -> TemplateSegment:
+        """将 TemplateClipConfigModel 转换为 TemplateSegment 域实体。"""
+        material_type = None
+        if model.config and isinstance(model.config, dict):
+            material_type = model.config.get("material_type")
+        return TemplateSegment(
+            id=model.id,
+            template_id=model.template_id,
+            segment_order=model.order,
+            duration_min=model.min_duration,
+            duration_max=model.max_duration,
+            material_type=material_type,
+            created_at=model.created_at,
+            updated_at=model.updated_at,
+        )
+
    @staticmethod
    def _category_model_to_entity(model: TemplateCategoryModel) -> TemplateCategory:
        return TemplateCategory(
@@ -26,6 +26,7 @@ class GeneratedVideo:
    video_fingerprint: dict[str, Any] | None = None
    is_duplicate: bool = False
    duplicate_of: str | None = None
+    duplicate_rate: float | None = None
    generated_at: datetime = field(default_factory=lambda: datetime.now(timezone.utc))
    created_at: datetime = field(default_factory=lambda: datetime.now(timezone.utc))

@@ -79,7 +79,7 @@ def get_token(repo, action="pull"):
    token_url = AUTH_URL + "?service=" + SERVICE + "&scope=" + scope
    req = urllib.request.Request(token_url)
    req.add_header("Authorization", "Basic " + base64.b64encode((USERNAME + ":" + PASSWORD).encode()).decode())
-    with urllib.request.urlopen(req) as resp:
+    with urllib.request.urlopen(req, timeout=30) as resp:
        data = json.loads(resp.read())
        return data.get("token", "")

@@ -89,7 +89,7 @@ def get_tags(repo, token):
    url = "https://" + REGISTRY + "/v2/" + NAMESPACE + "/" + repo + "/tags/list?n=1000"
    req = urllib.request.Request(url)
    req.add_header("Authorization", "Bearer " + token)
-    with urllib.request.urlopen(req) as resp:
+    with urllib.request.urlopen(req, timeout=30) as resp:
        data = json.loads(resp.read())
        return data.get("tags", []) or []

@@ -99,7 +99,7 @@ def http_get_json(url, token, accept_header):
    req = urllib.request.Request(url)
    req.add_header("Authorization", "Bearer " + token)
    req.add_header("Accept", accept_header)
-    with urllib.request.urlopen(req) as resp:
+    with urllib.request.urlopen(req, timeout=30) as resp:
        return json.loads(resp.read()), resp.headers


@@ -194,7 +194,7 @@ def delete_manifest(repo, digest, token):
    req.add_header("Accept", ACCEPT_MANIFEST_OCI)
    req.add_header("Accept", ACCEPT_MANIFEST_V2)
    try:
-        with urllib.request.urlopen(req) as resp:
+        with urllib.request.urlopen(req, timeout=30) as resp:
            return True, resp.status
    except urllib.error.HTTPError as e:
        return False, str(e.code) + " " + e.read().decode()[:200]
@@ -216,7 +216,7 @@ def gitea_get_open_prs():
        req = urllib.request.Request(url)
        req.add_header("Authorization", "token " + GITEA_TOKEN)
        try:
-            with urllib.request.urlopen(req) as resp:
+            with urllib.request.urlopen(req, timeout=30) as resp:
                data = json.loads(resp.read())
                if not data:
                    break
@@ -241,7 +241,7 @@ def gitea_get_pr_commits(pr_number):
    req = urllib.request.Request(url)
    req.add_header("Authorization", "token " + GITEA_TOKEN)
    try:
-        with urllib.request.urlopen(req) as resp:
+        with urllib.request.urlopen(req, timeout=30) as resp:
            data = json.loads(resp.read())
            return [c.get("sha", "") for c in data]
    except Exception as e:
@@ -408,7 +408,7 @@ def cleanup_repo(repo, keep_count, dry_run, protected_tags, pr_sha=None, pr_open
                req = urllib.request.Request(url)
                req.add_header("Authorization", "token " + GITEA_TOKEN)
                try:
-                    with urllib.request.urlopen(req) as resp:
+                    with urllib.request.urlopen(req, timeout=30) as resp:
                        data = json.loads(resp.read())
                        if not data:
                            break
@@ -46,7 +46,7 @@ def ensure_git_repo(api_url, repo, token, pr_number):
    # 获取PR的源分支
    pr_api_url = f"{api_url}/repos/{repo}/pulls/{pr_number}"
    req_obj = urllib.request.Request(pr_api_url, headers={"Authorization": f"token {token}"})
-    with urllib.request.urlopen(req_obj) as resp:
+    with urllib.request.urlopen(req_obj, timeout=15) as resp:
        pr = json.loads(resp.read())
    head_branch = pr["head"]["ref"]

@@ -120,7 +120,7 @@ def get_changed_files(pr_number, api_url, token):
    """获取PR中变更的文件列表"""
    url = f"{api_url}/pulls/{pr_number}/files?limit=100"
    req = urllib.request.Request(url, headers={"Authorization": f"token {token}"})
-    with urllib.request.urlopen(req) as resp:
+    with urllib.request.urlopen(req, timeout=15) as resp:
        files = json.loads(resp.read())
    return [f["filename"] for f in files if f["status"] != "removed"]

@@ -129,7 +129,7 @@ def get_pr_head_branch(pr_number, api_url, token):
    """获取PR的来源分支名"""
    url = f"{api_url}/pulls/{pr_number}"
    req = urllib.request.Request(url, headers={"Authorization": f"token {token}"})
-    with urllib.request.urlopen(req) as resp:
+    with urllib.request.urlopen(req, timeout=15) as resp:
        pr = json.loads(resp.read())
    return pr["head"]["ref"]

@@ -244,7 +244,7 @@ def main():
        # 获取PR信息
        pr_info_url = f"{api_url}/repos/{repo}/pulls/{pr_number}"
        req_pr = urllib.request.Request(pr_info_url, headers={"Authorization": f"token {token}"})
-        with urllib.request.urlopen(req_pr) as resp:
+        with urllib.request.urlopen(req_pr, timeout=15) as resp:
            pr_info = json.loads(resp.read())
        pr_author = pr_info.get("user", {}).get("login", "")
        print(f"PR作者: {pr_author}")
@@ -256,7 +256,7 @@ def main():
        try:
            commits_url = f"{api_url}/repos/{repo}/pulls/{pr_number}/commits?limit=3"
            req_commits = urllib.request.Request(commits_url, headers={"Authorization": f"token {token}"})
-            with urllib.request.urlopen(req_commits) as resp_commits:
+            with urllib.request.urlopen(req_commits, timeout=15) as resp_commits:
                commits = json.loads(resp_commits.read())
            latest_msg = commits[0].get("commit", {}).get("message", "") if commits else ""
            if skip_marker in latest_msg:
@@ -13,6 +13,7 @@ before="${GITHUB_EVENT_BEFORE:-}"
 after="${GITHUB_SHA:-}"
 repo="${GITHUB_REPOSITORY:-}"
 base="${GITHUB_API_URL:-}"
+ZERO="0000000000000000000000000000000000000000"

 # Gitea Actions 中 push 事件的前一个 SHA 在 event payload 的 before 字段
 if [ -z "$before" ] && [ -n "${GITHUB_EVENT_PATH:-}" ] && [ -f "$GITHUB_EVENT_PATH" ]; then
@@ -28,8 +29,76 @@ fi

 echo "改动范围检测: before=${before:-<empty>} after=${after}"

+# ── 安全回溯：确保 diff 基准覆盖所有未构建的改动 ──
+# 问题：concurrency 取消机制会导致前端改动被跳过。被取消的 push 的改动不会被
+# 后续 push 的 diff 覆盖到，因为 GITHUB_EVENT_BEFORE 只指向上一次 push 的 SHA。
+# 修复：查询最近一次**实际构建了 web 镜像**的成功 push run，用其 head_sha 作为
+# diff 基准。这样被取消/跳过的 run 的改动都会被包含在当前 diff 中。
+if [ -n "$before" ] && [ "$before" != "$ZERO" ] && [ -n "${GITHUB_TOKEN:-}" ]; then
+  BRANCH="${GITHUB_REF_NAME:-}"
+  if [ -n "$BRANCH" ]; then
+    SAFE_BASE=$(python3 -c "
+import json, subprocess, sys
+
+base = '${base}'
+repo = '${repo}'
+token = '${GITHUB_TOKEN}'
+branch = '${BRANCH}'
+cur_sha = '${after}'
+
+def check_run(run_id):
+    \"\"\"Check if this run actually built the web image.\"\"\"
+    try:
+        r = subprocess.run(
+            ['curl', '-sf', '--max-time', '10',
+             '-H', f'Authorization: token {token}',
+             f'{base}/repos/{repo}/actions/runs/{run_id}/jobs'],
+            capture_output=True, text=True, timeout=15)
+        if r.returncode != 0:
+            return False
+        jobs = json.loads(r.stdout).get('jobs', [])
+        return any(
+            'Build Staging Web' in j.get('name', '')
+            and j.get('conclusion') == 'success'
+            for j in jobs
+        )
+    except Exception:
+        return False
+
+try:
+    r = subprocess.run(
+        ['curl', '-sf', '--max-time', '15',
+         '-H', f'Authorization: token {token}',
+         f'{base}/repos/{repo}/actions/runs?status=success&event=push&branch={branch}&per_page=30'],
+        capture_output=True, text=True, timeout=20)
+    if r.returncode != 0:
+        sys.exit(0)
+    d = json.loads(r.stdout)
+    runs = d.get('workflow_runs', []) if isinstance(d, dict) else d
+    for run in runs:
+        sha = run.get('head_sha', '')
+        if sha and sha != cur_sha:
+            if check_run(run['id']):
+                print(sha)
+                break
+except Exception:
+    pass
+" 2>/dev/null || true)
+
+    if [ -n "$SAFE_BASE" ] && [ "$SAFE_BASE" != "$before" ]; then
+      echo "🔒 安全回溯: 使用最近实际构建 web 的 commit ${SAFE_BASE:0:8} 替代 before=${before:0:8}"
+      before="$SAFE_BASE"
+    elif [ -z "$SAFE_BASE" ]; then
+      echo "⚠️ 未找到历史成功构建 web 的 push run，保守走全量构建"
+      echo "skip_backend=false" >> "$OUTPUT"
+      echo "skip_frontend=false" >> "$OUTPUT"
+      exit 0
+    fi
+  fi
+fi
+
 FILES=""
-if [ -n "$before" ] && [ "$before" != "0000000000000000000000000000000000000000" ]; then
+if [ -n "$before" ] && [ "$before" != "$ZERO" ]; then
  # Gitea 1.26.x compare API 的顶层 files 字段不填充（始终为空），
  # 但响应里每个 commit 条目自带的 files 完整可用；聚合区间内所有提交的 files 即可。
  API_URL="${base}/repos/${repo}/compare/${before}...${after}?per_page=300"
@@ -0,0 +1,34 @@
+#!/bin/bash
+# CI job 运行前自检 —— 防 Gitea 1.26 调度 bug：
+# 被 concurrency cancel 的 run，其已派发的 job 可能在数小时后被调度器复活重跑
+# （2026-08-30 实测：18:46 取消的 run，21:46 复活 attempt=2，与正常流水线抢 runner 1.5 小时）。
+# job 启动第一步先实时查本 run 状态，若已终结（cancelled/skipped），容器立即退出不烧资源。
+#
+# 用法：在重量级 job（Validate/Unit Tests/Integration/Frontend/Build/E2E 类）的
+#       首个 step 中加入：bash scripts/ci/ci_run_selfcheck.sh
+set -u
+
+echo "▶ CI 运行前自检（僵尸 run 复活防护）..."
+
+if [ -z "${GITHUB_API_URL:-}" ] || [ -z "${GITHUB_REPOSITORY:-}" ] || [ -z "${GITHUB_TOKEN:-}" ] || [ -z "${GITHUB_RUN_ID:-}" ]; then
+  echo "⚠️  缺少 GITHUB_* 环境变量（GITHUB_RUN_ID/GITHUB_TOKEN），跳过自检（不阻断）"
+  exit 0
+fi
+
+CONCLUSION=$(curl -sf -H "Authorization: token ${GITHUB_TOKEN}" --max-time 10 \
+  "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/actions/runs/${GITHUB_RUN_ID}" 2>/dev/null \
+  | python3 -c "import json,sys; print(json.load(sys.stdin).get('conclusion') or 'running')" 2>/dev/null || echo "unknown")
+
+case "${CONCLUSION}" in
+  cancelled|skipped)
+    echo "::error::本 run(${GITHUB_RUN_ID}) 状态已为 ${CONCLUSION}，却仍被调度器派发执行——判定为 Gitea 复活的僵尸 job，立即退出，不再占用 runner"
+    exit 78  # EX_CONFIG：非零退出让调度器知晓该 job 终结（避免被当成功空跑）
+    ;;
+  running|unknown|"")
+    # unknown 时不阻断（API 临时故障），正常执行
+    echo "✅ run ${GITHUB_RUN_ID} 状态正常（${CONCLUSION:-running}），继续执行"
+    ;;
+  *)
+    echo "✅ run ${GITHUB_RUN_ID} 结论=${CONCLUSION}（已终结但非取消），继续执行"
+    ;;
+esac
@@ -0,0 +1,15 @@
+[Unit]
+Description=CI Transient Fault Auto-Retry
+After=network.target
+
+[Service]
+Type=oneshot
+# 安全设置
+ProtectSystem=strict
+ReadWritePaths=/opt/act-runner-docker/ci-retry-state /var/log
+PrivateTmp=true
+NoNewPrivileges=true
+
+# Token 从环境文件加载（不要用明文写在 unit 里）
+EnvironmentFile=/opt/act-runner-docker/ci-retry-state/env
+ExecStart=/opt/act-runner-docker/ci-retry-state/ci_transient_retry.sh
@@ -0,0 +1,305 @@
+#!/bin/bash
+# ============================================================================
+# CI 瞬态故障自动重试脚本
+# ============================================================================
+#
+# 解决什么问题：
+#   当某个 runner 出现瞬态故障（磁盘满、docker 挂掉、网络抖动等），分配到该
+#   runner 的 job 会在 "Set up job" / "Checkout code" 阶段就失败。这种失败
+#   与代码无关，换到其他 runner 重跑就能通过，但 CI 没有内置重试机制，只能
+#   人工干预。
+#
+# 检测逻辑：
+#   1. 查询最近 30 分钟内完成的 workflow runs
+#   2. 找到 conclusion=failure 的 run
+#   3. 检查失败 run 中的 jobs：
+#      a. 失败的 job 的第一个 step 必须是 "Checkout code" 或 "Set up job"
+#         且 conclusion=failure（说明 runner 环境有问题，不是代码问题）
+#      b. 同一 run 中至少有 2 个成功的 job（排除代码本身全挂的情况）
+#      c. 失败 job 数量 < 总 job 数量的 50%（多数成功=瞬态故障）
+#   4. 满足以上条件 → 判定为瞬态故障 → 自动 re-run 整个 workflow
+#
+# 安全措施：
+#   - 每个 run 最多自动重试 1 次（通过状态文件跟踪，避免死循环）
+#   - 状态文件 24 小时后自动清理
+#   - 所有操作记录日志，便于审计
+#
+# 用法：
+#   ./scripts/ci/ci_transient_retry.sh              # 正常运行
+#   ./scripts/ci/ci_transient_retry.sh --dry-run     # 只检查不重试
+#   ./scripts/ci/ci_transient_retry.sh --verbose     # 详细日志输出
+#
+# 环境变量：
+#   GITEA_API_TOKEN  - Gitea API token（必须设置）
+#   GITEA_API_URL    - Gitea API 基础地址（默认 https://git.xiaoxiajianji.com/api/v1）
+#   GITEA_REPO       - 仓库路径（默认 xiaoxia/xiaoxia-saas）
+#   RETRY_STATE_DIR  - 重试状态目录（默认 /opt/act-runner-docker/ci-retry-state）
+#   LOG_FILE         - 日志文件（默认 /var/log/ci-auto-retry.log）
+#
+# 部署方式：
+#   1. 将本脚本复制到 CI 服务器（如 /opt/act-runner-docker/ci-retry-state/）
+#   2. 创建 env 文件: echo 'GITEA_API_TOKEN=xxx' > /opt/act-runner-docker/ci-retry-state/env
+#   3. 安装 systemd timer:
+#      cp scripts/ci/ci_transient_retry.{service,timer} /etc/systemd/system/
+#      systemctl daemon-reload
+#      systemctl enable --now ci_transient_retry.timer
+#   或用 crontab:
+#      */5 * * * * GITEA_API_TOKEN=xxx bash /opt/act-runner-docker/ci-retry-state/ci_transient_retry.sh
+# ============================================================================
+set -eu
+
+# ---- 配置 ----
+GITEA_API_URL="${GITEA_API_URL:-https://git.xiaoxiajianji.com/api/v1}"
+GITEA_REPO="${GITEA_REPO:-xiaoxia/xiaoxia-saas}"
+RETRY_STATE_DIR="${RETRY_STATE_DIR:-/opt/act-runner-docker/ci-retry-state}"
+LOG_FILE="${LOG_FILE:-/var/log/ci-auto-retry.log}"
+WINDOW_MINUTES=30          # 检查最近 N 分钟内的 run
+MAX_RETRY_PER_RUN=1        # 每个 run 最多重试次数
+STATE_TTL_HOURS=24         # 状态文件过期时间（小时）
+MIN_SUCCESS_JOBS=2         # 至少 N 个 job 成功才算瞬态故障
+MAX_FAIL_RATIO=50          # 失败 job 占比上限（%）
+
+# ---- 参数解析 ----
+DRY_RUN=false
+VERBOSE=false
+for arg in "$@"; do
+  case "$arg" in
+    --dry-run)  DRY_RUN=true ;;
+    --verbose)  VERBOSE=true ;;
+  esac
+done
+
+# ---- 日志 ----
+log() {
+  local level="$1"; shift
+  local ts
+  ts=$(date -u +%Y-%m-%dT%H:%M:%SZ)
+  local msg="[$ts] [$level] $*"
+  echo "$msg" >> "$LOG_FILE" 2>/dev/null || true
+  if [ "$level" = "ERROR" ] || [ "$level" = "WARN" ] || $VERBOSE || $DRY_RUN; then
+    echo "$msg"
+  fi
+}
+log_info()  { log INFO "$@"; }
+log_warn()  { log WARN "$@"; }
+log_error() { log ERROR "$@"; }
+log_debug() { $VERBOSE && log DEBUG "$@" || true; }
+
+# ---- 前置检查 ----
+if [ -z "${GITEA_API_TOKEN:-}" ]; then
+  log_error "GITEA_API_TOKEN 未设置，退出"
+  exit 1
+fi
+
+mkdir -p "$RETRY_STATE_DIR" 2>/dev/null || true
+mkdir -p "$(dirname "$LOG_FILE")" 2>/dev/null || true
+
+API_BASE="${GITEA_API_URL}/repos/${GITEA_REPO}/actions"
+
+log_info "========== CI 瞬态故障检测开始 =========="
+$DRY_RUN && log_info "[DRY-RUN 模式] 不会实际触发重试"
+
+# ---- 清理过期状态文件 ----
+find "$RETRY_STATE_DIR" -name "*.retry" -mmin "+$((STATE_TTL_HOURS * 60))" -delete 2>/dev/null || true
+
+# ---- 临时文件（用 mktemp 避免引号/转义问题） ----
+TMPDIR_WORK=$(mktemp -d)
+trap "rm -rf $TMPDIR_WORK" EXIT
+
+# ---- 查询最近完成的 runs ----
+curl -sf -H "Authorization: token ${GITEA_API_TOKEN}" \
+  "${API_BASE}/runs?status=completed&limit=20" > "${TMPDIR_WORK}/runs.json" 2>/dev/null || echo '[]' > "${TMPDIR_WORK}/runs.json"
+
+# ---- 主分析逻辑（全部用 python3，避免 bash JSON 处理陷阱） ----
+python3 - "$TMPDIR_WORK" "$API_BASE" "$GITEA_API_TOKEN" "$RETRY_STATE_DIR" \
+  "$WINDOW_MINUTES" "$MIN_SUCCESS_JOBS" "$MAX_FAIL_RATIO" "$MAX_RETRY_PER_RUN" \
+  "$DRY_RUN" "$VERBOSE" "$LOG_FILE" << 'PYEOF'
+import json, sys, os, subprocess, urllib.request
+from datetime import datetime, timezone, timedelta
+
+tmpdir = sys.argv[1]
+api_base = sys.argv[2]
+token = sys.argv[3]
+state_dir = sys.argv[4]
+window_min = int(sys.argv[5])
+min_success = int(sys.argv[6])
+max_fail_pct = int(sys.argv[7])
+max_retry = int(sys.argv[8])
+dry_run = sys.argv[9] == "true"
+verbose = sys.argv[10] == "true"
+log_file = sys.argv[11]
+
+def log(level, msg):
+    ts = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
+    line = f"[{ts}] [{level}] {msg}"
+    try:
+        with open(log_file, "a") as f:
+            f.write(line + "\n")
+    except:
+        pass
+    if level in ("ERROR", "WARN") or verbose or dry_run:
+        print(line)
+
+# 加载 runs
+with open(os.path.join(tmpdir, "runs.json")) as f:
+    runs_data = json.load(f)
+if isinstance(runs_data, dict):
+    runs_data = runs_data.get("workflow_runs", runs_data.get("runs", []))
+
+cutoff = datetime.now(timezone.utc) - timedelta(minutes=window_min)
+
+# 筛选最近 N 分钟内完成的失败 runs
+candidates = []
+for r in runs_data:
+    completed_str = r.get("completed_at", "")
+    if not completed_str or completed_str.startswith("1970"):
+        continue
+    try:
+        completed = datetime.fromisoformat(completed_str.replace("Z", "+00:00"))
+        if completed < cutoff:
+            continue
+    except:
+        continue
+    if r.get("conclusion") != "failure":
+        continue
+    candidates.append(r)
+
+if not candidates:
+    log("INFO", f"没有发现最近 {window_min} 分钟内失败的 runs")
+    log("INFO", "========== 检测结束 ==========")
+    sys.exit(0)
+
+log("INFO", f"发现 {len(candidates)} 个失败的 runs 需要分析")
+
+retry_count = 0
+headers = {"Authorization": f"token {token}"}
+
+for run in candidates:
+    run_id = run["id"]
+    run_number = run.get("run_number", run_id)
+    branch = run.get("head_branch", "")
+    event = run.get("event", "")
+
+    log("INFO", f"分析 Run #{run_number} (id={run_id}) branch={branch} event={event}")
+
+    # 检查是否已重试
+    state_file = os.path.join(state_dir, f"{run_id}.retry")
+    prev_attempts = 0
+    if os.path.exists(state_file):
+        try:
+            with open(state_file) as f:
+                prev_attempts = int(f.read().strip())
+        except:
+            pass
+    if prev_attempts >= max_retry:
+        log("INFO", f"  Run #{run_number} 已重试过 {prev_attempts} 次，跳过")
+        continue
+
+    # 获取 jobs
+    try:
+        req = urllib.request.Request(
+            f"{api_base}/runs/{run_id}/jobs",
+            headers=headers
+        )
+        with urllib.request.urlopen(req, timeout=15) as resp:
+            jobs_data = json.loads(resp.read())
+    except Exception as e:
+        log("ERROR", f"  获取 jobs 失败: {e}")
+        continue
+
+    jobs = jobs_data.get("jobs", [])
+    total = len(jobs)
+    success_count = 0
+    fail_count = 0
+    transient_jobs = []
+    code_fail_jobs = []
+
+    for j in jobs:
+        conclusion = j.get("conclusion", "")
+        if conclusion == "success":
+            success_count += 1
+        elif conclusion == "failure":
+            fail_count += 1
+            steps = j.get("steps", [])
+            if steps:
+                first = steps[0]
+                fname = first.get("name", "").lower()
+                fconc = first.get("conclusion", "")
+                # 瞬态故障特征：第一个 step（checkout/setup）失败
+                if fconc == "failure" and any(kw in fname for kw in ["checkout", "set up", "setup"]):
+                    transient_jobs.append({
+                        "name": j["name"],
+                        "runner": j.get("runner_name", "?"),
+                    })
+                else:
+                    code_fail_jobs.append(j["name"])
+            else:
+                code_fail_jobs.append(j["name"])
+
+    skip_count = total - success_count - fail_count
+    log("INFO", f"  Jobs: total={total} success={success_count} fail={fail_count} skip={skip_count}")
+
+    if transient_jobs:
+        log("INFO", f"  瞬态故障 jobs: {', '.join(j['name'] for j in transient_jobs)}")
+    if code_fail_jobs:
+        log("INFO", f"  代码失败 jobs: {', '.join(code_fail_jobs)}")
+
+    # 判定
+    is_transient = False
+    reason = ""
+    if transient_jobs and not code_fail_jobs:
+        if success_count >= min_success:
+            is_transient = True
+            reason = f"所有失败 job 在 checkout/setup 阶段失败，{success_count} 个 job 成功"
+        else:
+            reason = f"checkout 失败但成功 job 数不足 ({success_count} < {min_success})"
+    elif transient_jobs and code_fail_jobs:
+        if fail_count < total * max_fail_pct / 100 and success_count >= min_success:
+            is_transient = True
+            reason = f"{len(transient_jobs)} 瞬态 + {len(code_fail_jobs)} 代码，但成功 job 占多数"
+        else:
+            reason = f"混合失败: {len(transient_jobs)} 瞬态 + {len(code_fail_jobs)} 代码"
+    elif code_fail_jobs:
+        reason = f"纯代码失败: {', '.join(code_fail_jobs[:3])}"
+    else:
+        reason = "无失败 job"
+
+    log("INFO", f"  判定: {reason}")
+
+    if not is_transient:
+        log("INFO", "  → 非瞬态故障，跳过")
+        continue
+
+    # 触发重试
+    log("WARN", f"  → 检测到瞬态故障！准备重试 Run #{run_number}")
+
+    if dry_run:
+        log("INFO", "  [DRY-RUN] 跳过实际重试")
+        continue
+
+    # 调用 re-run API
+    try:
+        req = urllib.request.Request(
+            f"{api_base}/runs/{run_id}/rerun",
+            headers={**headers, "Content-Type": "application/json"},
+            method="POST",
+            data=b""
+        )
+        with urllib.request.urlopen(req, timeout=30) as resp:
+            result = json.loads(resp.read())
+            new_run_id = result.get("id", "?")
+            new_status = result.get("status", "?")
+
+        # 记录重试状态
+        with open(state_file, "w") as f:
+            f.write(str(prev_attempts + 1))
+
+        log("INFO", f"  ✅ Re-run 成功! 新 Run ID: {new_run_id}, 状态: {new_status}")
+        retry_count += 1
+    except Exception as e:
+        log("ERROR", f"  ❌ Re-run 失败: {e}")
+
+log("INFO", f"========== 检测结束: 检查 {len(candidates)} 个失败 runs，重试 {retry_count} 个 ==========")
+PYEOF
+
+exit 0
@@ -0,0 +1,10 @@
+[Unit]
+Description=Run CI Transient Fault Auto-Retry every 5 minutes
+
+[Timer]
+OnBootSec=2min
+OnUnitActiveSec=5min
+AccuracySec=30s
+
+[Install]
+WantedBy=timers.target
@@ -19,6 +19,18 @@ for arg in "$@"; do
  BUILD_ARGS="$BUILD_ARGS --build-arg $arg"
 done

+# Web 镜像 cache bust：计算 apps/web/ 的 git tree hash
+# 当源码变化时 hash 变化，buildx 的 ARG 缓存键失效 → vite build 必定重新执行
+if [ "${DOCKERFILE##*/}" = "web.Dockerfile" ]; then
+  SOURCE_HASH=$(git rev-parse HEAD:apps/web 2>/dev/null || echo "")
+  if [ -n "$SOURCE_HASH" ]; then
+    echo "Web cache bust: SOURCE_HASH=${SOURCE_HASH}"
+    BUILD_ARGS="$BUILD_ARGS --build-arg SOURCE_HASH=${SOURCE_HASH}"
+  else
+    echo "⚠️  无法计算 apps/web tree hash，跳过 cache bust"
+  fi
+fi
+
 BUILDER_NAME="ci-builder-persist"
 if ! docker buildx inspect "$BUILDER_NAME" > /dev/null 2>&1; then
  echo "持久 builder 不存在，创建中..."
@@ -31,6 +31,18 @@ for arg in "$@"; do
  BUILD_ARGS="$BUILD_ARGS --build-arg $arg"
 done

+# Web 镜像 cache bust：计算 apps/web/ 的 git tree hash
+# 当源码变化时 hash 变化，buildx 的 ARG 缓存键失效 → vite build 必定重新执行
+if [ "${DOCKERFILE##*/}" = "web.Dockerfile" ]; then
+  SOURCE_HASH=$(git rev-parse HEAD:apps/web 2>/dev/null || echo "")
+  if [ -n "$SOURCE_HASH" ]; then
+    echo "Web cache bust: SOURCE_HASH=${SOURCE_HASH}"
+    BUILD_ARGS="$BUILD_ARGS --build-arg SOURCE_HASH=${SOURCE_HASH}"
+  else
+    echo "⚠️  无法计算 apps/web tree hash，跳过 cache bust"
+  fi
+fi
+
 # 确保持久 builder 存在并使用（幂等）
 if ! docker buildx inspect "$BUILDER_NAME" > /dev/null 2>&1; then
  echo "持久 builder 不存在，创建中..."
@@ -9,10 +9,16 @@ PR自动扫描器：扫描所有open PR，对CI全绿的进行自动审批/合
 import argparse
 import json
 import re
+import socket
 import time
 import urllib.error
 import urllib.request

+# 单次HTTP请求超时（秒），防止网络异常时永久阻塞占住runner
+HTTP_TIMEOUT = 15
+# 整次扫描墙钟上限（秒），到点主动退出（Gitea 1.26的timeout-minutes不可靠，脚本自保）
+DEFAULT_WALL_SECONDS = 240
+

 def api_request(token, repo, endpoint, method="GET", data=None):
    """Gitea API请求"""
@@ -29,7 +35,7 @@ def api_request(token, repo, endpoint, method="GET", data=None):
    ctx.verify_mode = ssl.CERT_NONE

    try:
-        resp = urllib.request.urlopen(req, context=ctx)
+        resp = urllib.request.urlopen(req, context=ctx, timeout=HTTP_TIMEOUT)
        return json.loads(resp.read().decode()), resp.status
    except urllib.error.HTTPError as e:
        body = e.read().decode()
@@ -39,6 +45,9 @@ def api_request(token, repo, endpoint, method="GET", data=None):
            except json.JSONDecodeError:
                return {"error": body}, e.code
        return {"error": str(e)}, e.code
+    except (urllib.error.URLError, socket.timeout, TimeoutError, OSError) as e:
+        # 网络不可达/超时：返回599让调用方按失败处理，绝不永久挂起
+        return {"error": f"request-failed: {e}"}, 599


 def get_open_prs(token, repo, base="develop"):
@@ -223,10 +232,12 @@ def add_pr_label(token, repo, pr_number, label):
    return code in (200, 201)


-def merge_pr(token, repo, pr_number):
+def merge_pr(token, repo, pr_number, deadline=None):
    """合并PR（squash merge）"""
-    # 等待几秒让状态同步
-    time.sleep(30)
+    # 等待几秒让状态同步（可被墙钟上限打断，最多等30秒）
+    wait_end = min(time.monotonic() + 30, deadline) if deadline else time.monotonic() + 30
+    while time.monotonic() < wait_end:
+        time.sleep(2)

    # 检查PR状态
    pr_data, code = api_request(token, repo, f"pulls/{pr_number}")
@@ -268,11 +279,23 @@ def main():
    parser.add_argument("--dry-run", default="false", help="试运行模式")
    parser.add_argument("--max-prs", type=int, default=20, help="最多处理的PR数")
    parser.add_argument("--skip-ai-review", action="store_true", help="跳过AI审查检查（强制审批）")
+    parser.add_argument(
+        "--max-wall-seconds",
+        type=int,
+        default=DEFAULT_WALL_SECONDS,
+        help="整次扫描墙钟上限（秒），到点主动退出，默认240",
+    )

    args = parser.parse_args()

    dry_run = args.dry_run.lower() == "true"

+    # 墙钟自保：Gitea 1.26 的 timeout-minutes 对卡死 job 不生效，脚本自己兜底
+    wall_deadline = time.monotonic() + args.max_wall_seconds
+
+    def wall_expired():
+        return time.monotonic() >= wall_deadline
+
    # required contexts（与分支保护一致）
    REQUIRED_CONTEXTS_FULL = [
        # 统一使用CI Gate作为合并门禁（与pr-automation和分支保护保持一致）
@@ -301,6 +324,9 @@ def main():
    ai_blocked_count = 0

    for pr in prs[: args.max_prs]:
+        if wall_expired():
+            print(f"\n⏰ 达到墙钟上限 {args.max_wall_seconds}s，停止处理剩余PR（下次调度继续）")
+            break
        pr_num = pr["number"]
        pr_title = pr["title"]
        head_sha = pr["head"]["sha"]
@@ -379,11 +405,14 @@ def main():
            approved = has_approval(args.token, args.repo, pr_num)

            if merge_ok and approved and not merge_failed:
+                if wall_expired():
+                    print("⏰ 达到墙钟上限，跳过本次合并（下次调度继续）")
+                    break
                if dry_run:
                    print("  🎯 [DRY-RUN] 将自动合并")
                else:
                    print("  🎯 执行自动合并...")
-                    ok, msg = merge_pr(args.token, args.repo, pr_num)
+                    ok, msg = merge_pr(args.token, args.repo, pr_num, deadline=wall_deadline)
                    if ok:
                        print(f"  ✅ 合并成功: {msg}")
                        merged_count += 1
@@ -50,8 +50,59 @@ if [ -z "$SRC_TAG" ]; then
 fi

 echo "✅ 源镜像: ${IMAGE}:${SRC_TAG}"
+
+# ====== 镜像内容校验（CI 加固 - 防止静默部署旧/损坏镜像） ======
+echo ""
+echo "--- 镜像内容校验 ---"
+
+# 1. 检查镜像 layers 有效性
+LAYERS=$(docker inspect --format='{{len .RootFS.Layers}}' "${IMAGE}:${SRC_TAG}" 2>/dev/null || echo "0")
+echo "源镜像 layers 数量: $LAYERS"
+if [ "$LAYERS" -eq 0 ]; then
+  echo "❌ 源镜像无有效 layers，可能为损坏镜像，拒绝 retag"
+  exit 1
+fi
+
+# 2. 镜像创建时间检查
+CREATED=$(docker inspect --format='{{.Created}}' "${IMAGE}:${SRC_TAG}")
+echo "源镜像创建时间: $CREATED"
+
+# 解析创建时间距现在多少小时
+CREATED_EPOCH=$(date -d "$CREATED" +%s 2>/dev/null || echo "0")
+NOW_EPOCH=$(date +%s)
+AGE_HOURS=$(( (NOW_EPOCH - CREATED_EPOCH) / 3600 ))
+echo "源镜像年龄: ${AGE_HOURS}小时"
+if [ "$AGE_HOURS" -gt 72 ]; then
+  echo "⚠️  警告: 源镜像已超过 72 小时，可能存在分支 tag 未更新的风险"
+  echo "   请检查最近几次 CI 是否对该服务成功构建过"
+fi
+
+# 3. Digest 审计记录
+SRC_DIGEST=$(docker inspect --format='{{.Id}}' "${IMAGE}:${SRC_TAG}")
+echo "源镜像 ID: $SRC_DIGEST"
+
+echo "--- 镜像内容校验通过 ---"
+echo ""
+
 docker tag "${IMAGE}:${SRC_TAG}" "${NEW_REF}"

 # 推新 SHA tag；分支 tag 若指向的就是源 digest 则无需重复，失败可忽略
 docker push "${NEW_REF}"
+
+NEW_DIGEST=$(docker inspect --format='{{.Id}}' "${NEW_REF}")
 echo "✅ retag 推送完成: ${NEW_REF} (from ${SRC_TAG})"
+echo "   源 digest: ${SRC_DIGEST}"
+echo "   新 tag digest: ${NEW_DIGEST}"
+
+# 写入 retag 审计记录（供 deploy 步骤参考）
+RETAG_LOG="/tmp/retag-audit-$(date +%s).txt"
+cat > "$RETAG_LOG" <<EOF
+timestamp=$(date -u +%Y-%m-%dT%H:%M:%SZ)
+image=${IMAGE}
+src_tag=${SRC_TAG}
+new_tag=${NEW_TAG}
+src_digest=${SRC_DIGEST}
+new_digest=${NEW_DIGEST}
+age_hours=${AGE_HOURS}
+EOF
+echo "审计记录: $RETAG_LOG"
@@ -1,18 +1,39 @@
 #!/bin/sh
-# CI 公共步骤：Checkout 代码（带重试）
+# CI 公共步骤：Checkout 代码（流式下载+解压，带重试）
 # 用法：直接 source 或调用，需要 GITHUB_TOKEN 环境变量
 set -eu

 python3 - <<'PY'
-import io, os, tarfile, time, urllib.request, urllib.error
+import io, os, sys, tarfile, time, urllib.request, urllib.error
+
 url = f"{os.environ['GITHUB_API_URL']}/repos/{os.environ['GITHUB_REPOSITORY']}/archive/{os.environ['GITHUB_SHA']}.tar.gz"
 request = urllib.request.Request(url, headers={"Authorization": f"token {os.environ['GITHUB_TOKEN']}"})
 last_err = None
+
 for attempt in range(5):
    try:
        with urllib.request.urlopen(request, timeout=120) as response:
-            archive = response.read()
-        break
+            # 流式读取：先读少量数据确认连接成功，再大块读取
+            first_chunk = response.read(8192)
+            buf = io.BytesIO()
+            buf.write(first_chunk)
+            while True:
+                chunk = response.read(65536)
+                if not chunk:
+                    break
+                buf.write(chunk)
+            buf.seek(0)
+            with tarfile.open(fileobj=buf, mode='r:gz') as tar:
+                root_prefix = tar.getmembers()[0].name.split('/', 1)[0] + '/'
+                for member in tar.getmembers():
+                    name = member.name
+                    if name == root_prefix[:-1]:
+                        continue
+                    if name.startswith(root_prefix):
+                        member.name = name[len(root_prefix):]
+                    if member.name:
+                        tar.extract(member, '.')
+            break
    except urllib.error.HTTPError as e:
        last_err = e
        if e.code >= 500 and attempt < 4:
@@ -31,14 +52,5 @@ for attempt in range(5):
        raise
 else:
    raise last_err
-with tarfile.open(fileobj=io.BytesIO(archive), mode='r:gz') as tar:
-    root_prefix = tar.getmembers()[0].name.split('/', 1)[0] + '/'
-    for member in tar.getmembers():
-        name = member.name
-        if name == root_prefix[:-1]:
-            continue
-        if name.startswith(root_prefix):
-            member.name = name[len(root_prefix):]
-        if member.name:
-            tar.extract(member, '.')
 PY
+# CI pipeline speedup batch 1
@@ -1,14 +1,20 @@
 #!/bin/bash
-# CI Validate: 代码质量与安全扫描（并行Job 1/3）
-# 包含：密钥扫描、格式检查、安全扫描、依赖漏洞、死代码检测、脚本语法校验
+# CI Validate: 安全扫描（validate-security）
+# 包含：密钥扫描、bandit 安全扫描（仅告警）、pip-audit 依赖漏洞（仅告警）、CI 脚本语法校验
 set -eu

-echo "=== CI Validate: 代码质量与安全扫描 ==="
+echo "=== CI Validate: 安全扫描 ==="

 # --- 密钥检测 ---
 echo ""
-echo "=== [1/6] Secret detection (detect-secrets) ==="
-python3 -m pip install -q detect-secrets
+echo "=== [1/4] Secret detection (detect-secrets) ==="
+python3 -m pip install -q --no-cache-dir detect-secrets || {
+  echo "⚠️  detect-secrets install failed, retrying without cache..."
+  python3 -m pip install -q --no-cache-dir --no-binary :all: detect-secrets || {
+    echo "❌ detect-secrets install failed after retry"
+    exit 1
+  }
+}
 detect-secrets --version

 detect-secrets scan \
@@ -56,23 +62,9 @@ for fpath, items in data.get('results', {}).items():
 fi
 echo "✅ Secret scan passed"

-# --- 代码质量检查（全量，PR 和 push 统一标准）---
-# 历史：PR 侧用增量检查以加速，但会导致 push 侧全量检查失败时 PR 侧感知不到
-# 现在统一全量检查，确保 CI 真正保护主分支（black/isort/ruff 全量仅多几十秒）
+# --- Bandit 安全扫描（仅告警）---
 echo ""
-echo "=== [2/6] Code quality checks (full scan) ==="
-SCAN_MODE="full"
-echo "Full scan mode"
-python3 -m compileall -q alembic apps packages tests scripts
-python3 -m black --check --fast alembic apps packages tests scripts
-python3 -m isort --check-only alembic apps packages tests scripts
-python3 -m ruff check apps packages tests --statistics
-
-echo "✅ Code quality checks passed"
-
-# --- Bandit 安全扫描（仅告警） ---
-echo ""
-echo "=== [3/6] Security scan (bandit, advisory only) ==="
+echo "=== [2/4] Security scan (bandit, advisory only) ==="
 set +e
 bandit -r apps packages -q -ll
 BANDIT_EXIT=$?
@@ -83,42 +75,34 @@ else
  echo "✅ Bandit security scan passed"
 fi

-# --- Pip-audit 依赖漏洞扫描（仅告警） ---
+# --- Pip-audit 依赖漏洞扫描（仅告警）---
 echo ""
-echo "=== [4/6] Python dependency vulnerability scan (pip-audit, advisory only) ==="
-python3 -m pip install -q pip-audit
-pip-audit --version
-EXIT_CODE=0
-for req_file in requirements.txt requirements-base.txt requirements-dev.txt; do
-  if [ -f "$req_file" ]; then
-    echo "--- Scanning $req_file ---"
-    pip-audit -r "$req_file" --desc on 2>&1 | head -40 || EXIT_CODE=$?
-    echo ""
-  fi
-done
-echo "pip-audit scan completed (advisory mode - warnings only, not blocking CI)"
-
-# --- Vulture 死代码检测（仅告警） ---
-echo ""
-echo "=== [5/6] Dead code detection (vulture, advisory only) ==="
-set +e
-python3 -m pip install -q vulture
-vulture --version
-echo "告警模式，不阻断CI。置信度>=90%建议尽快确认。"
-echo ""
-vulture apps packages scripts \
-  --exclude "tests,test,migrations,.gitea,docs,node_modules,site-packages,*/test_*.py,*/conftest.py" \
-  --min-confidence 70 \
-  2>&1 | sort -t'(' -k2 -rn | head -80
-echo ""
-echo "=== vulture scan summary ==="
-echo "发现潜在死代码（可能包含框架装饰器注册的函数，为误报）"
-echo "建议：定期人工审查高置信度（>=90%）条目"
-set -e
+echo "=== [3/4] Python dependency vulnerability scan (pip-audit, advisory only) ==="
+python3 -m pip install -q --no-cache-dir pip-audit || {
+  echo "⚠️  pip-audit install failed (cache issue?), retrying..."
+  python3 -m pip install -q --no-cache-dir pip-audit || {
+    echo "⚠️  pip-audit unavailable, skipping dependency vulnerability scan (advisory)"
+    pip-audit --version 2>/dev/null || true
+  }
+}
+if command -v pip-audit >/dev/null 2>&1 || python3 -m pip show pip-audit >/dev/null 2>&1; then
+  pip-audit --version
+  EXIT_CODE=0
+  for req_file in requirements.txt requirements-base.txt requirements-dev.txt; do
+    if [ -f "$req_file" ]; then
+      echo "--- Scanning $req_file ---"
+      pip-audit -r "$req_file" --desc on 2>&1 | head -40 || EXIT_CODE=$?
+      echo ""
+    fi
+  done
+  echo "pip-audit scan completed (advisory mode - warnings only, not blocking CI)"
+else
+  echo "⚠️  pip-audit not available, skipping dependency vulnerability scan (advisory)"
+fi

 # --- CI脚本语法校验 ---
 echo ""
-echo "=== [6/6] CI & shell scripts syntax validation ==="
+echo "=== [4/4] CI & shell scripts syntax validation ==="
 SYNTAX_ERROR=0
 # 检查所有 CI shell 脚本
 for script in scripts/ci/*.sh; do
@@ -154,4 +138,4 @@ fi
 echo "✅ All CI scripts syntax OK"

 echo ""
-echo "=== CI Validate: 代码质量与安全扫描 全部通过 ✅ ==="
+echo "=== CI Validate: 安全扫描 全部通过 ✅ ==="
@@ -0,0 +1,42 @@
+#!/bin/bash
+# CI Validate: 代码风格检查（validate-style）
+# 包含：Python 字节码编译、black 格式、isort 排序、ruff lint、vulture 死代码（仅告警）
+set -eu
+
+echo "=== CI Validate: 代码风格检查 ==="
+
+# --- Python 字节码编译 ---
+echo ""
+echo "=== [1/3] Python bytecode compilation ==="
+python3 -m compileall -q alembic apps packages tests scripts
+echo "✅ Bytecode compilation passed"
+
+# --- 代码格式检查（全量）---
+echo ""
+echo "=== [2/3] Code formatting (black + isort + ruff) ==="
+echo "Full scan mode"
+python3 -m black --check --fast alembic apps packages tests scripts
+python3 -m isort --check-only alembic apps packages tests scripts
+python3 -m ruff check apps packages tests --statistics
+echo "✅ Code formatting checks passed"
+
+# --- Vulture 死代码检测（仅告警）---
+echo ""
+echo "=== [3/3] Dead code detection (vulture, advisory only) ==="
+set +e
+python3 -m pip install -q --no-cache-dir vulture || echo "⚠️  vulture install failed, skipping dead code detection"
+vulture --version
+echo "告警模式，不阻断CI。置信度>=90%建议尽快确认。"
+echo ""
+vulture apps packages scripts \
+  --exclude "tests,test,migrations,.gitea,docs,node_modules,site-packages,*/test_*.py,*/conftest.py" \
+  --min-confidence 70 \
+  2>&1 | sort -t'(' -k2 -rn | head -80
+echo ""
+echo "=== vulture scan summary ==="
+echo "发现潜在死代码（可能包含框架装饰器注册的函数，为误报）"
+echo "建议：定期人工审查高置信度（>=90%）条目"
+set -e
+
+echo ""
+echo "=== CI Validate: 代码风格检查 全部通过 ✅ ==="
@@ -260,6 +260,90 @@ fi

 echo "All images pulled."

+# ====== 镜像内容校验（CI 加固 - 防止静默部署损坏/过期镜像） ======
+echo ""
+echo "=========================================="
+echo "  镜像内容校验"
+echo "=========================================="
+
+VERIFY_FAILED=0
+DEPLOY_MANIFEST="${GENERATED_DIR}/deploy-manifest.json"
+
+# 读取上次部署的 manifest（用于对比）
+PREV_MANIFEST=""
+if [ -f "$DEPLOY_MANIFEST" ]; then
+  PREV_MANIFEST=$(cat "$DEPLOY_MANIFEST")
+  echo "上次部署 manifest 存在"
+fi
+
+NEW_MANIFEST_LINES=""
+for svc in api worker web; do
+  img_var="REGISTRY_$(echo $svc | tr '[:lower:]' '[:upper:]')"
+  img_val=$(eval echo "\$$img_var")
+
+  # 1. 检查镜像是否存在
+  if ! docker image inspect "$img_val" >/dev/null 2>&1; then
+    echo "  ❌ $svc: 镜像不存在 ($img_val)"
+    VERIFY_FAILED=$((VERIFY_FAILED + 1))
+    continue
+  fi
+
+  # 2. 检查 layers 有效性
+  LAYER_COUNT=$(docker inspect --format='{{len .RootFS.Layers}}' "$img_val" 2>/dev/null || echo "0")
+  if [ "$LAYER_COUNT" -eq 0 ]; then
+    echo "  ❌ $svc: 镜像无有效 layers ($img_val)"
+    VERIFY_FAILED=$((VERIFY_FAILED + 1))
+    continue
+  fi
+
+  # 3. 获取 digest 和创建时间
+  IMG_ID=$(docker inspect --format='{{.Id}}' "$img_val")
+  IMG_CREATED=$(docker inspect --format='{{.Created}}' "$img_val")
+  IMG_SIZE=$(docker inspect --format='{{.Size}}' "$img_val")
+  echo "  ✅ $svc: ${LAYER_COUNT} layers, size=${IMG_SIZE}, created=${IMG_CREATED}"
+  echo "     id: $IMG_ID"
+
+  # 4. 对比上次部署
+  CHANGED="unchanged"
+  if [ -n "$PREV_MANIFEST" ]; then
+    PREV_ID=$(echo "$PREV_MANIFEST" | grep "\"${svc}_id\"" | sed 's/.*: *"\(.*\)".*/\1/' 2>/dev/null || echo "")
+    if [ -n "$PREV_ID" ] && [ "$PREV_ID" != "$IMG_ID" ]; then
+      CHANGED="changed"
+    elif [ -n "$PREV_ID" ] && [ "$PREV_ID" = "$IMG_ID" ]; then
+      CHANGED="unchanged"
+    else
+      CHANGED="unknown"
+    fi
+  else
+    CHANGED="first-deploy"
+  fi
+  echo "     vs last deploy: $CHANGED"
+
+  NEW_MANIFEST_LINES="${NEW_MANIFEST_LINES}  \"${svc}_id\": \"${IMG_ID}\",
+  \"${svc}_created\": \"${IMG_CREATED}\",
+  \"${svc}_layers\": ${LAYER_COUNT},
+"
+done
+
+if [ "$VERIFY_FAILED" -gt 0 ]; then
+  echo ""
+  echo "ERROR: $VERIFY_FAILED 个镜像校验失败，拒绝部署"
+  exit 1
+fi
+
+# 写入新 manifest
+cat > "$DEPLOY_MANIFEST" <<MANIFEST_EOF
+{
+  "deployed_at": "$(date -u +%Y-%m-%dT%H:%M:%SZ)",
+  "image_tag": "$IMAGE_TAG",
+${NEW_MANIFEST_LINES}  "verified": true
+}
+MANIFEST_EOF
+echo ""
+echo "部署 manifest 已更新: $DEPLOY_MANIFEST"
+echo "=========================================="
+echo ""
+
 echo "Backing up legacy assets from current web container..."
 if docker inspect xiaoxia-web-staging >/dev/null 2>&1; then
  _tmpdir="/tmp/legacy-assets-$$"
@@ -386,14 +386,14 @@ class TestSmartMatchFiltersExhausted:
            _fresh_asset("a-fresh-1"),
        ]
        resp = self._call(assets)
-        returned_ids = {item.asset.id for item in resp.items}
+        returned_ids = {item.id for item in resp.items}
        assert "a-fresh-1" in returned_ids
        assert "a-exhausted-1" not in returned_ids
        assert "a-exhausted-2" not in returned_ids
        # total_candidates 是过滤前的候选总数
        assert resp.total_candidates == 3
        # 返回的素材全部 usable=True
-        assert all(item.asset.usable for item in resp.items)
+        assert all(item.usable for item in resp.items)

    def test_all_exhausted_returns_empty(self):
        """全部素材已用尽时返回空列表（不报错，前端显示空结果）。"""
@@ -406,4 +406,49 @@ class TestSmartMatchFiltersExhausted:
        assets = [_fresh_asset("a-1"), _fresh_asset("a-2")]
        resp = self._call(assets)
        assert len(resp.items) == 2
-        assert all(item.asset.usable for item in resp.items)
+        assert all(item.usable for item in resp.items)
+
+
+class TestSmartMatchFlatStructure:
+    """P0 回归：smart-match 响应必须扁平——item 顶层直接可读素材字段，
+    前端 items.map(a => a.id) 不能再拿到 undefined（此前 item.asset 嵌套包装
+    导致 GET /assets/undefined 404 + from-assets 422，自动模式全链路断裂）。"""
+
+    def test_item_id_at_top_level(self):
+        """item.id 直接在顶层可读，不存在 item.asset 包装层。"""
+        assets = [_fresh_asset("a-flat-1"), _fresh_asset("a-flat-2")]
+        resp = TestSmartMatchFiltersExhausted()._call(assets)
+        ids = [item.id for item in resp.items]
+        assert ids == ["a-flat-1", "a-flat-2"]
+        # 嵌套 asset 字段已移除
+        assert all(not hasattr(item, "asset") for item in resp.items)
+
+    def test_item_is_asset_response_superset(self):
+        """条目携带 AssetResponse 全部关键字段 + usable/余量，前端可直接渲染卡片。"""
+        assets = [_fresh_asset("a-fields-1")]
+        resp = TestSmartMatchFiltersExhausted()._call(assets)
+        item = resp.items[0]
+        assert item.id == "a-fields-1"
+        assert item.name == "fresh-a-fields-1"
+        assert item.storage_key == "key/test-asset.mp4"
+        assert item.mime_type == "video/mp4"
+        assert item.duration == 60.0
+        assert item.status == "ready"
+        assert item.thumbnail_url is None or item.thumbnail_url.startswith("http")
+        # 余量/可用性字段顶层可读（isAssetUsable 依赖）
+        assert item.usable is True
+        assert item.used_duration == 0.0
+        assert item.available_duration == 60.0
+        assert item.used_ratio == 0.0
+        # 评分字段保留
+        assert 0.0 <= item.score <= 100.0
+        assert isinstance(item.breakdown, dict)
+
+    def test_score_and_breakdown_preserved(self):
+        """扁平化后评分字段不丢失。"""
+        assets = [_fresh_asset("a-score-1")]
+        resp = TestSmartMatchFiltersExhausted()._call(assets)
+        item = resp.items[0]
+        assert isinstance(item.score, float)
+        assert item.score > 0
+        assert isinstance(item.breakdown, dict) and item.breakdown
@@ -0,0 +1,221 @@
+"""Tests for duplicate_rate computation and API response."""
+
+import json
+import sys
+from pathlib import Path
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+# Mock cv2 and numpy before any imports that need them
+sys.modules.setdefault("cv2", MagicMock())
+
+ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(ROOT / "apps" / "api"))
+sys.path.insert(0, str(ROOT / "packages"))
+sys.path.insert(0, str(ROOT / "apps" / "worker"))
+
+
+class TestComputeDuplicateRate:
+    """Test VideoDeduplicator.compute_duplicate_rate."""
+
+    def _make_fingerprint(self, md5="abc123", phashes=None):
+        from video_processing.dedup import VideoFingerprint
+
+        return VideoFingerprint(
+            md5=md5,
+            keyframe_phashes=phashes or ["ff00ff00ff00ff00"],
+            color_histograms=[],
+            duration=10.0,
+            resolution=(1920, 1080),
+        )
+
+    def _make_existing_video(self, vid, fingerprint_dict):
+        from packages.domain import GeneratedVideo
+
+        return GeneratedVideo(
+            id=vid,
+            project_id="proj1",
+            generation_task_id="task1",
+            name=f"video-{vid}",
+            file_url=f"https://example.com/{vid}.mp4",
+            file_size=1000,
+            duration=10.0,
+            width=1920,
+            height=1080,
+            fps=25.0,
+            video_fingerprint=fingerprint_dict,
+        )
+
+    def test_no_existing_videos_returns_zero(self):
+        from video_processing.dedup import VideoDeduplicator
+
+        deduplicator = VideoDeduplicator()
+        fingerprint = self._make_fingerprint()
+        session = MagicMock()
+
+        with patch("video_processing.dedup.SQLAlchemyGeneratedVideoRepository") as MockRepo:
+            mock_repo = MockRepo.return_value
+            session.query.return_value.filter.return_value.order_by.return_value.limit.return_value.all.return_value = (
+                []
+            )
+            rate = deduplicator.compute_duplicate_rate(fingerprint, "proj1", "vid1", session)
+
+        assert rate == 0.0
+
+    def test_md5_match_returns_100(self):
+        from video_processing.dedup import VideoDeduplicator
+
+        from packages.adapters.sqlalchemy_impl.models import GeneratedVideoModel
+
+        deduplicator = VideoDeduplicator()
+        fingerprint = self._make_fingerprint(md5="exact_match_md5")
+        session = MagicMock()
+
+        existing = self._make_existing_video("existing1", {"md5": "exact_match_md5", "keyframe_phashes": ["aa"]})
+        # Create a mock model with the domain attributes
+        mock_model = MagicMock(spec=GeneratedVideoModel)
+        mock_model.id = existing.id
+        mock_model.project_id = existing.project_id
+        mock_model.video_fingerprint = existing.video_fingerprint
+        mock_model.generated_at = "2026-01-01"
+
+        with patch("video_processing.dedup.SQLAlchemyGeneratedVideoRepository") as MockRepo:
+            mock_repo = MockRepo.return_value
+            mock_repo._to_domain.return_value = existing
+            # Mock the session.query chain
+            session.query.return_value.filter.return_value.order_by.return_value.limit.return_value.all.return_value = [
+                mock_model
+            ]
+            rate = deduplicator.compute_duplicate_rate(fingerprint, "proj1", "vid1", session)
+
+        assert rate == 100.0
+
+    def test_phash_similarity_computed(self):
+        from video_processing.dedup import VideoDeduplicator
+
+        from packages.adapters.sqlalchemy_impl.models import GeneratedVideoModel
+
+        deduplicator = VideoDeduplicator()
+        fingerprint = self._make_fingerprint(md5="different_md5", phashes=["ff00ff00ff00ff00"])
+        session = MagicMock()
+
+        existing = self._make_existing_video(
+            "existing1",
+            {"md5": "other_md5", "keyframe_phashes": ["ff00ff00ff00ff03"]},
+        )
+        mock_model = MagicMock(spec=GeneratedVideoModel)
+        mock_model.id = existing.id
+        mock_model.project_id = existing.project_id
+        mock_model.video_fingerprint = existing.video_fingerprint
+        mock_model.generated_at = "2026-01-01"
+
+        with patch("video_processing.dedup.SQLAlchemyGeneratedVideoRepository") as MockRepo:
+            mock_repo = MockRepo.return_value
+            mock_repo._to_domain.return_value = existing
+            session.query.return_value.filter.return_value.order_by.return_value.limit.return_value.all.return_value = [
+                mock_model
+            ]
+            rate = deduplicator.compute_duplicate_rate(fingerprint, "proj1", "vid1", session)
+
+        # hamming distance = 2, similarity = (1 - 2/64) * 100 = 96.875
+        assert rate == pytest.approx(96.88, abs=0.1)
+
+    def test_excludes_self_video(self):
+        from video_processing.dedup import VideoDeduplicator
+
+        from packages.adapters.sqlalchemy_impl.models import GeneratedVideoModel
+
+        deduplicator = VideoDeduplicator()
+        fingerprint = self._make_fingerprint(md5="same_md5")
+        session = MagicMock()
+
+        self_video = self._make_existing_video("vid1", {"md5": "same_md5", "keyframe_phashes": ["aa"]})
+        mock_model = MagicMock(spec=GeneratedVideoModel)
+        mock_model.id = self_video.id
+        mock_model.project_id = self_video.project_id
+        mock_model.video_fingerprint = self_video.video_fingerprint
+        mock_model.generated_at = "2026-01-01"
+
+        with patch("video_processing.dedup.SQLAlchemyGeneratedVideoRepository") as MockRepo:
+            mock_repo = MockRepo.return_value
+            mock_repo._to_domain.return_value = self_video
+            session.query.return_value.filter.return_value.order_by.return_value.limit.return_value.all.return_value = [
+                mock_model
+            ]
+            rate = deduplicator.compute_duplicate_rate(fingerprint, "proj1", "vid1", session)
+
+        assert rate == 0.0
+
+    def test_takes_max_similarity(self):
+        from video_processing.dedup import VideoDeduplicator
+
+        from packages.adapters.sqlalchemy_impl.models import GeneratedVideoModel
+
+        deduplicator = VideoDeduplicator()
+        fingerprint = self._make_fingerprint(md5="new_md5", phashes=["ff00ff00ff00ff00"])
+        session = MagicMock()
+
+        existing1 = self._make_existing_video("e1", {"md5": "md5_1", "keyframe_phashes": ["ff00ff00ff00ff0f"]})
+        existing2 = self._make_existing_video("e2", {"md5": "md5_2", "keyframe_phashes": ["ff00ff00ff00ff01"]})
+        mock_model1 = MagicMock(spec=GeneratedVideoModel)
+        mock_model1.id = existing1.id
+        mock_model1.project_id = existing1.project_id
+        mock_model1.video_fingerprint = existing1.video_fingerprint
+        mock_model1.generated_at = "2026-01-02"
+        mock_model2 = MagicMock(spec=GeneratedVideoModel)
+        mock_model2.id = existing2.id
+        mock_model2.project_id = existing2.project_id
+        mock_model2.video_fingerprint = existing2.video_fingerprint
+        mock_model2.generated_at = "2026-01-01"
+
+        with patch("video_processing.dedup.SQLAlchemyGeneratedVideoRepository") as MockRepo:
+            mock_repo = MockRepo.return_value
+            mock_repo._to_domain.side_effect = [existing1, existing2]
+            session.query.return_value.filter.return_value.order_by.return_value.limit.return_value.all.return_value = [
+                mock_model1,
+                mock_model2,
+            ]
+            rate = deduplicator.compute_duplicate_rate(fingerprint, "proj1", "vid1", session)
+
+        # max similarity: e2 distance=1, (1-1/64)*100 = 98.4375
+        assert rate == pytest.approx(98.44, abs=0.1)
+
+
+class TestDuplicateRateAPI:
+    """Test that duplicate_rate is returned in API responses."""
+
+    def test_video_item_response_has_duplicate_rate(self):
+        from app.schemas.video_center import VideoItemResponse
+
+        resp = VideoItemResponse(
+            id="v1",
+            project_id="p1",
+            generation_task_id="t1",
+            name="test.mp4",
+            file_url="https://example.com/test.mp4",
+            file_size=1000,
+            duration=10.0,
+            width=1920,
+            height=1080,
+            fps=25.0,
+            duplicate_rate=75.5,
+        )
+        assert resp.duplicate_rate == 75.5
+
+    def test_video_item_response_duplicate_rate_default_none(self):
+        from app.schemas.video_center import VideoItemResponse
+
+        resp = VideoItemResponse(
+            id="v1",
+            project_id="p1",
+            generation_task_id="t1",
+            name="test.mp4",
+            file_url="https://example.com/test.mp4",
+            file_size=1000,
+            duration=10.0,
+            width=1920,
+            height=1080,
+            fps=25.0,
+        )
+        assert resp.duplicate_rate is None
@@ -747,3 +747,83 @@ class TestReuseRatioGate:
        assert len(clips_data) == 2
        # 耗尽素材被跳过，两个片段都分配给新鲜素材
        assert all(c["asset_id"] == "fresh" for c in clips_data)
+
+
+# ── P0 回归：from-assets 对 undefined/null/空串 asset_ids 容错 ──────────────
+# 线上事故：前端 smart-match 拿到 {asset: {...}} 包装层后 items.map(a=>a.id)
+# 全为 undefined，POST /clips/from-assets 携带 null → 422，自动模式预览断裂。
+
+
+class TestClipsFromAssetsInvalidIds:
+    def test_schema_filters_null_and_empty_ids(self):
+        """请求 schema 在 pre 阶段剔除 null/空串/空白 id，不触发 422。"""
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1", None, "", "  ", "a2"])  # type: ignore[list-item]
+        assert body.asset_ids == ["a1", "a2"]
+
+    def test_schema_all_invalid_raises(self):
+        """全部为非法 id 时 min_length=1 兜底报校验错误（前端得到 422 而非脏数据）。"""
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+        from pydantic import ValidationError
+
+        with pytest.raises(ValidationError):
+            ClipsFromAssetsRequest(asset_ids=[None, "", "  "])  # type: ignore[list-item]
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_route_filters_invalid_ids_and_uses_valid(self, mock_storage):
+        """路由层二次兜底：混有 null/空串时只用合法 id 正常创建片段。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        mock_plan_svc = _make_plan_svc(replace_return_count=2)
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get = MagicMock(side_effect=lambda aid: _make_mock_asset(aid, 30.0))
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1", None, "", "a2"])  # type: ignore[list-item]
+
+        with _patch_segments(_segments(2, dur_min=3.0, dur_max=5.0)):
+            result = create_clips_from_assets_editor(
+                template_id="tpl-001",
+                body=body,
+                background_tasks=MagicMock(),
+                plan_id=TEST_PLAN_ID,
+                services=(MagicMock(), mock_plan_svc),
+                asset_repo=mock_asset_repo,
+                db=MagicMock(),
+                current_user=_make_auth_user(),
+            )
+
+        assert result.created_count == 2
+        clips_data = _get_clips_data_from_call(mock_plan_svc)
+        assert {c["asset_id"] for c in clips_data} == {"a1", "a2"}
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_route_all_empty_ids_raises_400(self, mock_storage):
+        """schema 被绕过直接调路由、且 id 全非法时，路由 400 而非 500/422。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+
+        mock_plan_svc = _make_plan_svc()
+        mock_asset_repo = MagicMock()
+
+        body = MagicMock()
+        body.asset_ids = [None, "", "  "]
+
+        with _patch_segments(DEFAULT_SEGMENTS):
+            with pytest.raises(HTTPException) as exc_info:
+                create_clips_from_assets_editor(
+                    template_id="tpl-001",
+                    body=body,
+                    background_tasks=MagicMock(),
+                    plan_id=TEST_PLAN_ID,
+                    services=(MagicMock(), mock_plan_svc),
+                    asset_repo=mock_asset_repo,
+                    db=MagicMock(),
+                    current_user=_make_auth_user(),
+                )
+        assert exc_info.value.status_code == 400
+        mock_plan_svc.replace_all_clips_transactional.assert_not_called()
@@ -0,0 +1,93 @@
+"""Tests for get_asset_recent_use_counts and smart-match high-use exclusion."""
+
+import json
+import sys
+from pathlib import Path
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(ROOT / "apps" / "api"))
+sys.path.insert(0, str(ROOT / "packages"))
+
+
+class TestGetAssetRecentUseCounts:
+    """Test asset_segment_tracker.get_asset_recent_use_counts."""
+
+    def _make_asset_model(self, asset_id, used_time_ranges=None):
+        from packages.adapters.sqlalchemy_impl.models import AssetModel
+
+        model = AssetModel(
+            id=asset_id,
+            name=f"asset-{asset_id}",
+            file_type="video",
+            status="ready",
+            asset_library_id="lib1",
+            project_id="proj1",
+            file_size=1000,
+            file_url=f"https://example.com/{asset_id}.mp4",
+            uploaded_by_user_id="user1",
+        )
+        model.classification_result = json.dumps(
+            {
+                "used_time_ranges": used_time_ranges or [],
+            }
+        )
+        return model
+
+    def test_empty_asset_ids_returns_empty(self):
+        from app.services.asset_segment_tracker import get_asset_recent_use_counts
+
+        db = MagicMock()
+        result = get_asset_recent_use_counts(db, [])
+        assert result == {}
+
+    def test_no_usage_returns_zero(self):
+        from app.services.asset_segment_tracker import get_asset_recent_use_counts
+
+        db = MagicMock()
+        model = self._make_asset_model("a1", [])
+        db.query.return_value.filter.return_value.all.return_value = [model]
+
+        result = get_asset_recent_use_counts(db, ["a1"])
+        assert result == {"a1": 0}
+
+    def test_counts_distinct_plan_ids(self):
+        from app.services.asset_segment_tracker import get_asset_recent_use_counts
+
+        db = MagicMock()
+        ranges = [
+            {"start": 0, "end": 5, "plan_id": "plan1", "created_at": "2026-08-01T00:00:00"},
+            {"start": 5, "end": 10, "plan_id": "plan1", "created_at": "2026-08-01T00:01:00"},
+            {"start": 0, "end": 5, "plan_id": "plan2", "created_at": "2026-08-02T00:00:00"},
+            {"start": 0, "end": 5, "plan_id": "plan3", "created_at": "2026-08-03T00:00:00"},
+        ]
+        model = self._make_asset_model("a1", ranges)
+        db.query.return_value.filter.return_value.all.return_value = [model]
+
+        result = get_asset_recent_use_counts(db, ["a1"])
+        assert result == {"a1": 3}
+
+    def test_limits_to_recent_n(self):
+        from app.services.asset_segment_tracker import get_asset_recent_use_counts
+
+        db = MagicMock()
+        ranges = [
+            {"start": 0, "end": 5, "plan_id": f"plan{i}", "created_at": f"2026-08-{i+1:02d}T00:00:00"}
+            for i in range(10)
+        ]
+        model = self._make_asset_model("a1", ranges)
+        db.query.return_value.filter.return_value.all.return_value = [model]
+
+        result = get_asset_recent_use_counts(db, ["a1"], recent_video_count=5)
+        assert result["a1"] == 5
+
+    def test_missing_asset_defaults_to_zero(self):
+        from app.services.asset_segment_tracker import get_asset_recent_use_counts
+
+        db = MagicMock()
+        db.query.return_value.filter.return_value.all.return_value = []
+
+        result = get_asset_recent_use_counts(db, ["missing_asset"])
+        assert result == {"missing_asset": 0}
@@ -425,6 +425,12 @@ class TestSmartMatchEndpoint:
        for item in data["items"]:
            assert "quality" in item["breakdown"]
            assert "duration" in item["breakdown"]
+        # P0 回归：扁平结构——素材字段在 item 顶层，无 asset 包装层
+        for item in data["items"]:
+            assert item["id"]
+            assert "asset" not in item
+            assert item["mime_type"].startswith("video/")
+            assert "usable" in item

    def test_limit_parameter(self):
        project, library, assets = _make_test_data()
@@ -463,7 +469,7 @@ class TestSmartMatchEndpoint:
        assert resp.status_code == 200
        data = resp.json()
        assert len(data["items"]) == 1
-        assert data["items"][0]["asset"]["mime_type"] == "image/png"
+        assert data["items"][0]["mime_type"] == "image/png"
        # total_candidates should only count filtered-by-kind assets (1 image, not 3 videos)
        assert data["total_candidates"] == 1

@@ -0,0 +1,209 @@
+"""统一模板 segments 数据源单元测试。
+
+验证 template_repository 从 template_clip_configs 读取 segments，
+写入走 template_clip_configs，回退兼容 template_segments。
+"""
+
+from __future__ import annotations
+
+import sys
+import uuid
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[2] / "apps" / "api"))
+
+import pytest
+from sqlalchemy import create_engine
+from sqlalchemy.orm import sessionmaker
+
+from packages.adapters.sqlalchemy_impl.models import (
+    Base,
+    TemplateClipConfigModel,
+    TemplateModel,
+    TemplateSegmentModel,
+)
+from packages.adapters.sqlalchemy_impl.template_repository import (
+    SQLAlchemyTemplateRepository,
+)
+from packages.domain.template import Template, TemplateSegment
+
+
+@pytest.fixture()
+def session():
+    engine = create_engine("sqlite:///:memory:")
+    Base.metadata.create_all(engine)
+    Session = sessionmaker(bind=engine)
+    s = Session()
+    try:
+        yield s
+    finally:
+        s.close()
+
+
+@pytest.fixture()
+def repo(session):
+    return SQLAlchemyTemplateRepository(session)
+
+
+def _make_template(template_id=None, user_id="u1", name="测试模板", mode="one_take"):
+    tid = template_id or str(uuid.uuid4())
+    return Template(
+        id=tid,
+        user_id=user_id,
+        name=name,
+        mode=mode,
+        category="",
+        tags=[],
+        estimated_duration=30.0,
+        is_active=True,
+        segments=[],
+    )
+
+
+def _make_segment(template_id, order=1, material_type=None):
+    return TemplateSegment(
+        id=str(uuid.uuid4()),
+        template_id=template_id,
+        segment_order=order,
+        duration_min=5.0,
+        duration_max=10.0,
+        material_type=material_type,
+    )
+
+
+class TestCreateSegments:
+    def test_writes_to_clip_configs(self, repo, session):
+        tpl = _make_template()
+        repo.create(tpl)
+        seg = _make_segment(tpl.id, order=1)
+        repo.create_segments([seg])
+        clips = session.query(TemplateClipConfigModel).filter(TemplateClipConfigModel.template_id == tpl.id).all()
+        assert len(clips) == 1
+        assert clips[0].clip_type == "main"
+        assert clips[0].order == 1
+        assert clips[0].min_duration == 5.0
+
+    def test_material_type_stored_in_config(self, repo, session):
+        tpl = _make_template()
+        repo.create(tpl)
+        seg = _make_segment(tpl.id, order=1, material_type="voiceover")
+        repo.create_segments([seg])
+        clip = session.query(TemplateClipConfigModel).filter(TemplateClipConfigModel.template_id == tpl.id).first()
+        assert clip.config["material_type"] == "voiceover"
+
+
+class TestListSegments:
+    def test_reads_from_clip_configs(self, repo, session):
+        tpl = _make_template()
+        repo.create(tpl)
+        seg = _make_segment(tpl.id, order=1, material_type="voiceover")
+        repo.create_segments([seg])
+        result = repo.list_segments(tpl.id)
+        assert len(result) == 1
+        assert result[0].material_type == "voiceover"
+
+    def test_fallback_to_old_table(self, repo, session):
+        tpl = _make_template()
+        repo.create(tpl)
+        old = TemplateSegmentModel(
+            id=str(uuid.uuid4()),
+            template_id=tpl.id,
+            segment_order=1,
+            duration_min=3.0,
+            duration_max=8.0,
+            material_type="场景",
+        )
+        session.add(old)
+        session.commit()
+        result = repo.list_segments(tpl.id)
+        assert len(result) == 1
+        assert result[0].material_type == "场景"
+
+    def test_clip_configs_takes_priority(self, repo, session):
+        tpl = _make_template()
+        repo.create(tpl)
+        seg = _make_segment(tpl.id, order=1)
+        repo.create_segments([seg])
+        old = TemplateSegmentModel(
+            id=str(uuid.uuid4()), template_id=tpl.id, segment_order=1, duration_min=1.0, duration_max=2.0
+        )
+        session.add(old)
+        session.commit()
+        result = repo.list_segments(tpl.id)
+        assert len(result) == 1
+        assert result[0].duration_min == 5.0
+
+
+class TestListByUser:
+    def test_batch_loads_from_clip_configs(self, repo, session):
+        tpl = _make_template()
+        repo.create(tpl)
+        seg = _make_segment(tpl.id, order=1, material_type="人物")
+        repo.create_segments([seg])
+        result = repo.list_by_user("u1")
+        assert len(result) == 1
+        assert len(result[0].segments) == 1
+        assert result[0].segments[0].material_type == "人物"
+
+    def test_fallback_for_old_data(self, repo, session):
+        tpl = _make_template()
+        repo.create(tpl)
+        old = TemplateSegmentModel(
+            id=str(uuid.uuid4()), template_id=tpl.id, segment_order=1, duration_min=2.0, duration_max=6.0
+        )
+        session.add(old)
+        session.commit()
+        result = repo.list_by_user("u1")
+        assert len(result) == 1
+        assert len(result[0].segments) == 1
+        assert result[0].segments[0].duration_min == 2.0
+
+
+class TestCopyTemplate:
+    def test_copy_writes_to_clip_configs(self, repo, session):
+        tpl = _make_template()
+        repo.create(tpl)
+        seg = _make_segment(tpl.id, order=1, material_type="voiceover")
+        repo.create_segments([seg])
+        copied = repo.copy_template(tpl.id, "u1", "副本模板")
+        assert copied.id != tpl.id
+        clips = session.query(TemplateClipConfigModel).filter(TemplateClipConfigModel.template_id == copied.id).all()
+        assert len(clips) == 1
+        assert clips[0].config["material_type"] == "voiceover"
+
+    def test_copy_empty_segments(self, repo, session):
+        tpl = _make_template()
+        repo.create(tpl)
+        copied = repo.copy_template(tpl.id, "u1", "空副本")
+        assert len(copied.segments) == 0
+
+
+class TestDelete:
+    def test_delete_cleans_both_tables(self, repo, session):
+        tpl = _make_template()
+        repo.create(tpl)
+        seg = _make_segment(tpl.id, order=1)
+        repo.create_segments([seg])
+        old = TemplateSegmentModel(
+            id=str(uuid.uuid4()), template_id=tpl.id, segment_order=1, duration_min=1.0, duration_max=2.0
+        )
+        session.add(old)
+        session.commit()
+        repo.delete(tpl.id, "u1")
+        c1 = session.query(TemplateClipConfigModel).filter(TemplateClipConfigModel.template_id == tpl.id).count()
+        c2 = session.query(TemplateSegmentModel).filter(TemplateSegmentModel.template_id == tpl.id).count()
+        assert c1 == 0
+        assert c2 == 0
+
+    def test_delete_segments_by_template(self, repo, session):
+        tpl = _make_template()
+        repo.create(tpl)
+        seg = _make_segment(tpl.id, order=1)
+        repo.create_segments([seg])
+        old = TemplateSegmentModel(
+            id=str(uuid.uuid4()), template_id=tpl.id, segment_order=2, duration_min=1.0, duration_max=2.0
+        )
+        session.add(old)
+        session.commit()
+        count = repo.delete_segments_by_template(tpl.id)
+        assert count == 2