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
26 changed files with 1344 additions and 318 deletions
@@ -81,12 +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
@@ -125,14 +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: ''
@@ -156,11 +150,12 @@ jobs:
      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
@@ -184,17 +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
@@ -210,7 +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
@@ -223,7 +210,7 @@ jobs:
        CI_NOTIFY_WEBHOOK: ${{ secrets.CI_NOTIFY_WEBHOOK }}
      run: |
        set +e
-        NOTIFY_MODE=failure JOB_NAME="Validate - Code Quality" python3 scripts/ci_notify.py
+        NOTIFY_MODE=failure JOB_NAME="Validate - Style" python3 scripts/ci_notify.py
    - name: Report CI trace
      if: always()
      shell: sh
@@ -236,12 +223,16 @@ jobs:
        [ -f /tmp/ci_job_start_time ] && START_TIME=$(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:
@@ -260,6 +251,116 @@ jobs:
    - 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: |
@@ -285,6 +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
@@ -293,7 +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
@@ -306,7 +410,7 @@ jobs:
        CI_NOTIFY_WEBHOOK: ${{ secrets.CI_NOTIFY_WEBHOOK }}
      run: |
        set +e
-        NOTIFY_MODE=failure JOB_NAME="Validate - Type Check (mypy)" python3 scripts/ci_notify.py
+        NOTIFY_MODE=failure JOB_NAME="Validate - Python (mypy + alembic)" python3 scripts/ci_notify.py
    - name: Report CI trace
      if: always()
      shell: sh
@@ -319,92 +423,6 @@ jobs:
        [ -f /tmp/ci_job_start_time ] && START_TIME=$(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: 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: 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]
@@ -441,6 +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') }}
@@ -495,9 +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'
@@ -652,6 +668,7 @@ jobs:
      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') }}
@@ -858,19 +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
@@ -881,7 +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
@@ -1139,9 +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
@@ -1399,8 +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
@@ -1886,9 +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
@@ -1896,14 +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
@@ -1912,9 +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 }}
@@ -1926,9 +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: $RESULT_FRONTEND_LINT"
@@ -1968,9 +1970,9 @@ jobs:
          # 必填检查项（根据PR类型决定）
          # 通用检查（所有PR都必须过）
          REQUIRED_GENERAL=(
-            "validate-code-quality:$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"
@@ -2065,4 +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
@@ -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
@@ -608,6 +608,38 @@ 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 = [
@@ -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),
    )


@@ -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
@@ -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))

@@ -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}...${after}?per_page=300"
@@ -19,6 +19,18 @@ for arg in "$@"; 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..."
@@ -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 ✅ ==="
@@ -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
@@ -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}
@@ -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