Add pipeline governance orchestrator

.github/workflows/ci.yml

@@ -26,6 +26,10 @@ jobs:
         run: |
           python -m py_compile $(git ls-files '*.py')
 
+      - name: Unit tests
+        run: |
+          python -m unittest discover -s tests -p 'test_*.py'
+
       - name: Import and routing smoke check
         run: |
           python scripts/smoke_imports.py
@@ -34,15 +38,27 @@ jobs:
         run: |
           python scripts/verify_cta_resources.py
 
+      - name: Pipeline preflight dry-run
+        run: |
+          python scripts/preflight.py --input examples/example_article.md --skip-command-checks --json
+
+      - name: Orchestrator governance dry-run
+        run: |
+          python scripts/orchestrator.py --input examples/example_article.md --output-dir output/ci-dry-run --log /tmp/md2video-ci-pipeline.jsonl --dry-run --skip-command-checks --allow-dirty-output
+
       - name: JSON validation
         run: |
           python -c "import json; json.load(open('harness/video-rules.json'))"
           python -c "import json; json.load(open('rules/segment_types.json'))"
           python -c "import json; json.load(open('rules/storyboard_rules.json'))"
 
-      - name: Self report dry-run
+      - name: Self report no-write dry-run
+        run: |
+          python harness/self_report.py --no-write --json
+
+      - name: Privacy gate
         run: |
-          python harness/self_report.py
+          bash scripts/privacy-check.sh --full
 
       - name: Check LESSONS_LEARNED frontmatter
         run: |

README.md

@@ -94,6 +94,14 @@ storyboard_from_article(article, output_dir='output')
 
 # 运行自检
 python harness/self_report.py
+
+# 治理 dry-run（不调用外部素材/TTS 服务）
+python scripts/orchestrator.py \
+  --input examples/example_article.md \
+  --output-dir output/dry-run \
+  --dry-run \
+  --skip-command-checks \
+  --allow-dirty-output
 ```
 
 零外部 API 依赖即可运行语法检查和自检。即梦 CLI 仅在有 AI 素材需求时需要。
@@ -119,6 +127,31 @@ LESSONS_LEARNED.md 更新 + video-rules.json 规则演化
 ✅
 ```
 
+### Step -1：治理预检和运行证明
+
+正式生成前先跑治理 dry-run，确认本仓入口、规则、CTA 资源和自检链路是自洽的：
+
+```bash
+python scripts/preflight.py \
+  --input examples/example_article.md \
+  --skip-command-checks \
+  --json
+
+python scripts/orchestrator.py \
+  --input examples/example_article.md \
+  --output-dir output/dry-run \
+  --dry-run \
+  --skip-command-checks \
+  --allow-dirty-output
+```
+
+`scripts/orchestrator.py` 当前是治理外壳，不会调用付费或远程素材生成服务。它会写入：
+
+- `.md2video-pipeline.jsonl`：每个治理步骤的结构化日志
+- `output/dry-run/run-manifest.json`：输入 hash、仓库状态、环境版本、关键产物指纹和步骤结果
+
+CI 也会跑这一套 dry-run，防止入口契约、QR registry、导入路由或 self-report no-write 行为漂移。
+
 ### Step 0：分镜拆解
 
 **规则驱动，无需改代码。** 将文章输入 `storyboard_ai.py`，自动输出：

SKILL.md

@@ -129,6 +129,31 @@ harness.run("output/final_with_cta.mp4")
 
 完整 pipeline 示例见 `examples/example_pipeline.py`。
 
+### 3b. 治理 dry-run（推荐先跑）
+
+在调用真实 TTS、即梦或完整视频拼接前，先运行治理入口，确认规则、CTA、
+导入路由和 self-report no-write 行为没有漂移：
+
+```bash
+python scripts/orchestrator.py \
+  --input examples/example_article.md \
+  --output-dir output/dry-run \
+  --dry-run \
+  --skip-command-checks \
+  --allow-dirty-output
+```
+
+这个入口不调用付费或远程素材服务。它会生成：
+
+- `.md2video-pipeline.jsonl`：结构化步骤日志
+- `output/dry-run/run-manifest.json`：输入 hash、仓库状态、环境版本、关键产物指纹和步骤结果
+
+只需要预检时可直接运行：
+
+```bash
+python scripts/preflight.py --input examples/example_article.md --skip-command-checks --json
+```
+
 ## Pipeline 详解
 
 ### 数据流

docs/tasks/2026-06-08-pipeline-orchestrator-governance.md

@@ -0,0 +1,83 @@
+# Task Card: Pipeline Orchestrator and Preflight Governance
+
+Status: implemented
+Created: 2026-06-08
+Implemented: 2026-06-08
+
+## Objective
+
+Adopt the useful governance pattern from `md2wechat`: every pipeline run should
+have a stable preflight envelope, structured JSONL execution log, run manifest,
+no-write self-report validation, CI dry-run coverage, and a privacy gate.
+
+## Boundary
+
+In scope:
+
+- Add `scripts/preflight.py` for machine-readable pipeline readiness checks.
+- Add `scripts/orchestrator.py` as the governed dry-run entrypoint.
+- Write `.md2video-pipeline.jsonl` and `output/run-manifest.json` style proof.
+- Add `SelfReport.run(no_write=True)` and CLI `--no-write --json`.
+- Add unit tests for the governance contracts.
+- Add privacy scanning and CI coverage.
+
+Out of scope:
+
+- Paid or external video/material generation.
+- Replacing `examples/example_pipeline.py` with a full production runner.
+- Live E2E rendering with TTS or remote services.
+- Automatic self-evolution rule generation beyond the existing `SelfReport`
+  behavior.
+
+## Implementation Notes
+
+- `scripts/preflight.py` checks:
+  - required command availability (`ffmpeg`, `ffprobe`) unless skipped
+  - governance JSON parseability
+  - storyboard animation routing
+  - CTA registry consistency
+  - TTS-sensitive input characters
+  - stale output artifacts
+- `scripts/orchestrator.py` runs the governance chain:
+  - preflight
+  - CTA registry verification
+  - import/routing smoke
+  - self-report no-write validation
+  - run manifest write
+- `harness/self_report.py --no-write --json` is safe for CI and local contract
+  checks because it does not write `LESSONS_LEARNED.md`, `video-rules.json`, or
+  `output/self_report.json`.
+- `scripts/privacy-check.sh --full` is wired into CI to block common secret
+  patterns and local user paths in tracked text.
+
+## Validation
+
+Run:
+
+```bash
+python -m unittest discover -s tests -p 'test_*.py'
+python scripts/preflight.py --input examples/example_article.md --skip-command-checks --json
+python scripts/orchestrator.py --input examples/example_article.md --output-dir /tmp/md2video-dry-run-output --log /tmp/md2video-pipeline.jsonl --dry-run --skip-command-checks --allow-dirty-output
+python harness/self_report.py --no-write --json
+bash scripts/privacy-check.sh --full
+python -m py_compile $(git ls-files '*.py')
+python scripts/smoke_imports.py
+```
+
+Expected:
+
+- unit tests pass
+- preflight JSON reports `ok: true`
+- orchestrator exits `0` and writes a JSONL log plus `run-manifest.json`
+- self-report no-write exits `0` without mutating governance files
+- privacy gate has no blocking findings
+- syntax and import smoke checks pass
+
+## Residual Risks
+
+- The orchestrator is currently a governance dry-run wrapper, not a full
+  article-to-video runner.
+- Full paid-service E2E remains outside this task and should stay explicit.
+- Future self-evolution hardening should add observation-layer generated checks,
+  companion tests, audit records, and rollback snapshots before any generated
+  rule can be promoted.

harness/self_report.py

@@ -151,7 +151,7 @@ def _generate_rule_id(self, category: str) -> str:
         safe_cat = category.lower().replace(" ", "_").replace("/", "_").replace("-", "_")
         return mapping.get(category, f"auto_{safe_cat}")
 
-    def auto_encode(self):
+    def auto_encode(self, write: bool = True):
         """
         自动将未编码的摩擦点加入 video-rules.json
 
@@ -203,7 +203,8 @@ def auto_encode(self):
                 "last_evolution": datetime.now().isoformat(),
                 "evolution_count": self.rules.get("autopoiesis", {}).get("evolution_count", 0) + new_rules_count,
             }
-            self._save_json(self.rules_path, self.rules)
+            if write:
+                self._save_json(self.rules_path, self.rules)
 
         return new_rules_count
 
@@ -405,21 +406,23 @@ def print_report(self):
 
         print("\n" + "=" * 60)
 
-    def run(self) -> Tuple[Path, Dict]:
+    def run(self, no_write: bool = False, print_human: bool = True) -> Tuple[Optional[Path], Dict]:
         """
         完整自检流程：
         1. 加载系统状态
         2. 自动编码摩擦点
-        3. 写入活记忆
-        4. 生成并保存报告
-        5. 打印报告
+        3. 写入活记忆（no_write=False 时）
+        4. 生成并保存报告（no_write=False 时）
+        5. 打印报告（print_human=True 时）
         """
         self.load_system_state()
-        self.auto_encode()
-        self.write_lessons()
+        self.auto_encode(write=not no_write)
+        if not no_write:
+            self.write_lessons()
         self.generate_report()
-        report_path = self.save_report()
-        self.print_report()
+        report_path = None if no_write else self.save_report()
+        if print_human:
+            self.print_report()
         return report_path, self.report
 
 
@@ -431,14 +434,19 @@ def main():
     parser.add_argument("--project-dir", default=".", help="项目根目录")
     parser.add_argument("--capture", nargs=3, metavar=("CATEGORY", "DESC", "RESOLUTION"),
                         help="捕获一个摩擦点：--capture '素材遗漏' 's22缺失' '补充生成'")
+    parser.add_argument("--no-write", action="store_true",
+                        help="只生成内存报告，不写 LESSONS_LEARNED.md、video-rules.json 或 output/self_report.json")
+    parser.add_argument("--json", action="store_true", help="输出 machine-readable JSON")
     args = parser.parse_args()
 
     report = SelfReport(project_dir=args.project_dir)
 
     if args.capture:
         report.capture_friction(args.capture[0], args.capture[1], args.capture[2])
 
-    report.run()
+    _, data = report.run(no_write=args.no_write, print_human=not args.json)
+    if args.json:
+        print(json.dumps(data, ensure_ascii=False, indent=2))
 
 
 if __name__ == "__main__":