tukuaiai d80122e9da chore: update prompts-library structure 1 mese fa
..
artifacts d80122e9da chore: update prompts-library structure 1 mese fa
state d80122e9da chore: update prompts-library structure 1 mese fa
README.md d80122e9da chore: update prompts-library structure 1 mese fa
hook_runner.sh d80122e9da chore: update prompts-library structure 1 mese fa
runner.py d80122e9da chore: update prompts-library structure 1 mese fa

README.md

workflow_engine

全自动开发闭环的轻量编排引擎,基于 文件事件 Hook + 状态机 实现。

目录结构

workflow_engine/
├── runner.py           # 状态机调度器
├── hook_runner.sh      # 文件监听 Hook (inotify)
├── state/
│   └── current_step.json   # 当前状态
└── artifacts/
    └── <run_id>/       # 每次运行的产物
        ├── step1.json
        ├── step2.json
        └── ...

快速开始

1. 手动模式(无 Hook)

# 启动新工作流
python runner.py start

# 查看状态
python runner.py status

2. 自动模式(Hook 监听)

# 终端 1: 启动 Hook 监听
./hook_runner.sh

# 终端 2: 启动工作流(状态变更会自动触发后续步骤)
python runner.py start

状态文件 Schema

{
  "run_id": "20251225T053800",
  "step": "step3",
  "status": "running|success|failed|completed|fatal_error",
  "payload_path": "artifacts/20251225T053800/step2.json",
  "verify": {"status": "success|failed", "details": "..."},
  "target_step": "step2|step5|done",
  "retry_count": 0
}

流程控制

step1 → step2 → step3 → step4 → step5
                                  │
                    ┌─────────────┴─────────────┐
                    │                           │
              verify=failed               verify=success
                    │                           │
                    ▼                           ▼
              target_step=step2           target_step=done
              (回跳重规划)                 (流程结束)

熔断机制

  • 同一任务最多重试 3 次
  • 超过后状态变为 fatal_error,需人工介入

TODO

  • 集成实际 LLM 调用(替换 runner.py 中的 MOCK)
  • 添加 CI 集成示例
  • 支持并行任务处理