fix-engine

blueif16's avatarfrom blueif16

从验收标准创建检查清单，编排executor→validator循环并带重试逻辑。当子协调器开始实现或收到验证失败时使用。

⭐0stars🔀0forks📁View on GitHub🕐Updated Dec 31, 2025
[Testing and QA]

When & Why to Use This Skill

The Fix-Engine is a sophisticated Claude skill designed to automate the software development lifecycle's implementation and validation phases. It bridges the gap between requirements and code by transforming PRD acceptance criteria into interactive markdown checklists. By orchestrating a robust 'executor-validator' loop with intelligent retry logic, it ensures that every code change is verified against predefined standards. This skill is essential for developers and AI agents seeking to automate debugging, progress tracking, and version control commits within a structured, self-healing development environment.

Use Cases

  • Requirement-to-Checklist Automation: Automatically parsing mini-PRDs to create markdown-based checklists for systematic feature verification and progress tracking.
  • Self-Correcting Development Loops: Managing an autonomous cycle of code implementation and validation that retries on failure, accumulating context to resolve complex bugs.
  • Automated Quality Gatekeeping: Ensuring that code changes are only committed to the repository after passing all validator checks, maintaining high code quality standards.
  • Detailed Error Reporting and Escalation: Generating structured YAML error reports when maximum retries are reached, providing developers with clear failure context and tested methods.
namefix-engine
description从验收标准创建检查清单，编排executor→validator循环并带重试逻辑。当子协调器开始实现或收到验证失败时使用。

修复引擎

所有者: 仅子协调器

职责

  1. 读取 section_path 中的 mini-prd.md
  2. 将验收标准转换为 checklist.md（markdown 复选框格式）
  3. 编排：executor → validator → (失败则重试)
  4. 每次验证后更新 checklist.md 的复选框状态
  5. 更新 meta.yaml 中的进度摘要
  6. 跟踪尝试次数并累积失败上下文
  7. 超过最大重试次数时准备详细错误报告

检查清单创建

从 mini-prd.md 的验收标准生成 checklist.md（markdown 格式）：

## 认证模块检查清单

- [ ] 用户可以使用邮箱/密码注册
- [ ] 登录时签发 JWT 令牌
- [ ] 受保护路由拒绝无效令牌
- [ ] 刷新令牌轮换正常工作

**进度: 0/4**

每次 validator 返回结果后，更新对应的复选框：

  • 通过的检查项：- [x]
  • 失败的检查项：- [ ]

同时更新底部的进度摘要。

执行循环

attempt = 0
max_attempts = 3

while attempt < max_attempts:
    if attempt == 0:
        # 使用 executor agent 实现 .task/mini-prd.md
        call executor agent to implement .task/mini-prd.md
    else:
        call executor(task: "fix", failures: last_failures, previous_attempts: history)

    # 使用 validator agent 检查 .task/checklist.md
    result = call validator agent to check against .task/checklist.md

    # 更新 checklist.md
    update_checklist(result.checkbox_states)

    # 更新 meta.yaml 进度
    progress = f"{result.passed_count}/{result.total_count}"
    update_meta_yaml(section_id, progress: progress)

    if result.all_passed:
        # 完成协议：提交所有更改
        git add -A
        git commit -m "${section_name}: ${summary}

        - ${change_1}
        - ${change_2}

        🤖 Generated with InfiStack"

        # 更新状态文件（最后一步）
        update_meta_yaml(section_id, status: "completed")
        return SUCCESS

    history.append({
        attempt: attempt,
        failures: result.failures,
        executor_notes: executor.notes,
        code_snippets: result.relevant_code
    })

    attempt++

return ESCALATE(error_report: history)

完成协议

当所有检查清单项通过时：

  1. 暂存所有更改：git add -A
  2. 提交并附带结构化消息：
git commit -m "${section_name}: ${summary}

- ${change_1}
- ${change_2}

🤖 Generated with InfiStack"
  1. 写入 COMPLETE 到状态文件
  2. 只有在此之后才发出完成信号

错误报告格式

error_report:
  section_id: "auth-module"
  attempts: 3
  final_failures:
    - check_id: check-2
      expected: "400响应"
      actual: "500响应"
      code_snippet: |
        // src/auth/validate.ts:45-62
        function validateEmail(email) { ... }
      tested_methods:
        - validateEmail()
        - UserService.create()
        - AuthController.register()
  history:
    - attempt: 1
      approach: "初始实现"
      failures: [...]
    - attempt: 2
      approach: "添加try-catch"
      failures: [...]
    - attempt: 3
      approach: "重构验证"
      failures: [...]

将error_report交给escalation-router。