AIDLC Bug Killer

多Agent协作Bug修复工作流。用于三个Agent协作修bug：SubAgent发现bug，主Agent修复，ReviewAgent确认。当需要协调多个Agent修复项目bug时使用此skill。

Security Scan

The skill is internally consistent with its stated purpose: an instruction-only multi‑agent bug-fixing workflow that reads and writes files under aidlc-docs/bug-reports and updates a changelog; it requests no external credentials or installs.

Purpose & Capability

Name and description match the runtime instructions: the skill coordinates SubAgent/MainAgent/ReviewAgent to create, update, and confirm bug report files inside aidlc-docs/bug-reports and requires no unrelated binaries, env vars, or external services.

Instruction Scope

Instructions are scoped to repository paths (pending/, waiting_confirm/, confirmed/) and a changelog file (aidlc-docs/aidlc-state.md). However, the guidance 'analyze and fix code' and 'run tests' is broad and leaves agent behaviour unspecified (how tests are run, what commands are executed, what other repo paths may be touched). That vagueness can lead to agents executing untrusted code or modifying files beyond the bug-report directory unless you limit agent permissions and enforce review policies.

Install Mechanism

No install spec and no code files — this instruction-only skill does not write binaries or fetch external code.

Credentials

No environment variables, credentials, or external endpoints are requested. The only implicit requirement is filesystem access to the project's aidlc-docs/ directory.

Persistence & Privilege

The skill will instruct agents to create, update, move, and never delete bug files and to mandate changelog updates. always:false (normal). Because agents can act autonomously by default, installing the skill enables automated repo file modifications unless you limit agent invocation or permissions — consider requiring human approval before merges or commits to protected branches.

Assessment

This skill appears to do what it says: coordinate agents that read/write bug-report files in your repository and run tests. Before installing, verify the following: 1) Only grant the skill/agents the minimal filesystem and repo permissions they need (restrict to a sandboxed repo or specific paths). 2) Ensure test execution happens in a sandboxed environment (CI runner or container) to avoid running untrusted code on sensitive hosts. 3) Require human review or protected-branch rules before automated commits/merges — the skill expects agents to update changelog and commit changes, so protect your main branches. 4) Back up the repo or ensure an audit trail (commits, signatures) so you can revert unwanted changes. 5) If you need stricter guarantees, ask the skill author for explicit lists of commands the agents will run and for safeguards around test execution and file access. If you cannot enforce these controls, treat automated fixes as high risk and require manual approval.

AIDLC Bug Killer - 三Agent协作Bug修复系统

目录结构

card-tower/
└── aidlc-docs/
    └── bug-reports/
        ├── pending/           # 待处理bug (SubAgent写入)
        ├── waiting_confirm/   # 修复中/待确认 (主Agent写入)
        └── confirmed/         # 已确认完成 (ReviewAgent确认)

角色分工

角色	职责	写入目录
SubAgent	分析代码，发现bug，生成报告	pending/
主Agent	读取bug，修复代码，更新状态	waiting_confirm/
ReviewAgent	验证修复，确认通过	confirmed/

工作流

1. SubAgent发现Bug

创建文件：aidlc-docs/bug-reports/pending/B-XXX-title.md

使用模板：references/bug-template.md

2. 主Agent修复Bug

1. 读取 pending/ 下的bug文件
2. 分析并修复代码
3. 更新bug文件状态
4. 移动到 waiting_confirm/

3. ReviewAgent确认

1. 读取 waiting_confirm/ 下的修复文件
2. 运行测试验证
3. 检查修复是否符合预期
4. 移动到 confirmed/

Changelog 必须更新规则

每次合并代码到 master 后，必须更新 aidlc-docs/aidlc-state.md 的 changelog：

### YYYY-MM-DD
| 时间 | commit | 变更内容 |
|------|--------|----------|
| HH:MM | abc1234 | 描述 |

这是 Sub-Agent 获取主Agent进展的唯一入口。即使只改了一行代码也要写。

关键规则

1. 一个bug一个文件 - 便于并行处理
2. 状态同步 - 文件状态必须与实际一致
3. 不删除bug - 已确认的移入confirmed，不删除
4. 详细记录 - 便于追溯和复盘