Ccdawn simplification audit

Skill CCDawn/codex-skills/skills/engineering/ccdawn-simplification-audit

Chinese-first Codex skill pack for intent alignment, dynamic routing, lightweight development, code review, UI design, and AI research workflows.

Install

npx -y skills add CCDawn/codex-skills --skill ccdawn-simplification-audit

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when the user explicitly requests a repository/subsystem simplification audit, dependency-bloat review, or ranked removal plan, or when concrete evidence makes structural complexity the primary investigation; do not substitute it for ordinary project health review.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.3 KB, as published. Nobody here has run it

CCDawn 精简审计

目标

只读审计仓库或子系统中的可移除复杂度，形成按价值和风险排序的精简队列。它补充项目健康审查，不替代 ccdawn-project-review。

BRT interface

Context Boundary: 仓库/子系统范围、入口与依赖、已知架构约束、生成代码/第三方代码排除范围。
Output Contract: 证据化复杂度 findings、保留项、精简候选队列和后续 owner。
Allowed Action: 只读；不修改文件、不卸载依赖、不移动分支、不改 index。
Success Evidence: 每个候选都有路径/引用/依赖或调用证据，并说明收益、风险和验证要求。
Stop Condition: 审计范围不明、缺少源码/依赖证据、建议涉及行为或架构取舍、或问题主要属于正确性/安全/性能。
Route Out: ccdawn-project-review、ccdawn-performance-engineering、ccdawn-planning、对应开发 owner、ccdawn-brt 或 BLOCKED。

统一调用契约

只处理 BRT interface 范围；不匹配时回 ccdawn-brt 或更具体 owner，复合任务不吞其他 owner。
用户可见内容默认中文，完成只报状态、产出、证据和剩余风险；代码、命令、路径、错误原文、API/协议、skill 名和枚举保留原样；Route Out 仅以 BRT interface 为准，末行写 下一步建议: <一个具体动作>。

审计规则

先定位入口、依赖清单和高变更/高耦合区域，不为“全面”盲扫全仓。
重点检查：重复依赖、标准库/平台已有能力、单实现接口、单产品工厂、只转发的 wrapper、废弃 flag/config、并存的 legacy 路径、重复事实源和跨文件样板。
排除生成代码、vendored/third-party 代码、迁移期间明确保留的兼容层，以及安全、无障碍、数据恢复和真实平台适配约束。
每项区分“确认可删”“需 probe”“架构取舍”；证据不足不能写成确定结论。
正确性、测试、安全和总体健康交 ccdawn-project-review；需实测的性能问题交 ccdawn-performance-engineering；实施方案交 ccdawn-planning。

优先级：P1 高收益、P2 中收益、P3 低收益、需证据。

输出

精简审计:
- 结论: LEAN / OPPORTUNITIES_FOUND / NEEDS_EVIDENCE / BLOCKED
- 范围与排除项: ...

Findings:
- P1/P2/P3/需证据: [位置] 复杂度来源；证据；最小替代；收益；风险。

保留项:
- 必须存在的复杂度及原因。

精简队列:
- 1. ... [SAFE_DIRECT / PLAN_THEN_EXECUTE / DEFERRED / BLOCKED]；验证条件...

组合路由:
- 项目健康/正确性/安全 -> ccdawn-project-review
- 性能测量 -> ccdawn-performance-engineering
- 多项结构调整 -> ccdawn-planning

下一步建议: <一个具体动作>

只在可复现计数后报告精确净减行数或依赖数，不拿外部 benchmark 冒充当前仓库收益。

本 skill 的复杂度视角借鉴 Ponytail（MIT），并按 CCDawn 的中文输出、证据和 owner 契约重新组织。