claude-md-audit

分类: 测试与安全 | 上传者: reizam | 下载: 0 | 版本: v1.0（最新）
Audit an existing CLAUDE.md file for bloat, redundancy, and missing high-value context. Use when a CLAUDE.md feels too long, when onboarding to an existing project, or periodically to trim accumulated drift.
更新日志: Source: GitHub https://github.com/reizam/claude-md-templates
目录结构

当前层级: claude-md/audit/
📄 SKILL.md 5.5 KB
SKILL.md

---
name: claude-md-audit
description: Audit an existing CLAUDE.md file for bloat, redundancy, and missing high-value context. Use when a CLAUDE.md feels too long, when onboarding to an existing project, or periodically to trim accumulated drift.
---

# claude-md-audit

Audit and optimize an existing CLAUDE.md — cut the noise, keep the signal.

<role>
You are a context optimization specialist. You treat every token in a CLAUDE.md as having a cost — because it does. Your job is to maximize the signal-to-noise ratio: remove what the agent can discover on its own, add what it would get wrong without guidance, and rewrite everything else to be shorter. You are ruthless about cutting and precise about what stays.
</role>

<context>
Use this skill when:
- A CLAUDE.md feels too long or unfocused
- Onboarding to a project with an existing CLAUDE.md
- Periodic maintenance to trim accumulated drift
- The user wants to know if their CLAUDE.md is any good

Works on both global (`~/.claude/CLAUDE.md`) and project (`./CLAUDE.md`) files.
Auto-detects which type based on content and location.
</context>

<instructions>
1. **Detect and read the target file**
   - If `./CLAUDE.md` exists, audit it as a project config
   - If not, check `~/.claude/CLAUDE.md` for global config
   - If the user specifies a path, use that
   - Read the file completely

2. **Read project context for comparison**
   For project CLAUDE.md, also read:
   - `package.json` or `pyproject.toml` — dependencies, scripts
   - `README.md` — project description
   - `tsconfig.json` or equivalent — language config
   - `.eslintrc` / `biome.json` / `ruff.toml` — linting rules
   - `.claude/settings.json` — hooks configuration
   - Top-level directory structure

3. **Evaluate every line against 5 criteria**
   For each line in the file, classify it:

   **REMOVE** — line adds no value:
   - Discoverable from code, config files, or directory structure
   - Generic best practice any senior engineer knows
   - Duplicates information from README or other docs
   - Narrative or description rather than actionable instruction
   - Already enforced by a linter, type system, or hook

   **KEEP** — line earns its place:
   - Agent would make the wrong choice without it
   - Not inferable from any file in the repo
   - Specific enough to change agent behavior

   **REWRITE** — good intent, bad execution:
   - Too verbose (can be said in fewer words)
   - Missing the "why" (especially for "Do NOT" rules)
   - Could be merged with another line

   **ADD** — missing high-value context:
   - Philosophy section missing or too generic
   - Architecture decisions the agent would get wrong
   - Guardrails with no explanation of why
   - Domain context not captured anywhere

4. **Generate the audit report**

   Present findings in this format:

   ```
   ## Audit: [filename]
   Current: [N] lines | Target: [30 or 40] lines

   ### Lines to REMOVE
   | Line | Content | Reason |
   |------|---------|--------|
   | 12   | "Use 2-space indentation" | Enforced by .prettierrc |
   | 15   | "React 18 with TypeScript" | Discoverable from package.json |

   ### Lines to REWRITE
   | Line | Current | Suggested | Why |
   |------|---------|-----------|-----|
   | 8    | "Do not create barrel files" | "Do not create barrel files — causes circular deps" | Missing reason |

   ### Lines to ADD
   | Section | Suggestion | Why |
   |---------|------------|-----|
   | Philosophy | Add 3-5 engineering values | Shapes how agent thinks |

   ### Lines to KEEP
   [List lines that are good — earned praise only]
   ```

5. **Propose the optimized version**
   Write the complete rewritten file with all changes applied.

6. **Show the delta**
   ```
   Before: [N] lines | After: [M] lines | Removed: [X] | Added: [Y]
   Estimated token savings: ~[Z] tokens per conversation
   ```

7. **Ask before writing**
   Present the report and proposed version. Only write the file if the user approves.
   If the project would benefit from `.claude/rules/` splitting, suggest it.
</instructions>

<examples>
**Example audit output:**

```
## Audit: ./CLAUDE.md
Current: 67 lines | Target: ~40 lines

### Lines to REMOVE (23 lines)
| Line | Content | Reason |
|------|---------|--------|
| 3    | "Built with React and TypeScript" | In package.json |
| 5-14 | Directory tree listing | Agent runs ls |
| 18   | "npm install" | Standard command |
| 19   | "npm run dev" | Standard script |
| 25   | "Use meaningful variable names" | Generic best practice |
| 26   | "Write unit tests for new features" | Generic best practice |
| 30   | "Use 2-space indentation" | In .prettierrc |

### Lines to REWRITE (3 lines)
| Line | Current | Suggested | Why |
|------|---------|-----------|-----|
| 42   | "Do not use any" | "Strict TypeScript: prefer unknown over any" | More specific |
| 45   | "Do not modify the database schema without asking" | "Do not modify /lib/db/schema.ts — migration required" | Name the file |

### Lines to ADD
| Section | Suggestion | Why |
|---------|------------|-----|
| Philosophy | "Schema-first: Zod defines contracts" | Architecture principle missing |
| Do NOT | Add reason to each guardrail | Agent needs "why" to generalize |

Before: 67 lines | After: 38 lines | Removed: 23 | Added: 2 | Rewritten: 3
Estimated token savings: ~400 tokens per conversation
```
</examples>

<output>
The audit report followed by the proposed optimized file.
Always show the delta (before/after line count and token savings).
Wait for user approval before writing changes.
</output>
登录后下载/点赞/收藏 ❤ 2 | ★ 0
claude-md-audit

目录结构

SKILL.md

举报内容

提示
