AgentSkillsCN

claude-md-optimizer

引导用户进行正念练习、冥想训练以及压力缓解技巧。适用于用户寻求放松、压力管理、呼吸练习或内在平和培育方面的帮助时使用。

SKILL.md
--- frontmatter
name: claude-md-optimizer
description: "Analyzes and optimizes CLAUDE.md files following Anthropic's official best practices. Use when reviewing existing CLAUDE.md for improvements, or when user mentions CLAUDE.md is too long or ineffective."

Optimize CLAUDE.md

Analyze, audit, and optimize CLAUDE.md files based on Anthropic's official best practices for effective Claude Code configuration.

Degrees of Freedom

This is a medium-freedom task. Follow the structured workflow but adapt recommendations based on the specific project's needs, tech stack, and team conventions.

Workflow

Copy this checklist and track your progress:

code
Optimization Progress:
- [ ] Step 1: Analyze current CLAUDE.md state
- [ ] Step 2: Audit against best practices checklist
- [ ] Step 3: Identify and prioritize issues
- [ ] Step 4: Generate recommendations
- [ ] Step 5: Apply or present changes

1. Analyze Current State

Read the existing CLAUDE.md (if present) and assess:

  • •Current structure and organization
  • •Content categories present
  • •Length and verbosity
  • •Presence of anti-patterns

2. Audit Against Best Practices

Compare against the checklist in references/best-practices.md.

Score each category:

  • •Present & Good - Follows best practices
  • •Present but Needs Work - Exists but could be improved
  • •Missing - Should be added
  • •Anti-pattern - Should be removed/fixed

3. Generate Recommendations

Prioritize changes by impact:

High Priority:

  • •Remove sensitive information (API keys, credentials)
  • •Fix anti-patterns (generic advice, excessive length)
  • •Add missing critical sections (project structure, commands)

Medium Priority:

  • •Improve organization and scannability
  • •Add missing helpful sections
  • •Condense verbose explanations

Low Priority:

  • •Formatting improvements
  • •Minor restructuring
  • •Optional enhancements

4. Apply Changes

Either:

  • •Suggest mode: Present recommendations for user approval
  • •Apply mode: Directly edit CLAUDE.md with improvements

Optimization Principles

Keep It Concise

CLAUDE.md is loaded into context every time. Total limit: 40k tokens (CLAUDE.md + all @referenced files combined).

markdown
# BAD: Verbose explanation
## Testing Guidelines
When running tests in this project, you should use the npm test command.
This will execute all test files and provide output. Make sure tests pass.

# GOOD: Concise and actionable
## Testing
npm test                    # Run all tests
npm run test:coverage       # With coverage report

Be Specific, Not Generic

Include YOUR project's actual patterns, not theoretical best practices.

markdown
# BAD: Generic advice
Follow clean code principles and write maintainable code.

# GOOD: Project-specific guidance
Use `logger` from @/utils, never console.log (tree-shaken in production).
Query keys must use queryKeys factory from @/lib/queryKeys.

Use Progressive Disclosure

For large documentation, split into separate files and reference them:

markdown
# In CLAUDE.md
See @.claude/rules/testing.md for testing standards.
See @docs/ARCHITECTURE.md for system design.

# NOT: 500 lines of testing docs inline

Include Actual Commands

Show the commands your team uses:

markdown
## Development
npm run dev     # Start dev server (localhost:3000)
npm test        # Run tests in watch mode
npm run build   # Production build

Use Emphasis for Critical Rules

For must-follow instructions, use emphasis markers:

markdown
IMPORTANT: Never commit directly to main.
YOU MUST run tests before pushing.
ALWAYS use logger utility, never console.log.

Output Format

When optimizing, provide:

  1. •Summary - Brief assessment (1-2 sentences)
  2. •Score - X/10 rating with breakdown
  3. •Critical Issues - Must-fix problems
  4. •Recommendations - Prioritized improvements
  5. •Optimized CLAUDE.md - The improved file (if applying changes)

Example Assessment

code
## CLAUDE.md Optimization Report

**Summary:** Good foundation but overly verbose. Contains generic advice that adds token cost without value.

**Score:** 6/10
- Structure: 8/10 (well-organized sections)
- Conciseness: 4/10 (too verbose, 2000+ lines)
- Specificity: 5/10 (mix of specific and generic)
- Security: 9/10 (no sensitive data found)

**Critical Issues:**
1. File exceeds 500 lines - should use modular rules files
2. Contains generic "best practices" not specific to project

**Recommendations:**
1. [HIGH] Move detailed rules to .claude/rules/*.md files
2. [HIGH] Remove generic coding advice Claude already knows
3. [MED] Add missing quick-start commands section
4. [LOW] Consolidate duplicate information in docs/

Resources

references/