claude-md-architect（Claude MD 架构师）

CLAUDE.md Architect Skill

Generate and optimize CLAUDE.md files for software projects by analyzing codebase context and applying Anthropic engineering best practices. Creates concise, maintainable project instructions that maximize Claude Code effectiveness while minimizing token consumption.

Core Principles

1. Conciseness over Completeness: Treat CLAUDE.md like frequently-used prompts, not exhaustive documentation
2. Evidence-Based Context: Analyze actual codebase before making recommendations
3. Incremental Refinement: Start minimal, iterate based on effectiveness
4. Token Efficiency: Every line should earn its presence in the context window
5. Human-Readable Structure: Use clear organization with XML tags for parsing

When to Use This Skill

Activate when user requests:

• "init CLAUDE.md" or "create project instructions"
• "optimize" or "improve" existing CLAUDE.md
• "setup Claude for my project"
• Project-specific instructions for Claude Code
• Starting work on a new project that needs documentation
• Existing CLAUDE.md is outdated or ineffective

Quick Decision Tree

User Request
    │
    ├─→ "init" or "create" → Initialize New CLAUDE.md
    │   └─→ Follow: @refs/initialization-workflow.md
    │
    ├─→ "optimize" or "improve" → Optimize Existing CLAUDE.md
    │   └─→ Follow: @refs/optimization-patterns.md
    │
    └─→ "integrate" or "setup MCP/slash commands" → Integration
        └─→ Follow: @refs/integration-strategies.md

Workflow Overview

Initialize New CLAUDE.md

Quick Steps:

1. Discover - Analyze codebase (language, framework, structure, tests)
2. Extract - Identify code style, commands, patterns
3. Generate - Select template, customize for project
4. Validate - Verify commands, check token count
5. Present - Show output with explanation

Detailed Instructions: @refs/initialization-workflow.md

Optimize Existing CLAUDE.md

Quick Steps:

1. Analyze - Evaluate token efficiency, accuracy, relevance
2. Identify - Find redundancy, outdated info, generic fluff
3. Refactor - Apply token reduction, restructure, improve clarity
4. Test - Verify 40%+ token reduction, retain critical info
5. Present - Show before/after with metrics

Detailed Instructions: @refs/optimization-patterns.md

Token Reduction Quick Reference

Before (Verbose)

When you are implementing new features in this codebase, it's very important that you always make sure to write comprehensive tests for all the functionality you add. We use Jest as our testing framework, and we expect all new code to have at least 80% code coverage.

After (Concise)

### Testing Requirements
- New features require tests (Jest, >80% coverage)
- Unit tests: Individual functions in `src/lib/`
- Integration tests: API endpoints with Supertest

Result: 70% token reduction, same information

Template Selection Guide

Quality Checklist

Before presenting CLAUDE.md:

• ✓ File is <400 lines (prefer <250)
• ✓ Commands verified against package.json/Makefile
• ✓ Patterns match actual codebase conventions
• ✓ No redundant info already in README
• ✓ Context loading is selective, not exhaustive
• ✓ Structure uses clear sections with headers
• ✓ Examples are concrete, not generic
• ✓ Security/testing requirements reflect actual needs

Token Budget Guidelines

If approaching max, split into:

• .claude/CLAUDE.md (essentials)
• .claude/ARCHITECTURE.md (reference, not auto-loaded)
• .claude/commands/*.md (workflows as slash commands)

Content Priority

Keep (High Value)

• Project-specific commands and scripts
• Code style conventions unique to this project
• File organization patterns
• Testing requirements and strategies
• Security constraints
• Performance considerations

Remove (Low Value)

• Generic software engineering advice
• Information already in README
• Obvious best practices (DRY, SOLID, etc.)
• Detailed framework explanations (use context7)
• Step-by-step tutorials (link to docs)

Context Loading Strategy

Use @-syntax ONLY for:

• Small, universally relevant files (<100 lines)
• Configuration affecting all changes (tsconfig,.eslintrc)
• Core type definitions used everywhere

Use On-Demand Loading:

When working on authentication:
  @src/lib/auth.ts
  @src/middleware/authenticate.ts

Avoid:

• Loading entire directories with @src/**
• Adding large files that aren't always needed
• Injecting documentation that duplicates official sources

Integration with Global CLAUDE.md

User has ~/.claude/CLAUDE.md (global):

• Universal principles (SOLID, DRY, YAGNI)
• MCP tool documentation
• Preferred communication style

Project CLAUDE.md should:

• Focus exclusively on project-specific guidance
• Avoid duplicating MCP tool usage (already global)
• Add project-specific MCP tool applications only

Example header:

# MyProject - Development Guide

> Note: General software engineering principles are in your global CLAUDE.md.
> This guide focuses on project-specific patterns and requirements.

MCP Tool Configuration

Detect opportunities:

• React/Vue/Angular imports → Suggest context7 for official docs
• UI component requests → Suggest magic MCP for patterns
• E2E test files → Suggest Playwright MCP for browser automation

Add to CLAUDE.md:

## MCP Tools Configuration

### context7 (Official Documentation)
Use for React hooks, Next.js routing, Prisma schema

### magic (UI Component Generation)
Use for new components, accessibility improvements

### Playwright (E2E Testing)
Use for user flows, visual regression, accessibility audits

Detailed Integration Guide: @refs/integration-strategies.md

Output Format

When Generating New CLAUDE.md

1. Display generated content in code block
2. Explain customizations based on analysis
3. Highlight key decisions
4. Suggest optional additions
5. Provide save location and next steps

When Optimizing Existing CLAUDE.md

1. Summarize issues found
2. Show before/after comparison for key sections
3. Present optimized full version
4. Quantify improvements (token savings %)
5. Suggest iteration approach

Detailed Templates: @refs/output-templates.md

Common Mistakes to Avoid

Success Metrics

Effective CLAUDE.md

• Claude requires fewer clarifying questions
• Code matches project conventions on first try
• Commands work without errors
• User rarely provides missing context manually

Ineffective CLAUDE.md

• Claude frequently asks "what framework?"
• Generated code doesn't match existing style
• User repeatedly provides same context files
• CLAUDE.md contains info never referenced

Philosophy

Remember:

• CLAUDE.md is a living document, not permanent documentation
• Start minimal, add based on actual friction points
• Remove guidance that doesn't improve Claude's output
• Measure effectiveness: fewer questions = better CLAUDE.md

When in doubt:

• Prefer concise over comprehensive
• Prefer specific over generic
• Prefer examples over explanations
• Prefer on-demand over auto-loading

Reference Files

• @refs/initialization-workflow.md - Complete guide for creating new CLAUDE.md files from scratch
• @refs/optimization-patterns.md - Techniques for reducing tokens and improving effectiveness
• @refs/integration-strategies.md - Integration with global config, MCP tools, slash commands
• @refs/output-templates.md - Standard formats for presenting results to users

Quick Start: When user requests CLAUDE.md creation/optimization, follow the decision tree above and consult relevant reference files for detailed instructions.