CLAUDE.md Documenter

CLAUDE.md Documenter

Create effective CLAUDE.md files that give Claude project context without over-documenting.

What is CLAUDE.md?

Observability

First: At the start of execution, display:

🔧 Skill "julien-dev-claude-md-documenter" activated

A markdown file that Claude automatically loads at the start of every conversation. It provides project context so you don't have to repeat yourself.

Location: ./​CLAUDE.md (project root) or ~/​.claude/​CLAUDE.md (global)

Core Principles

1. Keep It Short

• Target: 50-200 lines
• Maximum: 500 lines
• Why: Content is prepended to EVERY message (costs tokens)

2. Link, Don't Duplicate

❌ Don't:
## API Endpoints
POST /​api/​users - Creates user...
[50 more endpoints]

✅ Do:
## API
See: docs/​api.md or visit /​api-docs when running

3. Project-Specific Only

Include:

• Unique workflows
• Critical gotchas
• Common commands
• File structure

Don't include:

• Code style (use linters)
• General programming knowledge
• Complete API docs (link to them)

Minimal Template

# [Project Name]

[2 sentences: What is this? What problem does it solve?]

## Tech Stack
- [Primary language/​framework]
- [3-5 key dependencies]

Full deps: package.json

## Structure

app/ # Routes components/ # UI lib/ # Utils

## Commands
```bash
npm run dev    # Start dev server
npm test       # Run tests

All: Run npm run or see package.json

Important

• [Project-specific gotcha #1]
• [Project-specific gotcha #2]

References

• Architecture: [link]
• Contributing: CONTRIBUTING.md
• Style: .eslintrc

## Best Practices

**Source:** [Anthropic Engineering - Claude Code Best Practices](https://www.anthropic.com/engineering/claude-code-best-practices)

1. **~150-200 instructions max** - LLMs can follow ~150 instructions reliably
2. **Avoid code style rules** - "Never send an LLM to do a linter's job"
3. **Iterate over time** - Start minimal, add only what Claude misses
4. **Be specific** - "Run `npm test` before commit" > "Test your code"
5. **Provide motivation** - Explain WHY, not just WHAT

## File Locations

~/​.claude/​CLAUDE.md # Global (lowest priority) /​project/​CLAUDE.md # Project root /​project/​packages/​api/​CLAUDE.md # Subdirectory (highest priority)

Closer to working directory = higher priority.

## Common Mistakes

❌ **Too long** - 1000+ lines nobody reads
❌ **Duplicates docs** - Repeats README, API specs
❌ **Code style rules** - That's what linters are for
❌ **General knowledge** - Claude already knows React basics

## Quick Start

1. Create `CLAUDE.md` in project root
2. Start with the minimal template above
3. Fill in only unique project details
4. Link to existing docs (don't duplicate)
5. Keep under 200 lines

## External Resources

**Official:**
- [Claude Code Best Practices](https://www.anthropic.com/engineering/claude-code-best-practices)
- [Using CLAUDE.MD Files](https://www.claude.com/blog/using-claude-md-files)
- [Prompting Best Practices](https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/claude-4-best-practices)

**Community:**
- [Writing a Good CLAUDE.md - HumanLayer](https://www.humanlayer.dev/blog/writing-a-good-claude-md)
- [Complete Guide - Sid Bharath](https://www.siddharthbharath.com/claude-code-the-complete-guide/)

**More:** See [references/​external-resources.md](references/external-resources.md)

## Usage

When user asks to create/​improve a CLAUDE.md:

1. Ask about their project (tech stack, structure, key workflows)
2. Use the minimal template above
3. Fill in only project-specific details
4. Point to existing docs (README, CONTRIBUTING, etc.)
5. Keep it under 200 lines
6. Remind: iterate based on usage

## Skill Chaining

### Skills Required Before
- None (entry point skill)

### Input Expected
- User request to create/​improve CLAUDE.md
- Project context (tech stack, structure, workflows)
- Optional: existing CLAUDE.md to improve

### Output Produced
- **Format**: CLAUDE.md file in project root
- **Side effects**: Creates or modifies CLAUDE.md
- **Duration**: 5-10 minutes

### Compatible Skills After
**Recommandés:**
- **julien-dev-tools-skill-creator-pro**: If creating project-specific skills
- Git workflow: Commit the new CLAUDE.md

### Called By
- Direct user invocation: "Create a CLAUDE.md", "Improve my CLAUDE.md"
- New project setup workflows

### Tools Used
- `Read` (usage: read existing CLAUDE.md, README, package.json)
- `Write` (usage: create/​update CLAUDE.md)
- `Glob` (usage: discover project structure)

### Visual Workflow

User: "Create CLAUDE.md for this project" ↓ [THIS SKILL] ├─► Ask about project (tech, structure) ├─► Use minimal template ├─► Fill project-specific details ├─► Link to existing docs └─► Keep under 200 lines ↓ CLAUDE.md created ↓ [Optional] Git commit

### Usage Example

**Scenario**: Create CLAUDE.md for a React project

**Command**: "Create a CLAUDE.md for this project"

**Result**:
- CLAUDE.md created with ~100 lines
- Tech stack, structure, commands documented
- Links to README, CONTRIBUTING for details

---

**Remember:** The best CLAUDE.md is concise, specific, and links to details instead of duplicating them.