Creating Effective CLAUDE.md Files

Software Development with Claude Code

Dani Zysman

AI Curriculum Manager @ DataCamp

CLAUDE.md Best Practices

Start with /init, then iterate
- Add context when Claude makes mistakes
- Include convention changes
Goal: A living document that evolves with your project

The Project So Far

music-analytics-api/
  app.py           # Application entry point
  routes.py        # API endpoints
  models.py        # Artist/Album models
  analytics.py     # Statistics functions
  middleware.py    # Request logging
  config.py        # Session settings
  CLAUDE.md        # <- Project memory file

What Goes IN vs. OUT of CLAUDE.md

recraft: What to include vs leave out of CLAUDE.md

Rule: If it's longer than a sticky note, it doesn't belong.

Iterating on CLAUDE.md

> Add error handling to the /artists endpoint.

Done. I've added error handling to /artists (routes.py:9-19):
- Try-except block wrapping retrieval and serialization
- AttributeError handler if the data store isn't initialized
- General exception handler returning error details with 500 status

return jsonify({'error': f'Failed to retrieve artists: {str(e)}'}), 500

> Don't expose exception details to users, that's a security risk. Log internally 
and return a generic message.  Update CLAUDE.md.

Updated /artists to log errors internally and return a 
generic 500 message. CLAUDE.md updated with the convention: 
never expose exception details in responses.

Good vs. Bloated CLAUDE.md

Bloated (200+ lines)

# Inventory API
This is an Inventory Management 
API built with Flask...

## Full Source Code
### routes.py
from flask import jsonify...
[entire file pasted]

### models.py
[entire file pasted]

## API Documentation
GET /items returns...

Good (15 lines)

# Inventory API
Flask REST API for inventory.

## Commands
- Run: flask run --port 5001
- Test: pytest tests/

## Conventions
- Use type hints
- Keep routes thin

File References - Using @

> @routes.py what endpoints does this API have?

API Endpoints (6 GET routes)

Artists
- GET /artists              All artists
- GET /artists/<id>         Single artist by ID
- GET /artists/<id>/albums  Albums for an artist
Albums
- GET /albums               All albums
- GET /stats/albums         Album popularity statistics
- GET /stats/average-tracks Average track count per album

All endpoints return JSON.

CLAUDE.md + @ Together

# CLAUDE.md - Music Analytics API
A Flask-based REST API for music data analytics, providing endpoints for artists,
albums, and statistics.
...


## Key Files
@routes.py - API endpoint definition
@models.py - Artist and Album data models with in-memory store
...

CLAUDE.md gives persistent context. Use @ so Claude always knows your key files.

Three Levels of Context

Three levels of CLAUDE.md context

When to Use Each

CLAUDE.md (project, shared)
- Test commands, coding conventions, project structure
CLAUDE.local.md (project, personal)
- Local paths, API keys, personal workflow preferences
~/.claude/CLAUDE.md (global, personal)
- Style preferences across all projects
- Default behaviors you always want

Chapter 1 Recap

/init andCLAUDE.md → Persistent project context
Shift+Tab or /plan → Toggle Plan mode
think ... ultrathink → Deep reasoning (5 levels)

Let's Practice!

Software Development with Claude Code