It’s a constantly updated collection of the best practices:
- Skills & subagents (with clear agent architecture breakdown)
- MCP servers actually worth using (Context7, Playwright, DeepWiki, etc.)
- Proven workflows (Boris Cherny, Karpathy, RPI)
- Hidden slash commands & settings not obvious in docs
- Real usage tips, including insights from the creator of Claude Code
Plus real experience notes like:
- Keep CLAUDE. md under 150 lines
- Use manual /compact at ~50%
- Use iTerm instead of IDE terminal to avoid crashes
The best part:
You can set up a small team of custom AI agents in one evening - properly structured, connected to the right MCP servers, and tailored to your workflow (refactoring, testing, docs, research, etc.).
Link to the GitHub repo: https://lnkd.in/dxzZb956
♻️ Save this and share with others to help them level up!