Boris Cherny (Claude Code creator) recently dropped a threads on how his team at Anthropic uses Claude Code.

The key insight: they don't treat it as a static config. After every correction, they tell Claude "Update your CLAUDE.md so you don't make that mistake again." Claude writes a rule for itself. They review it, commit it to git. The mistake never happens again.

I cross-referenced his tweets with Anthropic's official docs and other best practices for CLAUDE.md and then packaged it into a starter kit:

  - Fill-in-the-blank templates for Next.js/TypeScript, Python/FastAPI, and a generic
  catch-all
  - The workflow patterns his team actually uses (plan mode, verification loops, subagent
  strategy)
  - Every claim cited back to the source tweet or doc

Repo: https://github.com/abhishekray07/claude-md-templates

What's in your CLAUDE.md that's made a measurable difference?