I’m reading your posts about your experimentation with Claude Code and I have a few suggestions :

• how do people use it with large project ? by setting it up properly, claude.md at the user level and repo level is bare minimum (these are two very different claude.md with different purpose)
• Use Obra Superpower to plan/execute dev tasks (literraly change the way claude behave)
• Do not “chat” with it, investigate code base, plan, execute and carefully augment the claude.md of the repo with carefully chosen information you want it to load in its context each time you launch claude in that repo
• your repo claude.md is not for humans or for documentation, its for claude; do not let it grow indefinitely, you’ll poison your context window
• If you really need to give specific documentation know-how to claude, build a separate repo only for documentation, generate documentation from code base and then launch agents to fact check it, configure your claude with a hook to update documentation or claude.md when necessary after each dev phases
• If you have to repeat more than 2 times an information of context about a code repo to claude, this information should be stored in claude.md

There is so much more to say but…perhaps I should write this down on a blog post :)