Why Project Memory Matters
Without a CLAUDE.md, you'll correct the agent on the same thing every conversation forever.
By default, every Claude Code session starts cold. The agent doesn't know your test runner, your naming conventions, your deploy target, or that secrets live in .env.local and never in code. If you don't tell it, you'll re-tell it on every new conversation, forever.
Project memory fixes this. A file at the root of your repo named CLAUDE.md is read by the agent on every turn. It's not a prompt; it's a permanent context block. Anything written there becomes 'something the agent always knows' for that project.
The right rule of thumb: if you correct the agent on the same thing twice, that correction belongs in CLAUDE.md. The file grows as your project's tacit knowledge gets surfaced. Six months in, your CLAUDE.md is the most underrated artifact in your repo.