Glossary
What is CLAUDE.md?
Claude Code's project memory file, loaded at the start of every session.
Updated 2026-09-09
CLAUDE.md is the file Claude Code reads when a session starts and then treats as standing instructions for that project.
It plays the same role as AGENTS.md, just scoped to Anthropic's tooling. There are two useful places to put one. A CLAUDE.md checked into the repo is shared with the team and covers project rules. A CLAUDE.md in your own config directory covers personal preferences on your machine and follows you across projects, which is where things like preferred commit style belong.
The practical value is not documentation, it is not repeating yourself. If you have corrected the same behaviour twice in a session, that correction belongs in the file. Deploy runs through Vercel, not the CLI. Never write to the production database. Use the existing Button component instead of a new one.
If you already maintain AGENTS.md, do not fork the content. Keep one file real and have the other reference it, or generate one from the other. Two drifting instruction files is worse than one, because you will not know which one the agent actually followed.
Where people get this wrong
- It loads on every session, so bloat is a tax you pay on every request. Prune it like code.
- Personal preferences do not belong in the repo copy your teammates share. Put those in your user level file.
- Rules the model cannot check are ignored in practice. Rules with a command, a path, or a file name get followed.