CLAUDE.md

CLAUDE.md is a Markdown file that holds the standing context Claude Code should know before you ask it to do anything. You place one at your project root for project-specific rules, and you can keep another in your home directory (~/.claude/CLAUDE.md) for conventions you want on every project. At the start of each session Claude reads these files, so the things you would otherwise re-explain in every prompt - your stack, your folder layout, your naming conventions, the libraries you prefer, the things you never want it to do - are already loaded.

A good CLAUDE.md is specific and operational. Name the framework and version, point to where the important code lives, state the commands for running and testing the app, and list the hard rules ('never edit files in /generated', 'always use the existing logger, not console.log'). Think of it as onboarding a sharp new developer: you would not make them guess your conventions, you would write them down. The payoff compounds, because every future session starts already aligned instead of drifting until you correct it.

Two things to avoid. First, secrets: never put API keys or credentials in CLAUDE.md, because it is read into context on every session and usually committed to the repo. Second, bloat: the file costs context-window tokens every time it loads, so a tight, high-signal file beats an exhaustive one. Keep it to the rules and context that actually change Claude's behavior, prune it as the project evolves, and let more situational instructions live in slash commands or the prompt itself.