Configuration

• Claude Code token & thinking limits — The env vars that control thinking budget, output length, and how much tool output enters the context — which models they apply to, when to set them, per-session vs. everywhere, and recommended numbers.
• Give an agent standing instructions — The file you drop in a repository to tell a coding agent how to behave — CLAUDE.md, AGENTS.md, and the per-tool equivalents — and how to write one that helps instead of bloats.
• Your first CLAUDE.md — A short starter CLAUDE.md you can copy into your first project, exactly where the file goes on macOS, Linux, Windows, and FreeBSD, and how to decide what earns a line in it.

← All tags