AGENTS.md (format)
Open format (agents.md) standardized under the Linux Foundation umbrella in December 2025: an `AGENTS.md` file at the repo root describing to AI agents (Claude Code, Cursor, Codex) how to navigate code, run tests, structure commits, respect conventions. Complement to DESIGN.md (the visual what), AGENTS.md covers the code-side how.
Strengths
- Single file for all agents, no more duplicated CURSOR.md, CLAUDE.md
- Linux Foundation standard, protects from vendor lock-in (each agent vendor must support it)
- Versioned in repo, new dev OR new agent onboarding identical
Limitations
- Loose format (free Markdown), easy to let rot, little automatic validation
- Does NOT replace a test runner or linter, it's documentation, not enforcement
Best for
- Any repo opening its doors to AI agents, first file to create after the README
- Multi-agent teams (Claude Code + Cursor + Codex in parallel) wanting a common guide