AGENTS.md

AGENTS.md is an open convention for a single markdown file, placed at a repository’s root, that tells an AI coding agent everything it needs to work on the project: what the code is, how to build and test it, the coding standards to follow, and the practices to respect. It emerged as tools converged on the same problem — every assistant wanted project context — and settled on one predictable, human-readable file instead of a proliferation of per-tool dotfiles.

• One file, many tools - A shared format read by OpenAI Codex, Google Jules, Cursor, Amp, and others, so guidance is authored once.
• Plain markdown - Build commands, test commands, and conventions live as ordinary prose an agent and a human can both follow.
• Repo-local and versioned - The file ships with the code, travels through pull requests, and evolves alongside the project.
• Neutral stewardship - Maintained under the Agentic AI Foundation within the Linux Foundation rather than owned by any single vendor.

For API teams, AGENTS.md is fast becoming the place to encode the operating manual an agent needs before it touches a codebase — the governance rules, the parts of the surface that must not drift, and the build/test loop that proves a change is safe. It is the widely-adopted cousin of the singular AGENT.md variant, and it sits in the same agentic stack as Agent2Agent and MCP — the connective tissue that turns raw AI assistance into governed, repeatable work.