AGENTS.md is a plain-text file at the root of a repository that gives AI coding agents the instructions they need to work in it: how the project is structured, what conventions to follow, what to run before committing, and where the team's deeper context lives. Where a README is written for humans evaluating or installing a project, AGENTS.md is written for the agent that is about to modify it.
The convention exists because agents, like new teammates, do their best work when they are briefed first. A good AGENTS.md keeps the always-true, repo-wide rules inline and points to richer, code-anchored context for the specifics of each area — so the agent loads the relevant decisions and invariants for the files it is about to touch, rather than guessing from the code alone.
Treated well, AGENTS.md is the entry point to a team's repo memory: it is where an agent discovers that a context layer exists, how to pull from it before editing, and how to persist what it learns afterward. It informs the agent; it does not constrain the human.
Sources: agents.md ↗