guide

AGENTS.md, explained

AGENTS.md is a plain markdown file at the root of a repository that tells AI coding agents what the project is and how to work inside it. Claude Code, Cursor, Codex, Copilot and several other tools look for it before they read your code, the same way a new engineer reads the README on their first morning. It is not code and it is not configuration: it is context, written in sentences, and every agent that supports it will treat it as standing instructions for the whole session.

The reason it matters is scope. An agent asked to fix a date format has no idea that the same file also renders invoices, so it happily refactors both. A good AGENTS.md closes that gap: it names the features that already work, the files each one owns, and the tests that would fail if a behaviour moved. With those three sentences in front of it, an agent stops treating your codebase as a blank page and starts treating it as someone's work. In practice this is the cheapest guardrail available — a text file that travels with the repo, needs no account, and works across tools.

Most teams get it wrong in the same two ways: the file grows into a second README that nobody updates, or it is written once at the start of a project and contradicts the code six weeks later. Stet exists to fix the second problem. The CLI writes AGENTS.md from the charters you record — locked behaviour, owned files, proving tests — and regenerates it when those charters change, so the instructions an agent reads today match the code it is about to edit.