AGENTS.md

One file that tells every coding agent how your repo works

AGENTS.md is plain Markdown at the root of your repository. Coding agents read it before they touch your code: how to build and test, which conventions matter and what to leave alone. It is stewarded by the Agentic AI Foundation and read by more than 20 agents.

Reviewed Oct 18, 2026.

A template that works

Short beats complete. Agents follow concrete commands and explicit boundaries far better than prose about values.

AGENTS.md

# AGENTS.md

## Project
One paragraph: what this repo is and who uses it.

## Commands (must pass before you finish)
- Install: pnpm install
- Typecheck: pnpm typecheck
- Test: pnpm test
- Lint: pnpm lint

## Conventions
- TypeScript strict mode. No `any` in new code.
- Server code lives in src/server, UI in src/components.
- Use the existing logger (src/lib/log.ts), not console.log.

## Boundaries
- Never edit generated files in src/gen/.
- Ask before changing database migrations or CI config.
- Do not add dependencies without saying why.

## Pull requests
- Small, focused changes. Explain the why in the description.
- Include the commands you ran and their result.

Rules of thumb

  • Put commands the agent must run to verify its work first. That is the part that changes outcomes.
  • Only write down what an agent cannot infer from the code: unusual conventions, sharp edges, things that look wrong but are intended.
  • In a monorepo, add an AGENTS.md per package. The nearest file to the code being edited takes precedence.
  • Move long, task-specific procedures into Agent Skills so they load only when needed.

Using Claude Code too?

Claude Code reads CLAUDE.md. Import AGENTS.md from it so both stay in sync.

CLAUDE.md

# CLAUDE.md
@AGENTS.md

Agents that read AGENTS.md

Also listed on agents.md: Jules, Warp, Devin, Windsurf, Kilo Code, Augment Code, UiPath, Semgrep and Phoenix.