How to use AGENTS.md generator
- Optionally paste your package.json; the commands fill themselves in.
- Check the install, dev, test, lint, typecheck and build commands.
- Add where things live, your conventions, and what agents must not do.
- Copy or download AGENTS.md and commit it at the root of your repository.
Why agents need instructions
A coding agent starting work in your repository has to work out how to install dependencies, run the tests, and check its own changes, and it has to guess the rules your team follows. Every wrong guess costs a failed run or a pull request you have to correct. AGENTS.md answers those questions once, in a file every major agent reads before it starts: OpenAI Codex, Cursor, Google's Jules, GitHub Copilot's coding agent, Aider, Zed and others.
Claude Code reads CLAUDE.md instead. Add a CLAUDE.md containing the single line @AGENTS.md and Claude Code imports the same instructions, so you maintain one file.
What this generator writes
Paste your package.json and the tool reads your package manager and your dev, test, lint, typecheck and build scripts, and turns them into exact commands such as pnpm test. It then adds a verify step telling agents to run the tests, typecheck and lint before proposing a change, which is the single most useful line an AGENTS.md can contain.
Add short lines about where code lives, the conventions that aren't obvious from the code, and what agents must never do, such as editing generated files or the lockfile by hand. Each becomes a bullet.
Keep it short and true
Agents read the whole file on every task, so every line should change what they do. Leave out what the code already shows, and update the file when commands change: an AGENTS.md with a broken test command is worse than none. Your package.json and answers stay in your browser and are never uploaded.
Example
A typical result has a title and one-line description, a Commands section listing pnpm install, pnpm dev, pnpm test and pnpm typecheck, the line telling agents to run the tests and typecheck before proposing a change, a Project structure section such as src/app: routes and src/lib: shared code, and a Don't section such as never edit pnpm-lock.yaml by hand or commit .env files.
Questions
- Which agents read AGENTS.md?
- AGENTS.md is read by OpenAI Codex, Cursor, GitHub Copilot's coding agent, Google's Jules and many others. Claude Code reads CLAUDE.md; a CLAUDE.md containing @AGENTS.md makes it read the same file.
- What makes a good AGENTS.md?
- Exact commands an agent can run to verify its work, short facts about where code lives, and the rules that aren't obvious from the code. Keep it short: agents read it on every task.
- Is my package.json uploaded?
- No. It's read in your browser to fill in the commands, and nothing is saved.