Technique
Write the repo rules down once
Agents A few minutes First logged
- InA repository an agent will touch
- ThenOne file at the root stating build, test and boundaries
- OutChanges that arrive already inside your rules
What the move is
Put the build command, the test command, the untouchable files and the definition of a complete change in a file at the repository root. Agents read it. So do new humans.
Why it works
This is what maintainers absorbing agent-written pull requests at scale actually did: repo-specific instruction files, pull request templates that auto-close when unmet, mandatory test plans, and coverage gates.
Step by step
- Create AGENTS.md (and/or CLAUDE.md) at the repository root.
- State the build command and the test command exactly, as commands, not prose.
- List the files and directories that must not be modified, and why.
- State what a change must include: tests, a scope, a size limit.
- Make CI enforce the parts that can be enforced, so the file is a description of the gate rather than a request.
Paste this
Read this repository and draft AGENTS.md for its root. Four things only: the build and test commands exactly as they are run, the files that must not be modified, what a change must include before it is finished, and one paragraph on the layout. If a command is not written down anywhere, say so instead of inferring one.
Make it yours
Add a short 'how this project is laid out' section. It is the single highest-value paragraph for both an agent and a new contributor.
Variations
- Add per-directory rules files where a subsystem has its own conventions.
- Keep a short 'things that look wrong but are deliberate' list to stop well-meant refactors.