One definition of done for everyone
Small teamsTeam with mixed agents and habits
One developer's agent ran lint and tests, another's ran nothing. Main was red most mornings.
How do I make every developer's AI agent run the same lint and tests before it finishes?
Commit an ArcRoom .arcroom.json file with your team's done-checks, such as lint, typecheck and unit tests. Every agent, on every laptop, runs the same checks before it may stop. Each laptop approves the checks once, and a changed check needs approval again. Keep done-checks fast and leave heavy suites to CI, since long test suites make stopping slow.
Also asked as: Some of our agents run tests and some don't, and main keeps breaking. How do we fix that? · How do I set one definition of done for all our AI agents? · Can every developer's AI agent use the same checks before stopping?
The team commits .arcroom.json with its done-checks: lint, typecheck, unit tests.
- Done-checks
Done-checks (.arcroom.json): pnpm lint ✓ · pnpm typecheck ✗ · pnpm test –
Step by step
- The team commits .arcroom.json with its done-checks: lint, typecheck, unit tests.
- Every agent, on every laptop, runs the same checks before it may stop.
- Checks are commands, so each laptop approves them once. A changed check needs approval again.
What the agent sees
Done-checks (.arcroom.json): pnpm lint ✓ · pnpm typecheck ✗ · pnpm test –
Not done: fix typecheck first.Compared with today's tools
Pre-commit hooks + required CI checks
Run lint and tests in a pre-commit hook and block merges while CI is red.
- Setup
- 1–2 hours
- Ongoing
- low once set up
ArcRoom
Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.
- Setup
- 5 min per person (init + arcroom checks approve)
- Ongoing
- None
CI stays the real gate and catches what agents skip; done-checks make the agent fix it in the same session instead of the next morning.
Where it stops
Long test suites make stopping slow. Keep done-checks fast and leave the heavy suites to CI.
Try it in your project
Works with the agents your team already uses, and any agent over MCP or the HTTP API.