Agent contract
The agent contract is the repo-specific rule set that keeps AI-assisted work safe and reviewable.
It should answer:
- when the agent must stop and ask
- what it must verify before claiming success
- which areas are risky
- which areas are safer to edit
Example
Good contract guidance:
Pause before changing public SDK behavior or wire contracts.Run focused repo checks before claiming success.Transport boundaries and migration-sensitive code are risky.
Bad contract guidance:
Be careful.Run tests.Avoid risky files.
Why this matters
Agents behave much better when the repo boundaries are explicit. A thin but specific contract is more useful than a long page of generic caution.
Next steps
- Continue to System prompt strategy.
- Finish with Validation and handoff.