I've been building software with multiple AI agents, and one recurring problem led us to create Agent Role Contracts (ARC).
An implementer says "done." A reviewer says "passed." But sometimes the task scope has changed, the review refers to an older version, or it's unclear who is responsible for the next step.
We built ARC to check the consistency of role, scope, review, and handoff declarations before execution. It doesn't enforce runtime permissions or prove that work was completed.
When an AI agent is assigned a task outside its declared scope—or asked to review its own work—things can go wrong before execution even starts.
Agent Role Contracts checks role, authority, task scope, reviewer separation, and handoff declarations offline.
Try the 3-minute PASS → FAIL → PASS demo with no API key or agent runtime. Free and MIT-licensed.
It checks declarations, not runtime permissions.