dos-goal-gate

To stop autonomous coding agents from falsely claiming completion, you can wire a grounded stop gate. This hook refuses to end the agent loop until self-reported progress is corroborated by external evidence like git ancestry checks.

Evidence gating in agent workflows is a verification mechanism that breaks a goal into checkable effects. It requires external witnesses, such as git history, to verify each claimed effect before allowing the agent to stop working.

You verify an agent's refactoring claims by applying claim-to-effect verification. This process checks git ancestry and effect-level witnesses to ensure the agent's self-reported progress is actually backed by repository evidence before stopping.

Yes, enforcing grounded completion for long-running agent tasks requires stop-hook integration. The stop hook intercepts the agent's self-reported done claims and blocks them until external evidence verifies the work actually happened.

You should use claim verification for autonomous coding loops, migrations, and long-running refactors. It is necessary whenever an agent might self-report completion before the repository provides external evidence that the work is actually finished.

Yes, you can use this stop gate for migrations without custom scripts by telling the agent to wire the grounded stop gate. It automatically breaks the goal into checkable effects and verifies each one against external witnesses before stopping.