Agent Action Integrity
Definition
Agent Action Integrity is the discipline of proving that AI agents perform high-risk actions correctly, once, and according to policy.
It is the category Postcept is building toward. The first and missing primitive is Proof-of-Completion against the system of record, the layer that closes the completion gap.
The four guarantees
Agent Action Integrity rests on four guarantees for every high-risk action an agent takes:
Verify completion
Check claimed actions against the system of record, existence, amount, identity, status, and duplicates.
Preserve operation context
Keep idempotency keys, OAuth state, and operation IDs stable across retries, timeouts, and handoffs.
Detect duplicate actions
Block the same operation from completing twice after a retry or lost idempotency key.
Produce proof
Issue signed completion receipts for audit, recovery, and customer communication.
Where it sits in the stack
Authorization decides whether an agent may act. Observability records what happened in the trace. Workflow engines orchestrate execution. Agent Action Integrity sits on top of all of them and asks the question none of them answer: did the action actually complete correctly in the system of record?
Postcept measures the result with a single reliability number, Verified Completion Rate, and backs it with signed completion receipts.
Bring integrity to your agent's actions.
Add Proof-of-Completion to your high-risk agent workflows, verified, proven, and recoverable.