Run Receipt

A multi-step agent run is not complete until a receipt exists: a cryptographically signed, tamper-evident record of what ran, what failed, what was not observed, and who authorized each action. If no receipt exists, the run did not happen — you have no defensible audit trail.

This is observability as a first-class output, not an afterthought. Without receipts, incident investigations turn into "what did it actually do?" archaeology.

When to run

• An agent completes a multi-step task
• An agent fails mid-task
• A production incident requires investigation
• Compliance or audit asks "what did the agent do?"
• The agent made an irreversible action (deployment, deletion, financial transaction)

If you cannot produce a receipt for the last 10 agent runs, your observability is logs and hope.

What is a receipt

A run receipt is a cryptographically signed audit record of one agent action. Each receipt is: