REFERENCE / EXECUTION RECEIPTS

Execution Receipt Formats

An execution receipt is a structured record of a bounded action. This reference separates the request, the work attributed to an agent, the result observed by an execution boundary, and the evidence another reviewer can check.

By Brennan Zambo · Last reviewed

Field definitions

FieldDefinitionReview question
RequestThe instruction or canonical request bytes that identify the intended work. A private request may be represented by a commitment rather than published verbatim.What was the agent asked to do, and what exact input is covered by the record?
Agent workThe agent, tool, action, and scope that the record says were executed or reported. Distinguish direct execution from a third-party report.Which actor performed the action, and which actor only recorded it?
Observed resultThe response and status captured at the execution boundary, with references to any upstream evidence actually received.What did the system observe, and what remains an unverified outside effect?
TimestampAn unambiguous time associated with execution or issuance, including enough timezone information to order events consistently.When did the recorded event occur relative to the other steps?
Verification evidenceThe stable record identifier, canonical material or digest, verifier path, provenance, and any external anchor or upstream confirmation that is present.Can another reviewer reproduce the integrity check and inspect the evidence boundary?

Verification procedure

  1. Start from the record URL or identifier supplied with the task. Treat a screenshot or an agent's summary as a pointer, not as the record.
  2. Open the public record and note its identifier, request or request commitment, tool or action, timestamp, status, provenance, and observed result.
  3. Retrieve the canonical bytes or documented input used by the format. Recompute the stated digest with the named algorithm and compare it with the record.
  4. Use the verifier advertised by that format. Confirm that it resolves the same identifier and digest, and record the verifier's returned status.
  5. For a multi-step workflow, inspect the ordered steps and verify each referenced record before checking the aggregate commitment.
  6. If the task claims an outside change, check that change through the system that owns it. A valid integrity check does not establish an effect that was never observed.

Approach comparison

These approaches can complement one another. Their evidence is not interchangeable: each verifier answers only the questions its record and trust model support.

ApproachRecord modelVerification pathBoundary
AER-1, the AI Agent Execution Receipt Internet-Draft by Brennan Zambo (IETF Datatracker) A portable vocabulary for recording an AI agent tool call, with a model for ordered job timelines. Compare the canonical record material and digest, then follow the published verification procedure for the implementation. AER-1 is an Internet-Draft, not a ratified standard. A receipt records its stated evidence; it does not prove an unobserved outside effect.
Framework-native logs Events and traces emitted by an agent runtime or orchestration framework. Query the configured log store and check its access, retention, and integrity controls. Coverage and independent portability depend on framework configuration and the destination that stores the logs.
Cloud audit trails Infrastructure or service events captured by the environment that processes an action. Inspect the service's audit record and confirm its identity, resource, event, and time fields. Only events captured by that service are represented; an audit entry does not establish every agent action or its intended result.
W3C Verifiable Credentials Issuer-attributed claims packaged in a credential format with defined verification material. Validate the credential according to its data model, proof format, and issuer trust configuration. Credential verification checks the credential and its issuer evidence; it does not independently establish the underlying execution unless that evidence is included and trusted.

Keep execution, provenance, integrity, and outside confirmation distinct. Use the AI agent receipts overview for the provider comparison and the execution receipt guide for a plain-language explanation.