Skip to main content
GUIDE · VERIFICATION

What is an AI agent work receipt?

Updated · 5 min read

SHORT ANSWER

What is an AI agent work receipt?

An AI agent work receipt is a structured record an agent hands back with finished work: what was requested, who acted and with what authority, what changed, the evidence, the outcome and whether it was verified. It differs from a log because it is about the result, not every step. Agent Work Receipt v0.1, published by OrgX, is an open JSON Schema for it that you can validate without an account.

What a receipt contains

  • Intent: the task or outcome the agent was asked for, and by whom.
  • Authority: what the agent was allowed to do, and any approvals it got.
  • Actions: what it changed, in summary (files, pull requests, messages, records).
  • Evidence: links a reviewer can open (pull request, test run, published URL, dashboard).
  • Outcome and verification: whether the result was checked, how, and the verdict.
  • Cost and lineage: what it cost where known, and which earlier work it builds on.

Receipt, log or trace?

  • A trace records every model and tool call. It is for debugging the agent.
  • An audit log records events as they happen. It is for reconstructing what occurred.
  • A receipt summarises one unit of finished work with its evidence. It is for deciding whether to accept the work and for the next person or agent who picks it up.

Agent Work Receipt v0.1

OrgX publishes Agent Work Receipt v0.1 as a framework-neutral JSON Schema, with example receipts from Codex, Claude Code and OpenClaw and two invalid examples that show what the validator rejects (missing authority, unresolved evidence). You can emit receipts locally, validate them at useorgx.com/agent-work-receipts or through the validation API without an account, and import them into a shared OrgX record only when a person authorizes it.

Other uses of the term

Some projects use "agent receipt" for cryptographically signed logs of individual agent actions. That is a related idea at a lower level: proving a specific call happened. A work receipt is about the deliverable and whether it was verified, and it can link to signed action logs as evidence.

Questions people ask

Who writes the receipt, the agent or a person?

The agent drafts it from what it did; verification and acceptance are added by checks and by a person. A receipt where the agent marks its own work verified should not be trusted, and OrgX does not let an agent approve its own work.

Do receipts work with Claude Code and Codex?

Yes. The schema ships with example receipts from both. In OrgX, Codex sessions produce receipts more reliably today than Claude Code sessions, which need the OrgX hooks installed.

Is the schema free to use?

Yes. The schema and validator are public and do not need an OrgX account.

Related

Proof for AI-delivered work

OrgX is proof for AI-delivered work — the operating record your agents and tools share.

Start free