Agency Receipts · The receipt layer for agent actions

A Signed Record That It Happened

An agent that acts for a human leaves no standardized, verifiable record of what it did. Agency Receipts is the per-action receipt an agent signs when it acts — readable in one sentence, verifiable offline, chained to the agent's own history.

A receipt answers one question: did this agent action happen? The Agency Receipt (sm-arp) is a signed, human-readable record that an agent took a specific action, for a specific principal, under a specific authority — Ed25519-signed over a JCS-canonical body, hash-chained to the issuing agent's prior receipts, and verifiable by anyone holding the issuer's public key. It sits above MCP (tool integration) and A2A (agent-to-agent transport), recording for the human the outcome an agent produced through them.

What is Agency Receipts → The primitives →

What a receipt binds

A receipt binds, in one signed object, the facts a human and a third party need to understand and dispute one agent action.

(a) It happened

The action occurred

Bound to the signed action object and its human_summary — one plain-language sentence, ≤ 280 Unicode code points, that a person can read without tooling.

(b) Under what authority

The authority chain

Bound to authority_chain — identifiers of the delegation grants the action relied on. In v0.1 this is descriptive: a verifier notes it but does not yet evaluate the underlying grant.

Out of scope. A receipt proves that an agent's action happened — signed, timestamped, one-sentence-readable — not that it was authorized. Full pre-action authorization — the permit and the grant behind it — is a companion concern, kept deliberately separate. See Agent Envelope →