Authorised transaction and signed-receipt layer
RECEIPT_SIGNEDAgentProof
Consequential actions leave evidence that outlives the session.
- Outcome
- Authority
- AgentProof
- Evidence
- Definition — AgentProof
- AgentProof is the transaction and signed-receipt layer for consequential agent actions: it binds authority to an exact prepared state, executes exactly once with recovery, and emits an independently verifiable signed receipt, keeping proposer, authority, executor, signer and verifier separate.
Before a consequential action runs, AgentProof pins the exact prepared state and the authority that permits it. Execution is exactly-once and recoverable. What comes out is a receipt anyone holding the public key can check.
- action
- merge_pull_request
- prepared state
- sha256:9f2c41ab6e0d7c5183ba0e77c4d21f9a
- authority
- envelope/repo-write@exp-2026-08-29T09:00Z
- proposer
- agent:coding-harness
- executor
- agentproof:executor-01
- signer
- key:ed25519:AP-3f8c
- verifier
- opstruth:report-2f19
- outcome
- committed_once
Illustrative specimen. Field values show the shape of a receipt, not a real transaction.
Illustrative receipt fields. The digest covers the action, the exact prepared state and the authority it was executed under.
Capabilities
What AgentProof does
Each capability exists to make one class of failure impossible to hide.
- Exact prepared state
- Authority is granted against a specific prepared state, not a description of it. If the state drifts, the authority no longer applies.
- Exactly-once execution
- A transaction commits once. Retries after a partial failure resolve, they do not repeat the effect.
- Recovery
- An interrupted transaction resolves to a known terminal outcome rather than an ambiguous one.
- Independently verifiable receipts
- A receipt binds action, prepared state digest, authority and signer. Verification needs the receipt and the public key — not access to the system that produced it.
- Separation of duties
- Proposer, authority, executor, signer and verifier are distinct roles. No single component both acts and vouches for the action.
Explicit limits
What AgentProof does not do
Stated limits are part of the design. A component that could do everything could not be trusted to judge anything.
- Does not grant itself authority
- Does not sign for actions it proposed
- Does not execute against drifted state
Separation of duties
Five roles that must not collapse into one.
A receipt is only meaningful if the party that proposed the action is not the party that authorised, executed, signed and verified it.
- 01
Proposer
Prepares the action and the exact state it would apply to. Holds no authority to execute.
- 02
Authority
Grants a scoped, expiring envelope bound to that prepared state. Does not perform work.
- 03
Executor
Applies the action exactly once, with idempotent recovery after interruption.
- 04
Signer
Signs the receipt over the action, prepared state and authority reference.
- 05
Verifier
Checks the receipt independently, without trusting the executor's account of events.
Relationships
How this layer relates to the others
Proof & State is one system of three independent layers. Each holds authority the others do not.
Execution
DoneState
Give the agent room to work without giving up control.
Verification
OpsTruth
Verification that holds no power to change what it judges.
DoneState executes; it cannot verify itself. OpsTruth verifies read-only and never writes. AgentProof authorises consequential actions and signs receipts. The separation is the product.