AI work accountability

independent verification

Make agent work provable.

Give agents room to work without giving up control. Proof & State separates execution, verification and receipts so done means something you can inspect.

Explore the system View on GitHub

Proof & State is AI work accountability infrastructure: DoneState executes under declared authority, AgentProof authorises consequential actions and signs receipts, and OpsTruth verifies the result read-only from outside the execution path.

Evidence rail

01
Outcome
Prose objective from a human
02
Authority
Scope, budget, expiry declared up front
03
DoneState
Admission, lease, durable transitions
run/7c41
04
AgentProof
Consequential action, signed receipt
ap1:7c4e9d2b
05
OpsTruth
Independent read-only inspection
report/2f19
06
VERIFIED
Attestation closes the run state

A run only reaches VERIFIED through a component that could not have caused it.

One system. Independent layers.

Three layers that refuse to vouch for each other.

DoneState executes under declared authority. AgentProof authorises consequential actions and signs receipts. OpsTruth verifies from outside the execution path. The separation is the product.

Layer 01 — ExecutionAWAITING_VERIFICATION

DoneState

Durable execution and control plane

A run begins with a prose outcome and an explicit authority envelope. DoneState decides whether that run is admissible, holds the lease while it executes, records every transition durably, and stops at AWAITING_VERIFICATION until an independent attestation arrives.

Admission control
Budgets and leases
Idempotency
Durable state transitions

DoneState in detail Source

DoneState — run staterun/7c41

ADMITTED
Outcome accepted inside authority envelope
EXECUTING
Lease held, budget consuming
VALIDATING
Local checks and build evidence gathered
AWAITING_VERIFICATION
Halted — independent attestation required
VERIFIED
Attestation received, state closed

Layer 02 — VerificationREAD_ONLY

OpsTruth

Independent read-only verifier

OpsTruth reads the system from outside the execution path. It gathers evidence, states plainly what that evidence supports, and refuses to upgrade an absence of evidence into a pass.

Repository and stack inspection
Tests, build and CI evidence
Secrets and configuration checks
Routes and runtime evidence

OpsTruth in detail Source

OpsTruth — verification reportread-only

Illustrative OpsTruth verification report showing Verified, Risky and Unproven checks
Check	Surface	State
Build reproduces at commit Artefact matches source tree	build	verified
Test suite executed Exit status recorded in CI evidence	tests	verified
No credentials in tracked files Scan clean at this commit	secrets	verified
Declared routes respond One route returns an error status	runtime	risky
Deployed artefact matches commit No deployment evidence available	deployment	unproven

Layer 03 — AuthorisationRECEIPT_SIGNED

AgentProof

Authorised transaction and signed-receipt layer

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.

Exact prepared state
Exactly-once execution
Recovery
Independently verifiable receipts

AgentProof in detail Source

AgentProof — signed receiptap1:7c4e9d2b83a15f60c8e7d419ab35f2c0

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.

Architecture

Execution on one side. Verification on the other.

A human objective and an authority envelope enter DoneState. Consequential effects emit AgentProof receipts. OpsTruth observes repository, CI and runtime evidence independently, and verification is what closes the state.

Proof & State — system topologyexecution ⟂ verification

Execution side

may act

Inputobjective
Human objective + authority
A prose outcome and an explicit authority envelope enter the system together.
Execution / control planedonestate
DoneState
Admits the run, holds the lease, records durable transitions, halts at AWAITING_VERIFICATION.
Authorised transactionsagentproof
AgentProof
Consequential effects are bound to prepared state and emit a signed receipt.

Verification side

observes only

Surfaceevidence
Observable evidence
Repository at an exact commit, CI output, configuration, routes, runtime and deployment artefacts.
Independent verifieropstruth
OpsTruth
Reads only. Classifies each check as Verified, Risky or Unproven.
Outputattestation
Attestation closes state
The run closes when an independent attestation says observed state matches the claim.

Self-verification is not proof

Full system topology Run state model Authority model

Evidence

The agent can say it is done. Your system should prove it.

Every run produces inspectable artefacts: a classified verification report, a signed receipt digest and a durable state timeline you can read after the session is gone.

OpsTruth — verification reportread-only

Illustrative OpsTruth verification report showing Verified, Risky and Unproven checks
Check	Surface	State
Build reproduces at commit Artefact matches source tree	build	verified
Test suite executed Exit status recorded in CI evidence	tests	verified
No credentials in tracked files Scan clean at this commit	secrets	verified
Declared routes respond One route returns an error status	runtime	risky
Deployed artefact matches commit No deployment evidence available	deployment	unproven

AgentProof — signed receiptap1:7c4e9d2b83a15f60c8e7d419ab35f2c0

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.

DoneState — run staterun/7c41

ADMITTED
Outcome accepted inside authority envelope
EXECUTING
Lease held, budget consuming
VALIDATING
Local checks and build evidence gathered
AWAITING_VERIFICATION
Halted — independent attestation required
VERIFIED
Attestation received, state closed

Specimens show the shape of each artefact. They are illustrative, not records of real runs.

Ecosystem

Built for coding agents. Not tied to one.

Accountability belongs to the run, not the harness. The coding agent is a replaceable component.

Codex
CLI coding harness
Claude Code
Terminal coding agent
Cursor
Editor-resident agent
Lovable
Application build agent
OpenClaw
Open harness
Replit
Hosted agent environment

These are ecosystem targets and compatible workflows. Naming a harness is not a claim of partnership, endorsement, certification or affiliation.

Principles

Trust earned by structure, not by assurance.

Three positions the whole system is built on. Each one is enforced by architecture rather than by policy.

Developers

Start with the verifier.

Point the read-only verifier at a repository you already trust and see what it can and cannot evidence.

Quickstart Open source

verify a repository

$ npx opstruthscanning repository, stack, tests, build, CI, config, routesverified 3 · risky 1 · unproven 1# read-only: no deploys, no writes, no restarts

Definition — Proof & State: Proof & State is AI work accountability infrastructure: DoneState executes under declared authority, AgentProof authorises consequential actions and signs receipts, and OpsTruth verifies the result read-only from outside the execution path.

Give agents authority. Keep the proof.

Read the architecture, inspect the source, and decide for yourself whether the separation holds.

Read the architecture View on GitHub