Skip to content

Accountability infrastructure for autonomous engineering

Developers

Built for coding agents. Not tied to one.

Accountability belongs to the run, not to the harness that happens to be executing it. Swapping agents should not change what counts as evidence.

  • Harness
  • Run
  • Evidence

Ecosystem targets

Compatible workflows

These are the environments the tooling is designed to sit alongside.

  • Codex

    CLI coding harness

  • Claude Code

    Terminal coding agent

  • Cursor

    Editor-resident agent

  • Lovable

    Application build agent

  • OpenClaw

    Open harness

  • Replit

    Hosted agent environment

Integration surfaces

Where each layer touches your stack

Read the columns as authority, not as features.

Integration surfaces by layer
Integration surfaces by layer
DimensionDoneStateOpsTruthAgentProof
Coding harnessRuns the work inside a DoneState runNot inspected as a trusted narratorMay propose, never authorise
Version controlScoped branch write under the envelopeRead at an exact commitPR-first for remote mutation
CI pipelineEmits evidence into the run recordRead as verification evidenceReceipts reference the CI evidence
Runtime and deploymentDenied by default in the envelopeObserved read-onlySigned if explicitly authorised

Endpoints

Canonical service and MCP addresses

Point tooling at these addresses. Any workers.dev hostname you may have on file is historical and is not a canonical link.

DoneState service
https://donestate.proofandstate.com
DoneState MCP
https://donestate.proofandstate.com/mcp
OpsTruth website
https://opstruth.io
OpsTruth MCP
https://mcp.opstruth.io/mcp

Approach

Adopt one layer at a time.

Most teams start with the verifier, because it changes nothing. It reads a repository and returns a classified report; there is no write path to be nervous about.

The second step is usually durable execution: putting agent runs under DoneState so that an interrupted session leaves a known state and a run cannot mark itself done.

Receipts come last, when consequential actions are involved. They are worth the overhead exactly when an action cannot simply be retried.

Give agents authority. Keep the proof.

The harness is replaceable. The evidence should not be.