
Your coding agent says tests pass. Prove it.
OpenHarnX locks the tests you agree on before an AI coding agent touches you code, runs them in a sandbox the agent can't reach, and gives you a review brief: - What's verified - What's certain - What needs your judgement No model grades its own work. Open source, Apache 2.0 works with any coding agent with the built-in Claude code integration.
A source that found nothing is a measurement. A source that has not run is a gap. Neither means the launch lacks the thing.