Use other coding agents
OpenHarnX reads the change and the test runs, not the agent, so any agent's work can be checked with the CLI or the CI gate.
Verification does not depend on which agent wrote the code. There is no built-in integration for agents other than Claude Code in 0.1.1, and the table below says what has been tested.
Run it by hand#
Lock the tests before the agent starts, and verify when it finishes:
ohx init --lock-tests # once, or ohx contract new for a task
# ... let the agent work ...
ohx verify --sandbox srt
ohx reportohx verify exits 0 for READY or NO REGRESSIONS and 10 otherwise, so a script or an agent's own "run this when done" step can use the exit code. Paste the "What remains unverified" section back to the agent if it is blocked.
Or let CI judge the pull request#
Any agent that opens pull requests can be judged by the CI gate. The gate runs the base branch's tests against the change and uses the base's policy, so the agent cannot alter what judges it.
What has been tested#
| Agent | Integration | Tested |
|---|---|---|
| Claude Code | Built-in Stop hook (ohx hook install) | End to end, including a headless run that a hook verdict reached |
| GitHub Copilot coding agent | None needed: the CI gate judges its pull requests | Replayed after the fact on 20 merged pull requests in github/spec-kit |
| Codex, Cursor, OpenCode, Antigravity, Aider and others | ohx verify by hand, or the CI gate | Not tested with these agents yet |
Built-in integrations for other agents are on the roadmap. Each depends on the agent offering a way to run a command when it finishes, and ships only after a live run with that agent is recorded. A trial report from one of them helps decide the order.
History of changes already merged#
ohx history --last 20Shows what the gate would have said about recent commits, and marks the ones made by coding agents (--agents-only keeps only those). It writes ohx-history/history.md and history.json.