Clarity Harness
Keep the run. Keep the verdict. Check the next version.
For builders who want to operate their own agent tooling and carry a useful correction into the next iteration.
Free · OSS · Coming soon
Free OSS, coming soon. The waitlist provides release updates; it is not an installation or delivery commitment.
01
Start with what the agent actually did.
The intended workflow connects a run to its inputs, outputs and tool activity, so the review has a concrete case to inspect.
02
Keep the correction with the case.
A useful judgment records what was missed and the constraint behind it. Leave uncertainty visible rather than turning every output into a lesson.
run: representative agent output
verdict: accepted, failed or unresolved
correction: the missed constraint
next check: replay this case on the revision03
Use the case to judge the next iteration.
The next run should be compared with the reviewed case. More stored history is useful when it helps you decide what improved and what still needs work.