Ultraqa

This is an autonomous QA loop that runs your tests, build, lint, or typecheck commands, spawns an architect to diagnose failures, applies fixes, then reruns until everything passes or it hits five cycles. The neat part is the early exit logic: if the same failure appears three times, it stops and surfaces the root cause instead of thrashing. You invoke it with flags like `--tests` or `--build`, and it handles the entire diagnose-fix-verify cycle without you babysitting it. It's opinionated about cleanup, explicitly deleting state files on completion rather than just marking them inactive. Best used when you know what passing looks like but don't want to manually iterate on flaky tests or build errors.