I built a QA agent that audits itself nightly. It has filed 79 findings against its own releases.

Most "AI QA agents" audit your repo from scratch every time and end with LGTM. I wanted the opposite:...