Run an adversarial audit loop on a code/infrastructure change with the reviewer subagent as independent reviewer. Loops build -> audit -> fix -> re-audit until the reviewer returns GO or the round cap is hit. Collects all clarifying questions in one "Open Questions" list at the end so you can walk away during the loop. Use when you say "/quality-gate", "audit the X until it passes", "send X to a reviewer subagent for review and fix", or asks to harden something before paid traffic / a client hand-off / a go-live.
