What problem does it solve? Single-pass AI code reviews are unreliable: one reviewer misses issues, hallucinates problems, or rubber-stamps its own work. This Skill enforces an adversarial review process where two independent judges review the same target blindly, only agreed-upon findings get fixed, and every fix is re-judged before completion. ## Core Features & Use Cases - Blind Dual Review: Launches two parallel judge agents with identical targets and criteria, then synthesizes findings into confirmed, suspect, contradiction, and INFO buckets. - Fix-and-Rejudge Loop: Delegates confirmed issues to a surgical fix agent, then immediately re-runs both judges before any commit, push, or session summary. - Severity Discipline: Classifies warnings as real only when normal intended use can trigger them, downgrading theoretical issues to INFO to reduce noise. - Use Case: Before merging a PR, ask for a judgment day review of the changed files; the Skill runs both judges, fixes only confirmed CRITICAL and real WARNING findings after your approval, and re-judges until it reaches JUDGMENT: APPROVED or escalates to you. ## Quick Start Ask the AI to run judgment day on the files changed in your current branch and fix any confirmed issues.
