What problem does it solve? Single-reviewer code reviews miss blind spots and reflect one model's biases. This Skill runs adversarial multi-model reviews where independent LLM reviewers challenge the same diff, then synthesizes their findings into a prioritized verdict so real issues surface and noise gets filtered out. ## Core Features & Use Cases - Multi-Model Adversarial Review: Spawns one reviewer per configured model, each applying the same rubric and code-quality lens, so agreement across models signals high-confidence findings. - Lead Judgment Synthesis: Categorizes every finding as Act On, Consider, Noted, or Dismissed, with rationale, filtering nitpicks and hypothetical issues using full conversation context. - Structured Verdict Output: Produces an intent statement, reviewer summary, prioritized findings, and an agreement map showing where models converged or diverged. - Use Case: Before merging a feature branch, ask for an adversarial review of the diff; multiple models independently flag a race condition and a missing test, while style-only suggestions are dismissed with explanations. ## Quick Start Ask the assistant to interrogate the current branch diff and produce an adversarial multi-model review verdict.
