What problem does it solve? Single-model reviews can be biased, overconfident, or miss critical issues on high-stakes outputs. This Skill dispatches independent critics from different model families, detects disagreement, and resolves conflicts through adjudication so important decisions get multi-perspective validation. ## Core Features & Use Cases - Parallel Multi-Model Panel: Dispatches three critics across different provider families (nemotron, deepseek, mimo) concurrently using background task dispatch. - Consensus & Adjudication Logic: Applies a decision table for majority verdicts, ties, and low-confidence splits, escalating to an oracle adjudicator only when critics disagree. - Domain Rubric Library: Provides structured scoring rubrics for code review, architecture, planning, and content evaluation. - Use Case: Before merging a security-sensitive code change, run the critique pipeline to collect three independent verdicts, adjudicate any split, and receive a severity-sorted findings list with a concrete recommendation. ## Quick Start Use the critique skill to run a multi-model review of this draft response and give me a synthesized verdict with combined findings.
