What problem does it solve? Research artifacts like ideas, experiment plans, and paper drafts often contain hidden weaknesses, unsupported claims, or fatal flaws that a single author or single model cannot catch. This Skill provides independent cross-model review with structured scoring, severity-ranked weaknesses, and actionable fixes. ## Core Features & Use Cases - Cross-Model Independent Review: Sends the artifact to a separate Review LLM via MCP, keeping the reviewer's judgment independent from the primary model's pre-judgments. - Three Difficulty Levels: Standard single-round review, hard multi-round rebuttal dialogue (up to 3 rounds), and adversarial mode that actively hunts for fatal flaws like data leakage or incorrect proofs. - Four Review Focuses: Method correctness, evidence rigor, writing quality, or completeness (missing baselines, ablations, related work). - Wiki Entity Mapping: Maps findings to specific wiki claims needing stronger evidence and knowledge gaps, without modifying the wiki itself. - Use Case: Before submitting a paper draft, run an adversarial review focused on evidence to surface unsupported claims and missing controls, then hand the report to /refine for revision. ## Quick Start Review the wiki idea 'sparse-lora-for-edge-devices' at adversarial difficulty with a focus on evidence and give me the structured report.
