What problem does it solve? Getting rigorous, adversarial feedback on research ideas, papers, and experimental results is hard without access to senior reviewers. This Skill orchestrates a multi-round critical review of your research from an external reviewer backend (Codex or manual review) with maximum reasoning depth, simulating a NeurIPS/ICML-level review process. ## Core Features & Use Cases - Multi-Backend Review: Routes review requests to Codex MCP or a manual review MCP, maintaining thread continuity across rounds via saved threadIds. - Adversarial Review Workflow: Compiles a comprehensive research brief, then iterates through rounds of criticism, rebuttal, and follow-up until claims, experiments, and narrative converge. - Structured Deliverables: Produces a self-contained review document with round-by-round summaries, a claims matrix, prioritized TODOs with compute estimates, and paper outlines. - Use Case: You have a draft ML paper with preliminary results. Invoke this Skill to get a brutal mock review, identify missing experiments, and receive a minimal experiment package ranked by acceptance lift per GPU week. ## Quick Start Ask the assistant to review my research using the research-review skill with the topic of my current paper draft.
