What problem does it solve? Before investing months into a research project, you need to know whether your proposed method has already been published. This Skill systematically checks a research idea against recent literature (arXiv, ICLR, NeurIPS, ICML) and produces a brutally honest novelty assessment so you avoid duplicating existing work. ## Core Features & Use Cases - Claim Decomposition: Extracts 3-5 core technical claims from your method description and checks each one individually for novelty. - Multi-Source Literature Search: Queries arXiv, Google Scholar, and Semantic Scholar with multiple query formulations, filtered to 2024-2026 venues and preprints. - Cross-Model Verification: Sends a dossier of your method and candidate papers to a reviewer model (e.g., gpt-5.5 via Codex MCP) for an independent novelty judgment. - Structured Novelty Report: Outputs per-claim novelty ratings, a closest-prior-work table, a 0-10 score, and a PROCEED / PROCEED WITH CAUTION / ABANDON recommendation. - Use Case: You have an idea for a new attention mechanism. Run this Skill to discover that a similar mechanism appeared in a recent arXiv preprint, then reposition your contribution around the remaining delta before writing any code. ## Quick Start Ask the assistant to run a novelty check on your proposed method, for example: "Check the novelty of my idea: using retrieval-augmented pruning to compress mixture-of-experts models."
