What problem does it solve? Researchers risk wasting months implementing methods that already exist in the literature. This Skill systematically checks whether a proposed method or idea is novel before committing to implementation, comparing it against recent papers from arXiv, major ML venues, and web sources. ## Core Features & Use Cases - Claim Extraction: Breaks a method description into 3-5 core technical claims that each need independent novelty verification. - Multi-Source Literature Search: Searches arXiv, Google Scholar, Semantic Scholar, and recent conference proceedings (ICLR, NeurIPS, ICML) with multiple query formulations per claim. - Cross-Model Verification: Sends findings to an external LLM reviewer via the llm-chat MCP for strict, high-rigor feedback on closest prior work. - Structured Novelty Report: Produces a scored report with per-claim novelty ratings, closest prior work table, an overall PROCEED / PROCEED WITH CAUTION / ABANDON recommendation, and positioning suggestions. - Use Case: Before implementing a new mechanistic interpretability probing method, run a novelty check to discover a nearly identical 2025 arXiv preprint and reposition the contribution around a different experimental setting. ## Quick Start Ask the agent to run a novelty check on your proposed method by describing the idea, for example: "Check the novelty of using sparse autoencoders to steer attention heads in small language models."
