Проверка научной новизны идеи

★ 7.1 · research

novelty-check is a Claude Code skill that systematically verifies whether a proposed research idea or method has already appeared in the academic literature. It starts by extracting 3–5 core technical claims from the method description, then runs multi-source searches via WebSearch across arXiv, Google Scholar, and Semantic Scholar, covering ICLR, NeurIPS, and ICML 2025–2026 as well as recent preprints. A cross-model verification step follows: the method description and all candidate papers are written to a NOVELTY_DOSSIER.md file and passed to gpt-5.6-sol through Codex MCP with xhigh reasoning effort. The final report rates novelty for each claim (HIGH / MEDIUM / LOW), lists closest prior work in a structured table, assigns an overall score out of 10, and issues a PROCEED / PROCEED WITH CAUTION / ABANDON recommendation — making it the right tool for ML and AI researchers who need an honest novelty assessment before committing to months of experiments.