novelty-check

Extract technical claims and search arXiv and ML venues for prior work.

Updated May 29, 2026
One-click install
npx skills add https://github.com/Mang30/myskills --skill novelty-check-mang30
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: novelty-check
Source: https://github.com/Mang30/myskills/tree/main/skills/novelty-check
Command: npx skills add https://github.com/Mang30/myskills --skill novelty-check-mang30

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It helps prevent wasted effort by checking whether a proposed research method or idea has already been done in recent literature.

Core Features & Use Cases

  • Claim-based novelty verification: Extracts 3–5 core technical claims from a method description to test.
  • Multi-source literature search: Runs web searches (including 2024–2026 filters), checks key venues (ICLR/NeurIPS/ICML 2025/2026), and inspects recent arXiv preprints.
  • Abstract-level overlap checking: Fetches abstracts and related-work context for likely overlapping papers.
  • Cross-model reviewer synthesis: Uses a dedicated reviewer model via Codex MCP to compare found prior work against the proposed method and produce an overall delta.
  • Structured novelty report: Outputs a table of closest prior work, a novelty score (X/10), an explicit recommendation (PROCEED / PROCEED WITH CAUTION / ABANDON), and positioning guidance.

Quick Start

Use the novelty-check skill to verify whether your proposed method is novel given its description, especially when you want to do a "查新" or "有没有人做过" check before implementing.

Frequently Asked Questions about novelty-check

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify if my research idea has already been published in recent literature?

A novelty check extracts 3–5 core technical claims from your proposed method and searches across recent arXiv preprints and major ML venues like ICLR, NeurIPS, and ICML to identify overlapping prior work. It outputs a structured novelty report with a score and closest matching papers.

What is the best way to check if a machine learning method is novel before implementation?

Using a claim-based novelty verification process is the best way to check if a machine learning method is novel. It runs multi-query web searches with 2024–2026 filters across major ML venues, fetches abstracts for overlap checking, and outputs a scored novelty assessment with positioning guidance.

Can I perform a literature search for prior work verification on arXiv and conferences like ICLR or NeurIPS?

Yes, prior work verification searches across arXiv and major ML conferences including ICLR, NeurIPS, and ICML for 2025 and 2026. It fetches abstracts and related-work context to evaluate method, mechanism, and experimental-setting overlaps against your proposed research.

How does a novelty check handle unresolved or unverified papers during a related work search?

During a related work search, unresolved or inaccessible papers are explicitly marked as [UNVERIFIED] rather than fabricating citations. This ensures the reviewer synthesis produces an honest novelty score and an explicit PROCEED, PROCEED WITH CAUTION, or ABANDON recommendation based only on verified overlaps.

What limitations should I expect when checking if someone has already done my proposed research mechanism?

When checking if someone has already done your proposed research mechanism, limitations include reliance on 2024–2026 search filters and the potential inability to fetch some recent preprints, which are marked as [UNVERIFIED] rather than fabricated. Novelty scores are bounded by accessible venue publications and arXiv preprints.