novelty-assessment — Claude Code skill: check if your idea is new
Runs a research idea through literature search and returns a verdict: novel or not. Useful before you commit to a direction.
npx skills add lingzhi227/agent-research-skills --skill novelty-assessment -a claude-codeWhat novelty-assessment does
The skill takes an idea description, a title, or a JSON file and hunts for published work that overlaps with it. Search runs in rounds, up to 10: each round builds a query, reviews the top results with abstracts, and scores the overlap. The decision is binary — novel or not_novel — with a justification and a confidence level.
Two scripts do the work. novelty_check.py drives the search-and-evaluate loop and writes novelty_report.json; search_semantic_scholar.py queries papers on its own. Sources are Semantic Scholar, arXiv and OpenAlex. A harsh-critic persona is built in: a trivial extension of existing work does not count as novel, the idea needs a meaningfully different approach or formulation. At least 3 search rounds are required before anything is called novel.
When the idea passes, the skill points out the closest papers and the gap it fills, which feeds straight into Related Work. Methodology novelty and application novelty are judged separately. The install pulls dependencies from the repository, so make sure Python and search API access are in place before the first run.
What it helps with
- check whether someone already published a similar paper
- find the closest papers for a Related Work section
- rule out trivial extensions before starting the project
- articulate how your idea differs from existing work
- catch recent arXiv preprints on your topic
Example prompts for the agent
- /novelty-assessment "graph neural networks for protein folding prediction"
- check the novelty of this idea: adaptive quantization for LLM inference on edge devices
- assess the idea in idea.json and list the closest papers
How to install novelty-assessment
1Skills CLI (Vercel)
npx skills add lingzhi227/agent-research-skills --skill novelty-assessment -a claude-code2Manually: put the files in place
~/.claude/skills/novelty-assessment/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/novelty-assessment/
1Skills CLI (Vercel)
npx skills add lingzhi227/agent-research-skills --skill novelty-assessment -a codex2Manually: put the files in place
.agents/skills/novelty-assessment/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add lingzhi227/agent-research-skills --skill novelty-assessment -a cursor2Manually: put the files in place
.cursor/skills/novelty-assessment/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install lingzhi227/agent-research-skills novelty-assessment2Manually: put the files in place
.github/skills/novelty-assessment/SKILL.md1Skills CLI (Vercel)
npx skills add lingzhi227/agent-research-skills --skill novelty-assessment -a gemini-cli2Platform CLI
gemini skills install https://github.com/lingzhi227/agent-research-skills --path skills/novelty-assessmentManually: put the files in place
.devin/skills/novelty-assessment/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add lingzhi227/agent-research-skills --skill novelty-assessment -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about novelty-assessment
How do I install novelty-assessment in Claude?
Run npx skills add lingzhi227/agent-research-skills --skill novelty-assessment -a claude-code. The skill lands at ~/.claude/skills/novelty-assessment/SKILL.md.
What does novelty-assessment do?
It searches the literature for papers that overlap with your idea and returns novel or not_novel with a justification. It also lists the closest papers and how your idea differs.
How many search rounds does it need?
The skill requires at least 3 rounds before calling an idea novel, and allows up to 10. No verdict comes earlier.
