sanmu-paper-search: Find and Deduplicate Academic Papers in Claude Code or Codex
Search scholarly metadata by topic or seed paper, follow references and citing papers, and merge records by DOI or other stable identifiers. Each run records the sources actually queried and their status.
中文Install in Claude Code、Codex
The standalone core CLI uses the Python 3.10+ standard library. Live queries require network access. Crossref provides public metadata queries; some other providers require their own API credentials.
- Download the complete skills/sanmu-paper-search/ directory at the pinned commit, including SKILL.md, scripts/, and references/.
- Place it in ~/.claude/skills/ for Claude Code or ~/.agents/skills/ for Codex. In the skill directory, use python scripts/paper_search.py doctor and python scripts/paper_search.py --help to inspect configuration and arguments.
- Start with one provider, a public topic, a small limit, and a fresh output directory. Configure other providers as needed.
Example task
From the skill directory, use this small Crossref query for a public mathematics topic:
python scripts/paper_search.py search "matrix factorization" --sources crossref --limit 3 --out ./demo-crossrefWhat to check
When the request succeeds, demo-crossref/papers.json contains up to three Crossref candidates, and demo-crossref/run.json records the search. Check the run record if the provider is rate limited or returns zero matches.
Original SKILL.md on GitHub · Updated
Frequently asked question
Can I cite a conclusion from a metadata-only search result?
Use the search results to identify candidate papers. Read the accessible abstract or paper before citing its claims, and record what you actually reviewed.