| name | semanticscholar-skill |
| description | Use when searching Semantic Scholar for academic papers, citations, authors, similar papers, DOI/arXiv/PMID metadata, BibTeX candidates, citation graphs, or literature discovery for research papers, theses, surveys, or related work. |
Semantic Scholar Skill
Use this skill for real Semantic Scholar API searches and metadata lookup. It keeps literature discovery grounded in returned API data instead of memory.
Requirements
- Python 3.
- Python package
requests.
- Optional environment variable
S2_API_KEY for higher rate limits.
- Skill directory:
~/.codex/skills/semanticscholar-skill when globally installed.
If S2_API_KEY is missing, anonymous requests may still work with stricter rate limits. Do not ask for an API key unless rate limits block the task or the user wants higher throughput.
Core Rule
For real API work, write one temporary Python script that imports s2.py, performs all planned queries in one run, and prints Markdown/JSON/BibTeX output. Avoid many separate API calls from the shell.
Workflow
Read references/workflow.md for query patterns and examples. Read references/source-map.md for provenance and license notes.
- Classify the request: broad topic, exact title, DOI/arXiv/PMID, author, citation traversal, or recommendations.
- Build focused queries with synonyms, exclusions, year filters, field filters, and result limits.
- Run one Python script using
s2.py.
- Deduplicate and rank results by relevance to the user's research question.
- Return a table plus follow-up actions.
Output Contract
For each important paper, include:
- title
- authors
- year
- venue
- DOI, arXiv, URL, or Semantic Scholar ID when available
- citation count if available
- why it is relevant
source_status: normally candidate until metadata is verified elsewhere
next_action: usually verify_metadata, add_to_zotero, check_with_scite, cite_in_related_work, or do_not_cite_yet
- default
zotero_status=not_added and scite/support_status=not_checked when handing candidates to a thesis literature matrix
- BibTeX, JSON, or Zotero follow-up when requested
Mark metadata as unverified if the API result is incomplete or suspicious. Use publisher, DOI, arXiv, PubMed, Scite, or Zotero follow-up for final citation verification.