Claude Code Plugins

Community-maintained marketplace

Feedback

semantic-scholar

@bdambrosio/Cognitive_workbench
8
0

Search academic papers. Returns Collection of JSON Notes with fields text (full paper text via GROBID when PDF available, otherwise abstract), metadata.title, metadata.authors, metadata.year, metadata.citations, metadata.uri (alias: pdf_url), metadata.venue (Level 4 tool).

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name semantic-scholar
type python
description Search academic papers. Returns Collection of JSON Notes with fields text (full paper text via GROBID when PDF available, otherwise abstract), metadata.title, metadata.authors, metadata.year, metadata.citations, metadata.uri (alias: pdf_url), metadata.venue (Level 4 tool).
schema_hint [object Object]
examples {"type":"semantic-scholar","value":"transformer architecture","out":"$papers"}, {"type":"project","target":"$papers","fields":["metadata.title","metadata.year"],"out":"$titles"}, {"type":"project","target":"$papers","fields":["metadata.uri"],"out":"$urls"}, {"type":"filter-structured","target":"$papers","where":"metadata.citations > 100","out":"$high_impact"}

Semantic Scholar Search Tool (Level 4)

Input

  • Query string (e.g., "attention mechanisms in neural networks")

Output Structure

  • Collection ID containing one structured Note per paper result
  • Each Note contains JSON with uniform structure:
    • text (full paper text extracted via GROBID when PDF available, otherwise abstract)
    • metadata (title, authors, uri/pdf_url, citations, year, venue, etc.)
    • metadata.uri (alias: metadata.pdf_url) may be null for paywalled papers
  • When GROBID is configured and PDF is available, text contains full paper content with section headers. Otherwise, text contains the abstract.
  • Metadata fields (title, authors) are automatically enhanced from GROBID parsing when API values are missing or empty.
{
  "text": "Introduction\nThis paper presents a novel approach...\n\nMethods\nWe propose a transformer architecture...\n\nResults\nOur experiments demonstrate...",
  "format": "paper",
  "metadata": {
    "title": "Attention Is All You Need",
    "authors": ["Ashish Vaswani", "Noam Shazeer", "..."],
    "year": 2017,
    "citations": 75000,
    "venue": "NeurIPS",
    "pdf_url": "https://arxiv.org/pdf/1706.03762.pdf",
    "uri": "https://arxiv.org/pdf/1706.03762.pdf",
    "paper_id": "...",
    "doi": "..."
  },
  "char_count": 45230
}

Configuration

  • Requires SEMANTIC_SCHOLAR_API_KEY environment variable.
  • Requires grobid_url in YAML config (llm_config.grobid) for full text extraction. When GROBID is configured, papers with open PDF URLs are automatically parsed to extract full text and enhance metadata.

Common Workflows

Search and summarize:

{"type":"semantic-scholar","value":"BERT model","out":"$papers"}
{"type":"summarize","target":"$papers","focus":"what is BERT","out":"$summary"}

Get full text (GROBID configured):

{"type":"semantic-scholar","value":"GPT architecture","out":"$papers"}
{"type":"pluck","target":"$papers","field":"text","out":"$full_texts"}

Filter results:

{"type":"semantic-scholar","value":"neural networks","out":"$papers"}
{"type":"filter-collection","target":"$papers","predicate":"citations > 1000","out":"$top_papers"}