scholar-checking

Scholar Checking — Claude Code Plugin

A Claude Code plugin that verifies BibTeX citations against academic databases to catch hallucinated or incorrect references.

What it checks

For each entry in your .bib file:

Field	How it's verified
Title	Fuzzy match against database records
Authors	Author list comparison (family name + initial)
Year	Exact match, flags off-by-one (preprint vs. published)
Venue	Journal/conference name with abbreviation expansion
DOI	Confirms it resolves to the correct paper
arXiv ID	Validates against arXiv API

Databases queried

• CrossRef — DOI registry, best general coverage
• OpenAlex — 240M+ scholarly works, free
• Semantic Scholar — AI-powered academic search
• DBLP — Computer science bibliography
• arXiv — Preprint server

No API keys required. All APIs are queried at their free tier rate limits.

Installation

Add the marketplace, then install the plugin:

/plugin marketplace add kilojoules/scholar-checking-claude-skill
/plugin install scholar-checking@kilojoules-scholar-checking

Or for local development, point Claude Code at a local clone:

/plugin marketplace add /path/to/scholar-checking-claude-skill
/plugin install scholar-checking@kilojoules-scholar-checking

Usage

/check-refs path/to/references.bib

Options

• --fix — Generate corrected BibTeX entries for mismatches
• --context path/to/paper.tex — Cross-reference citation keys against your LaTeX file

Examples

/check-refs references.bib
/check-refs refs.bib --fix
/check-refs refs.bib --context paper.tex

Requirements

Python 3.7+ with:

• bibtexparser — BibTeX parsing
• requests — HTTP client

Install dependencies:

pip install bibtexparser requests

Optional: Semantic Scholar API key

For higher rate limits with Semantic Scholar, set:

export SEMANTIC_SCHOLAR_API_KEY=your_key_here

Get a free key at https://www.semanticscholar.org/product/api

How it works

1. Parses your .bib file and cleans LaTeX formatting
2. For entries with DOI or arXiv ID: does direct lookup (authoritative)
3. For other entries: searches by title across multiple databases
4. Compares metadata (title, authors, year, venue) using fuzzy matching
5. Claude Code reads the results and presents a human-readable report

Non-academic entries (software, websites, blogs) are automatically detected and skipped.

License

MIT