Claude Scholar plugin
2026-03-22 → 2026-04-02
I’ve been gradually building Claude Code skills for academic workflows and packaged them as claude-scholar. These skills may be useful for you.
What’s in it#
Eleven skills (for now). Some can be invoked as a slash command and some are for Claude. Note that skills can be composed. If you create more atomic skills, then other skills or agents can compose them to do their jobs. In other words, principles of software engineering are useful (obviously?).
Tools for Claude (mainly)#
These skills are usually invoked by other skills that I have, which are more customized to my workflow (and thus not part of this plugin). For instance, for my wiki, I have a /wikify skill that uses whatever information I throw at it (e.g., paper URL, paper title, arXiv ID, DOI, …) to find metadata about the paper and create a page in my wiki. These kinds of skills use some of the following skills to get verified metadata about the paper.
/arxiv-metadata— fetch structured metadata for arXiv papers (title, authors, date, categories, DOI). Handles multiple ID formats. I let Claude use this skill to find paper metadata when filing a paper into my wiki or other tasks./doi-bibtex— given a DOI, fetches the BibTeX entry and appends it to the project’s.bibfile, checking for duplicates. I don’t really invoke it myself. Once I have a DOI, Claude Code can download and add the bib entry to your.bibfile./openalex— query 240M+ scholarly works via the OpenAlex API for literature searches, citation analysis, and bibliometric queries. The helper script handles rate limiting, retries, and pagination. No API key required./verify-math— step-by-step verification of mathematical derivations using SymPy. Each derivation step is validated programmatically. Using this skill, Claude can check each step of a mathematical derivation.
Commands that you can invoke#
/check-refs— verify every citation in your LaTeX paper against academic databases, flag suspicious entries, and suggest missing DOIs/URLs. Uses bibsleuth. Can run standalone or as part of/presubmit-checks./latex-cleanup— systematic review for common LaTeX issues: style consistency, typography, cross-references, draft artifacts, my LaTeX pet peeves, etc. Identifies all.texfiles in the project (including\input/\includetargets)./presubmit-checks— pre-submission checklist that orchestrates multiple checks in parallel: references (check-refs), LaTeX cleanup (latex-cleanup), build verification, figure format checks (flags bitmap figures that should be vector), and front matter review (affiliations, acknowledgements, data availability). Presents a unified report organized by severity./arxiv-prep— automates arXiv submission packaging: cleans source files, removes cruft, verifies compilation, extracts metadata for the submission form, and produces a ready-to-upload tarball. Uses Google’s arxiv-latex-cleaner under the hood./critique-manuscript— structured self-review of your own manuscript. Evaluates the paper across seven review criteria (literature, methodology, causal claims, data quality, generalizability, mechanism, clarity) and produces a report with top risks, what to preserve, anticipated reviewer questions, and detailed comments anchored to specific sections. Optionally searches OpenAlex for missing literature. Designed strictly for self-review; declines to review others’ unpublished work and explains why./critique-figures— critiques academic figures using a mix of programmatic checks and vision analysis. Checks file format and resolution, colorblind risk, excessive categories, font legibility, overplotted scatterplots, and dynamite plots. Handles multi-panel figures by evaluating each panel individually. Report-only — does not modify figures./accessible-pdf— transforms a LaTeX document to produce tagged, screen-reader-friendly PDFs (PDF/UA-1, PDF/A-2b). Asks whether to edit in place or create a copy, applies preamble changes, switches to LuaLaTeX, and verifies tagging. See Accessible PDF from LaTeX.
Installation & Usage#
/plugin marketplace add yy/claude-scholar
/plugin install claude-scholar@yy-claude-scholar
/reload-plugins
You may need to restart Claude Code to make it function properly. Please note that these skills are early-stage products and not comprehensively tested, so proceed with caution. And if you use it for your paper, please check your venue’s AI use policy.
And of course, feedback is welcome!
Acknowledgements#
- K-Dense-AI/claude-scientific-skills — inspiration and starting point for the OpenAlex skill
- Sadamori Kojaku — inspiration for the math verification skill
- Trevor Campbell’s arXiv checklist — inspiration for the arXiv prep skill
- Google’s arxiv-latex-cleaner — used by the arXiv prep skill for source cleaning
See also#
- Kaicheng Yang — Claude Code Skills for Research — links several similar efforts