Research Skills
A four-skill Claude Code toolbox for academic work: medical-imaging literature reviews, stylized paper-to-slide images, PhD research proposals, and fidelity-first talk decks.
Last reviewed by the paperbanana team on Jul 13, 2026
Install
git clone https://github.com/luwill/research-skills ~/.claude/skills/research-skills
cd ~/.claude/skills/research-skills
cp -r medical-imaging-review paper-slide-deck research-proposal scholar-slides ~/.claude/skills/
cd ~/.claude/skills/scholar-slides && ./install.shWhat is Research Skills?
Research Skills is a collection of four independent Claude Code skills, each encapsulating a commonly used academic research task rather than one all-in-one pipeline. The repo description sums it up directly: "commonly used research experiences and processes are encapsulated into Agent skills." Instead of chaining every stage of a paper through a single suite, you install only the skill you need — medical-imaging-review for writing literature surveys, paper-slide-deck for stylized shareable slide images, research-proposal for PhD/grant proposal drafting, or scholar-slides for a faithful, editable talk deck.
The medical-imaging-review skill runs an 8-phase write-with-verify workflow (review-type routing, paradigm capture, init, collect-and-verify, outline/taxonomy, per-claim-verified writing, peer review, submission prep) tailored to six imaging domains — CCTA, lung, brain, cardiac, pathology, and retinal imaging — and matches the paper to the right reporting standard (PRISMA 2020, QUADAS-3, CLAIM 2024, or TRIPOD+AI). research-proposal instead targets PhD applications and 研究计划书/开题报告 with a 5-phase requirements → literature → outline → content → output flow and bilingual English/Chinese output.
The two slide-generation skills are deliberately split by purpose rather than merged: paper-slide-deck turns any content into stylized, AI-generated slide images (17 visual styles) meant to be read or shared, where text and numbers are baked into the image and not editable. scholar-slides does the opposite trade-off — vector KaTeX equations, real tables, cited figure crops, and an editable PPTX output — built for a live talk (组会 journal club or a conference/defense) where every number must stay traceable to the source. All four skills share the same underlying principle: citation and claim integrity is enforced with verification steps or an executable audit gate rather than left to the model’s judgment.
Core capabilities
Medical imaging literature reviews with reporting-standard routing
An 8-phase workflow (review-type routing → paradigm capture → init → collect-and-verify → outline/taxonomy → per-claim-verified writing → peer review → submission prep) that matches your review to PRISMA 2020, QUADAS-3, CLAIM 2024, or TRIPOD+AI, backed by an executable audit script (scripts/audit_manuscript.py) with its own fixture test suite.
Stylized paper-to-slide image generation
paper-slide-deck detects figures/tables in a PDF, maps them to a slide outline, and renders 17 visual styles (academic-paper, sketch-notes, minimal, etc.) via the Gemini API (gemini-3-pro-image), with a post-generation garbled-text proofread pass and PPTX/PDF export.
PhD research proposal drafting, bilingual
research-proposal runs a 5-phase requirements → literature → outline → content → output flow, adapts to STEM, humanities, and social-science conventions, and writes in Nature Reviews-style prose in either English or Chinese, targeting 2,000–4,000 words with 25–50 verified references.
Fidelity-first, editable talk decks
scholar-slides renders vector KaTeX equations, real HTML/OOXML tables, and cited figure crops into an editable PPTX (or reveal.js deck and vector PDF), grounding every number against the source and flagging anything unverifiable as [MISSING]/[UNVERIFIED] instead of silently filling it in.
Citation integrity as a shared design principle
Every reference-producing skill verifies citations exist (DOI/PMID/arXiv or Zotero) with matching author and year before they are used, and unverifiable entries are flagged rather than fabricated — the same rule applied across the review, proposal, and scholar-slides skills.
Pick-and-mix installation
Each of the four skills is a self-contained folder you copy independently into ~/.claude/skills/ — you are not required to install the whole suite to use one workflow.
What you can use it for
Writing a systematic review or survey on medical imaging AI
Use medical-imaging-review to route your paper to the right reporting standard (PRISMA/QUADAS-3/CLAIM/TRIPOD+AI) and run the collect-and-verify + per-claim-verified writing phases across domains like CCTA, lung, brain, cardiac, pathology, or retinal imaging.
Turning a finished paper into a shareable visual explainer
Feed a paper PDF into paper-slide-deck to auto-detect figures and generate a stylized image deck (e.g. watercolor or sketch-notes style) for a blog post or social share — not for a results-heavy talk, since text and numbers in the images are not editable.
Drafting a PhD application or grant research proposal
Run research-proposal to gather requirements, pull literature via WebSearch/Zotero/arXiv/PubMed, get an outline approved, then generate a full Nature Reviews-style proposal in English or Chinese with verified references.
Preparing a faithful deck for a journal club or thesis defense
Use scholar-slides when every equation, number, and citation must stay exact and projector-ready — it outputs an editable PPTX plus speaker notes with a bilingual talk-time estimate, in either journal-club or conference register.
Choosing the right slide tool for the occasion
The README explicitly separates the two slide skills by intent: paper-slide-deck for visual, shareable images where look matters more than precision, and scholar-slides for a live talk where equations and data must stay exact — pick based on whether the deck needs to be presented or just read.
How to get started
- 1
Clone the repository
git clone https://github.com/luwill/research-skills ~/.claude/skills/research-skills — this gives you all four skill folders in one place before you copy the ones you need.
- 2
Copy the skill(s) you want into your Claude Code skills directory
From inside the cloned repo, run cp -r medical-imaging-review paper-slide-deck research-proposal scholar-slides ~/.claude/skills/ (or copy just the ones you need) — Claude Code loads skills from ~/.claude/skills/.
- 3
Provision scholar-slides’ toolchain (only if you installed it)
scholar-slides carries Python + Node dependencies. Run cd ~/.claude/skills/scholar-slides && ./install.sh once to set up a .venv, npm packages, and Chromium, plus self-checks. Requires Python 3.11+ and Node 18+; Linux users need sudo apt-get install fonts-noto-cjk for Chinese decks.
- 4
Trigger the skill you need
Use the matching command or natural-language phrase: /medical-imaging-review or "write a survey"; /paper-slide-deck content.md --style watercolor; "research proposal" or "PhD proposal"; or "make slides" / "组会 PPT" / a paper PDF/arXiv/DOI link for scholar-slides.
How it compares to similar skills
Research Skills is a set of four narrow, independently installable skills rather than one chained pipeline. If you want a single suite that carries a paper through research, writing, and review with global integrity gates, or you work in a different discipline, one of these may fit better.
Academic Research Skills
Pick Academic Research Skills if you want one chained research → write → review → revise → finalize pipeline with cross-stage quality gates, instead of four separate point-solutions you trigger individually.
Medical Research Skills
Pick medical-research-skills if you need broader clinical/medical research tooling beyond literature-review writing, since the medical-imaging-review skill here is scoped to writing surveys, not full clinical workflows.
Academic PPTX Skill
Pick academic-pptx-skill if slide generation is your primary need and you want a dedicated PPTX-focused tool rather than choosing between this repo’s two differently-scoped slide skills.