claude-code-research-skills

A curated collection of Claude Code skills and plugins for academic / research paper workflows — figures, writing, polishing, peer-review simulation, literature search, and full research→write→review pipelines.

This is a personal install bundle. Most components are third-party work; original authorship and licenses are preserved. Three skills — paper-polish-pipeline, gpt-image, and nano-banana — are original work written for this collection (paper-polish-pipeline's staged methodology is adapted from a public article — see Attribution & Licenses). visiomaster is intentionally excluded (Windows + Microsoft Visio only — not usable on macOS/Linux).

Contents

skills/ — standalone skills (drop into ~/.claude/skills/ or a project .claude/skills/)

Skill Version Purpose Source License
paper-framework-figure-studio-pro 3.2.15b Human-in-the-loop paper framework/architecture figure workflow (S0–S5) c-narcissus/paper-framework-figure-studio-pro MIT-0
drawio-skill 1.14.0 Generate editable .drawio diagrams, export PNG/SVG/PDF/JPG via draw.io CLI Agents365-ai/drawio-skill MIT
research-paper-writing — ML/CV/NLP paper writing quality (section structure, flow, self-review) Master-cai/Research-Paper-Writing-Skills MIT
paper-polish-pipeline 1.0 Bilingual staged paper-polishing pipeline: diagnose → section optimize → abstract/contributions → language polish (reduce AI-sounding phrasing) → pre-submission/defense risk check, with research-integrity guardrails Original for this repo (methodology adapted from a public article) MIT
gpt-image 1.0 Generate / edit / describe / batch images from the terminal via OpenClaw + OpenAI gpt-image-2 (ChatGPT/Codex subscription); good for figures, diagrams, infographics Original for this repo MIT
nano-banana 1.0 Generate / batch images via Google nano-banana (Gemini image) through the Antigravity CLI (agy); strong in-image text rendering for posters/infographics Original for this repo MIT
autofigure 1.0 Editable vector paper figures (method/architecture/pipeline): OpenAI gpt-image-2 raster → local SAM3 + RMBG-2.0 + OpenAI gpt-5.5 SVG re-draw → editable SVG + margin-free PDF wrapper original for this repo; underlying tool ResearAI/AutoFigure-Edit MIT (wrapper)

plugins/ — Claude Code plugins (install via plugin marketplace)

Plugin Version Skills Source License
nature-skills 1.0.0 10 skills: figure, polishing, writing, reviewer, citation, data, reader, response, paper2ppt, academic-search (MCP) Yuan1z0825/nature-skills MIT
academic-research-skills (ARS) 3.12.0 18 skills/commands; 4 core (deep-research, academic-paper, academic-paper-reviewer, academic-pipeline) + ars-* commands Imbad0202/academic-research-skills (Edward Cheng-I Wu) CC BY-NC 4.0

Install

Standalone skills

cp -R skills/<name> ~/.claude/skills/          # user-level (all projects)
# or into a project: cp -R skills/<name> .claude/skills/

Restart Claude Code so it registers.

Plugins (recommended: original marketplaces)

# nature-skills
claude plugin marketplace add Yuan1z0825/nature-skills
claude plugin install nature-skills@nature-skills

# academic-research-skills (ARS)
claude plugin marketplace add Imbad0202/academic-research-skills
claude plugin install academic-research-skills@academic-research-skills

(The plugins/ copies here are a snapshot for reference/backup.)

macOS dependencies

Component Needs Install
drawio-skill export draw.io desktop CLI brew install --cask drawio
ARS / nature DOCX output Pandoc brew install pandoc
ARS APA-7 PDF compile Tectonic brew install tectonic
CJK PDF (Chinese) Static CJK serif font brew install --cask font-noto-serif-cjk-tc

⚠️ ARS PDF + Chinese: font gotcha

ARS LaTeX templates hardcode \setCJKmainfont{Source Han Serif TC VF}. That variable font crashes Tectonic (findFontByName panic on variable-font .ttc). Before compiling a Chinese PDF, change that line to a static font:

\setCJKmainfont{Noto Serif CJK TC}

Same glyphs (Adobe/Google co-developed). English-only papers don't touch CJK fonts and compile fine.

nature-academic-search MCP server

nature-skills ships an MCP server (plugins/nature-skills/skills/nature-academic-search/mcp-server/) for multi-source literature search (CrossRef + PubMed + arXiv by default; Scopus / ScienceDirect optional via Elsevier pybliometrics config).

Register it at user scope (NOTE: writing ~/.claude/.mcp.json does not work — Claude Code only reads project-root .mcp.json; use claude mcp add):

MCP_DIR="$HOME/.claude/mcp_servers/academic-search"
mkdir -p "$MCP_DIR"
cp -R plugins/nature-skills/skills/nature-academic-search/mcp-server/* "$MCP_DIR/"

claude mcp add academic-search -s user \
  -e [email protected] \
  -- uv run --python 3.13 --no-project --directory "$MCP_DIR" \
  --with 'mcp>=1.0.0,<2.0.0' --with 'requests>=2.28.0,<3.0.0' --with 'toml>=0.10.2,<2.0.0' \
  --with 'lxml>=4.9.0,<6.0.0' --with 'pybliometrics>=4.4.1,<5.0.0' \
  python academic_search_server.py

Requires uv. --python 3.13 is required because mcp needs Python ≥3.10 (don't rely on macOS system Python 3.9). PUBMED_EMAIL is required by NCBI; NCBI_API_KEY is optional for higher rate limits.

Attribution & Licenses

Each subdirectory retains its upstream LICENSE/NOTICE. Notably academic-research-skills is CC BY-NC 4.0 (non-commercial, attribution required) — see plugins/academic-research-skills/LICENSE and NOTICE.md. All other components are MIT / MIT-0. This collection is a non-commercial backup; respect each upstream license for any redistribution or use.