Claude Code Research Skills Bundle
kaelyang935-commits/claude-code-research-skills-bundle
Bundle de skills para pesquisa científica, de busca a submissão
Ver no GitHubPacote vendorizado de 21 skills comunitárias para Claude Code, organizadas em 5 estágios (busca, leitura, experimento, escrita, revisão) e 16 tarefas de pesquisa acadêmica. Inclui ferramentas de busca bibliográfica, análise de dados, escrita de papers e geração de apresentações, com licenças preservadas (MIT + CC-BY-NC-4.0).
Instalação
npx convez-ai kaelyang935-commits/claude-code-research-skills-bundle1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Busca e síntese de literatura acadêmica
- Análise de dados e geração de figuras científicas
- Redação e revisão de artigos científicos
- Criação de apresentações a partir de papers
- Detecção e redução de marcas de IA em textos acadêmicos
Veredito

Bundle bem estruturado para pesquisa acadêmica, com mapeamento claro de tarefas por estágio. Requer atenção às restrições de licença (CC-BY-NC-4.0) em 4 skills para uso comercial; cuidado também com reprodutibilidade via SHA pinned.
README
A tradução para o português deste README ainda está na fila.
🔬 claude-code-research-skills-bundle
按科研任務階段索引的社群 Claude Code 科研 Skill 中文打包工具包 A task-staged, vendored bundle of community Claude Code skills for scientific research
[!IMPORTANT] 本專案為非官方社群整合,與 Anthropic 無從屬關係。每個 skill 均保留其上游 原作者與授權;
skills/_noncommercial/之下為 CC-BY-NC 4.0(禁止商業使用、 不可重新授權)。完整授權對照見 CREDITS.md。This is an unofficial community bundle, not affiliated with Anthropic. Every skill retains its upstream author and license. Skills under
skills/_noncommercial/are CC-BY-NC 4.0 (no commercial use, no relicensing).
📖 繁體中文
目錄
這是什麼
本 repo 把散落在多個社群倉庫的 Claude Code 科研 skill 實際打包(vendor) 進來,
依「檢索 → 閱讀 → 實驗 → 寫作 → 返修」五階段重新編目,方便中文使用者按任務挑選。
每個 skill 都以固定 commit SHA 取得,並原封不動保留其 LICENSE / NOTICE / 署名。
| 範圍 | 授權 |
|---|---|
| 本 repo 原創文件與腳本(README、scripts、docs…) | MIT(LICENSE) |
skills/ 內各 skill | 保留上游授權(CREDITS.md) |
skills/_noncommercial/ | CC-BY-NC 4.0(⚠ 非商業、不可重新授權) |
取得方式可重現:
bash scripts/vendor.sh(依 pinned SHA 重新打包)、bash scripts/check-licenses.sh(輸出授權報告)。
🗺 五階段 16 任務速查表
| # | 任務 | 主 skill | 搭配 skill | 判斷重點 |
|---|---|---|---|---|
| [檢索調研] | ||||
| 01 | 文獻檢索 | nature-academic-search | deep-research / nature-citation | 檢索式、資料庫、多源交叉、引用核驗優先 |
| 02 | 研究熱點匯總 | deep-research | nature-academic-search / nature-reader | 主題聚類、趨勢識別、爭議點、證據空白 |
| 03 | 深度文獻調研 | deep-research | nature-academic-search / academic-pipeline ⚠NC | 系統性檢索、偏倚檢查、證據綜合、完整報告 |
| 04 | 科學假設提出 | deep-research | nature-reviewer / nature-academic-search | 導師式追問 + 審稿式反駁打磨假設 |
| [閱讀匯報] | ||||
| 05 | 文獻閱讀 | nature-reader | pdf(內建) / deep-research review | 全文精讀、中英對照、圖表鄰近解釋 |
| 06 | 文獻匯報 PPT | nature-paper2ppt | nature-reader / nature-figure | 直接產出 PPTX、講稿備註、關鍵圖解讀 |
| [實驗分析] | ||||
| 07 | 組學數據分析 | omics-analysis(待自建) | deep-research / nature-figure | 暫無專門 skill,建議固化 R/Python 生信流程 |
| 08 | 實驗方案設計 | deep-research | nature-academic-search / nature-reviewer | 先設計假設與對照,再查參數並做審稿式壓力測試 |
| 09 | 結果作圖及統計 | nature-figure | manuscript-writing(=ml-paper-writing) / nature-reviewer | 圖的結論、統計、面板邏輯、導出格式 |
| 10 | 實驗結果解讀 | manuscript-writing(=ml-paper-writing) | nature-writing / nature-reviewer | 避免過度解讀,寫清證據層級 |
| [寫作投稿] | ||||
| 11 | 論文寫作 | academic-paper ⚠NC | nature-writing / deep-research | 完整論文流水線 + Nature 風格論證互補 |
| 12 | 論文降低 AI 率 | ai-check(=aigc-detector) | aigc-down-skill / humanize(=humanizer-zh) | 先診斷 AI 痕跡,再做中文學術保量改寫 |
| 13 | 投稿格式調整 | academic-paper ⚠NC | nature-polishing / nature-citation / nature-data | 格式轉換、LaTeX、引用與數據聲明 |
| [返修回復] | ||||
| 14 | 返修意見分析 | nature-response | academic-paper-reviewer ⚠NC / nature-reviewer | 先拆意見、分優先級,再規劃回復與補實驗 |
| 15 | 返修實驗設計 | nature-response | deep-research / nature-academic-search | 圍繞 reviewer 證據缺口設計最小充分實驗 |
| 16 | 回復審稿人 | nature-response | nature-polishing / manuscript-writing(=ml-paper-writing) | 逐點回復、語氣控制、正文一致性 |
[!WARNING] ⚠NC = 該 skill 為 CC-BY-NC 4.0,位於
skills/_noncommercial/,禁止商業使用。 商業情境請改用 MIT 替代:deep-research用research-zh(Weizhena, MIT); 論文寫作/審稿可改以nature-writing+nature-reviewer+ml-paper-writing組合。 名稱對照見 CREDITS.md。
💡 不知道從哪開始? 先用
deep-research把問題拆開(釐清研究問題、列出證據 空白);有明確的論文寫作 / 返修任務時,再進入對應階段的主 skill。
功能特色
檢索 / 調研
nature-academic-search— 多源文獻檢索 + 引文核驗 + 引用檔(.nbib/.ris/.bib)轉換(PubMed/CrossRef/arXiv/Scopus)。research-zh(MIT deep-research) — 中文 deep-research 流程:先生成調研 outline,再對每個 item 開獨立 agent 深調並彙整報告。_noncommercial/deep-research(⚠NC) — 13-agent、7 模式的完整研究流水線(系統性回顧、後設分析、事實查核、蘇格拉底式引導)。
閱讀 / 匯報
nature-reader— 全文中英對照、圖表就位、逐塊保留原文錨點的精讀 Markdown,不退化成純摘要。nature-paper2ppt— 由論文直接產出中文 PPTX(含講稿備註與自審修正迴圈)。
作圖 / 數據
nature-figure— 投稿級 Nature 風格多面板科研繪圖(Python 或 R,先定結論再作圖)。nature-data— Data Availability 聲明、資料庫選擇、FAIR metadata 清單。
寫作 / 潤色 / 引用
nature-writing— 由 claims/結果/圖表起草、重構 Nature 風格各章節。nature-polishing— 中→英學術潤色與 LaTeX 排版修正。nature-citation— 為段落自動補 Nature/CNS 嚴格引用並導出參考管理器格式。ml-paper-writing(= manuscript-writing) — 面向 NeurIPS/ICML/ICLR/ACL 的 ML 論文寫作(LaTeX 模板 + 引用核驗)。_noncommercial/academic-paper(⚠NC) — 12-agent、10 模式完整論文流水線(多文類、多引用格式、Pandoc 導出)。
降 AI 率 / 人工化
aigc-detector(= ai-check) — 學術文本 AI 痕跡偵測、改寫建議、可處理 .docx 並輸出報告。aigc-down-skill— 中文畢業論文降 AIGC 率專項(基於真實降率實驗歸納的模式)。humanizer-zh(= humanize) — 去除文本 AI 生成痕跡,使其更自然(中英皆可)。writing-anti-ai— 基於維基「AI 寫作特徵」清單偵測並修復套路化表述。
審稿 / 返修
nature-reviewer— 以審稿人視角產出 3 份 reviewer report + 交叉綜述。nature-response— Nature 系列返修的逐點回復信草擬/審核/修訂。review-response— 從意見分析到專業 rebuttal 的系統化返修流程。_noncommercial/academic-paper-reviewer(⚠NC) — 模擬 5 位獨立審稿人(EIC+3 peers+Devil's Advocate)。_noncommercial/academic-pipeline(⚠NC) — 串接研究→寫作→誠信檢查→審稿→返修的端到端 orchestrator。
安裝方式
先取得本 repo:
git clone https://github.com/<USER>/claude-code-research-skills-bundle.git
cd claude-code-research-skills-bundle
# 想自行重現打包:bash scripts/vendor.sh
mkdir -p /path/to/project/.claude/skills
cp -R skills/nature-reader /path/to/project/.claude/skills/
mkdir -p ~/.claude/skills
cp -R skills/nature-figure ~/.claude/skills/
[!CAUTION]
skills/_noncommercial/之下為 CC-BY-NC 4.0:僅限非商業用途,不可重新授權。 商業場景請勿安裝該目錄,改用速查表中的 MIT 替代。
詳細安裝與更新流程見 docs/INSTALL.md。
各 Skill 詳細(mode / agent)
research-zh(Weizhena, MIT)— 子 skill
| 子 skill | 用途 |
|---|---|
research | 對話題初步調研,生成調研 outline |
research-deep | 讀 outline,為每個 item 開獨立 agent 深調 |
research-add-fields | 向現有 outline 補充欄位定義 |
research-add-items | 向現有 outline 補充調研對象 |
research-report | 將 deep 結果彙整為 markdown 報告 |
_noncommercial/deep-research(Imbad, ⚠NC)— 7 模式
full research / quick brief / paper review / lit-review / fact-check / Socratic guided / systematic review(+meta-analysis)。13-agent pipeline。
_noncommercial/academic-paper(Imbad, ⚠NC)— 10 模式
full / plan / outline / revision / revision-coach / abstract / lit-review / format-convert / citation-check / disclosure。12-agent pipeline。
_noncommercial/academic-paper-reviewer(Imbad, ⚠NC)— 模擬 5 審稿人
EIC + 3 peer reviewers + Devil's Advocate;full / re-review / quick / methodology / Socratic / calibration。
_noncommercial/academic-pipeline(Imbad, ⚠NC)— 10 階段 orchestrator
research → write → integrity check → review → revise → re-review → re-revise → final integrity check → finalize。
其餘 skill 為單一工作流,觸發條件詳見各自 SKILL.md 的 description。
建議設定
以下為選用進階設定,能提升科研長任務效果,但也增加 token 消耗。
- Agent Teams(多代理協作) —
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1,讓deep-research/academic-pipeline等多 agent skill 並行協作。 - Ralph Loop(自我迭代迴圈) — 對「寫作 → 自審 → 修訂」類任務反覆迭代到收斂;建議設明確停止條件,避免無限迴圈。
- Skip Permissions —
claude --dangerously-skip-permissions。
[!CAUTION] Skip Permissions 安全警告:此旗標會略過所有工具使用的人工確認,Claude 可 未經詢問直接讀寫檔案、執行指令。僅在你完全信任當前工作目錄內容、且於隔離環境 (容器 / 專用帳號 / 無敏感資料)時使用。
💡 token 消耗提醒:多 agent skill 與 Agent Teams / Ralph Loop 會大量消耗 token。 建議探索階段先小範圍跑,確認方向後再放大;長任務優先選 Claude Pro/Max 訂閱或設預算 上限,避免 API 費用失控。
授權條款
| 範圍 | 授權 |
|---|---|
| 本 repo 原創文件 / 腳本 | MIT © 2026 <USER> |
skills/ 下各 vendored skill | 保留上游原始授權(見下) |
-
MIT(可商業使用,須保留授權與署名):nature-skills 全系列(Yuan Yizhe)、claude-scholar 之 nature-response / ml-paper-writing / review-response / writing-anti-ai(Gaorui Zhang)、research-zh(Lan Zheng)、aigc-down-skill(cangtianhuang)、humanizer-zh(歸藏)、aigc-detector(free-revalution)。
-
CC-BY-NC 4.0 ——
skills/_noncommercial/(⚠ 禁止商業使用、不可重新授權):deep-research / academic-paper / academic-paper-reviewer / academic-pipeline,原作者 吳政宜 / Cheng-I Wu。指定署名:Based on Academic Research Skills by Cheng-I Wu https://github.com/Imbad0202/academic-research-skills
完整對照(作者 / URL / pinned SHA / 授權)見 CREDITS.md;摘要見 NOTICE。
重新散布時請保留各 skill 資料夾內的 LICENSE / NOTICE / ATTRIBUTION。
致謝與貢獻
感謝所有上游 skill 作者的開源貢獻(見 CREDITS.md)。本 repo 僅做整合與 中文編目,未修改任何 skill 的內容與授權。
歡迎 issue / PR:回報失效連結、提報新的科研 skill(請附明確授權)、或補上 omics-analysis。
新增 skill 請走 scripts/vendor.sh(pinned SHA)並確保 scripts/check-licenses.sh 通過。
📖 English
Table of Contents
- What is this
- 5 Stages, 16 Tasks
- Skills at a Glance
- Installation
- Configuration Tips
- Licensing
- Credits & Contributing
What is this
This repository vendors community Claude Code skills for scientific research —
scattered across several upstream repos — into one bundle, re-cataloged by the five
stages of a research workflow: Search → Read → Experiment → Write → Revise. Every
skill is pinned to a fixed commit SHA and ships with its original LICENSE /
NOTICE / attribution untouched.
| Scope | License |
|---|---|
| This repo's original files & scripts (README, scripts, docs…) | MIT (LICENSE) |
Each vendored skill under skills/ | Retains upstream license (CREDITS.md) |
skills/_noncommercial/ | CC-BY-NC 4.0 (⚠ non-commercial, no relicensing) |
Reproducible:
bash scripts/vendor.sh(re-vendor by pinned SHA) ·bash scripts/check-licenses.sh(license report).
🗺 5 Stages, 16 Tasks (Cheat Sheet)
| # | Task | Primary skill | Companion skills | What matters |
|---|---|---|---|---|
| [Search & Survey] | ||||
| 01 | Literature search | nature-academic-search | deep-research / nature-citation | Query design, databases, cross-source, citation verification first |
| 02 | Hotspot synthesis | deep-research | nature-academic-search / nature-reader | Topic clustering, trends, controversies, evidence gaps |
| 03 | Deep lit review | deep-research | nature-academic-search / academic-pipeline ⚠NC | Systematic search, bias check, evidence synthesis, full report |
| 04 | Hypothesis generation | deep-research | nature-reviewer / nature-academic-search | Mentor-style probing + reviewer-style rebuttal to sharpen hypotheses |
| [Read & Present] | ||||
| 05 | Paper reading | nature-reader | pdf (built-in) / deep-research review | Full-text close reading, bilingual, figures explained in place |
| 06 | Journal-club PPT | nature-paper2ppt | nature-reader / nature-figure | Direct PPTX, speaker notes, key-figure interpretation |
| [Experiment & Analysis] | ||||
| 07 | Omics data analysis | omics-analysis (TBD) | deep-research / nature-figure | No dedicated skill yet; codify an R/Python bioinformatics pipeline |
| 08 | Experiment design | deep-research | nature-academic-search / nature-reviewer | Hypothesis + controls first, then params, then reviewer stress-test |
| 09 | Figures & statistics | nature-figure | manuscript-writing(=ml-paper-writing) / nature-reviewer | Figure's conclusion, stats, panel logic, export format |
| 10 | Results interpretation | manuscript-writing(=ml-paper-writing) | nature-writing / nature-reviewer | Avoid over-interpretation; state evidence level clearly |
| [Write & Submit] | ||||
| 11 | Paper writing | academic-paper ⚠NC | nature-writing / deep-research | Full pipeline + Nature-style argumentation |
| 12 | Reduce AI score | ai-check(=aigc-detector) | aigc-down-skill / humanize(=humanizer-zh) | Diagnose AI traces first, then content-preserving academic rewrite |
| 13 | Submission formatting | academic-paper ⚠NC | nature-polishing / nature-citation / nature-data | Format conversion, LaTeX, citations & data statement |
| [Revise & Respond] | ||||
| 14 | Review analysis | nature-response | academic-paper-reviewer ⚠NC / nature-reviewer | Decompose comments, prioritize, plan response & added experiments |
| 15 | Revision experiments | nature-response | deep-research / nature-academic-search | Minimal-sufficient experiments around reviewers' evidence gaps |
| 16 | Reviewer response | nature-response | nature-polishing / manuscript-writing(=ml-paper-writing) | Point-by-point reply, tone control, consistency with the manuscript |
[!WARNING] ⚠NC marks CC-BY-NC 4.0 skills under
skills/_noncommercial/— no commercial use. For commercial contexts use MIT alternatives:research-zh(Weizhena) fordeep-research;nature-writing+nature-reviewer+ml-paper-writingfor writing/review. See CREDITS.md for the name mapping.
💡 Not sure where to start? Use
deep-researchto decompose the problem first (clarify the question, list evidence gaps); enter a stage's primary skill only once you have a concrete writing/revision task.
Skills at a Glance
Search / Survey — nature-academic-search (multi-source search + citation verification + .nbib/.ris/.bib conversion) · research-zh (MIT deep-research: outline → per-item deep agents → report) · _noncommercial/deep-research ⚠NC (13-agent, 7-mode research pipeline).
Read / Present — nature-reader (bilingual, figure-aware close reading, not summary-only) · nature-paper2ppt (paper → Chinese PPTX with speaker notes).
Figures / Data — nature-figure (submission-grade Nature-style plots, Python or R) · nature-data (Data Availability statement, repo selection, FAIR metadata).
Write / Polish / Cite — nature-writing (draft Nature-style sections from claims) · nature-polishing (ZH→EN polishing + LaTeX layout) · nature-citation (strict Nature/CNS citations + export) · ml-paper-writing (NeurIPS/ICML/ICLR ML-paper writing) · _noncommercial/academic-paper ⚠NC (12-agent, 10-mode pipeline).
Reduce AI / Humanize — aigc-detector (AI-trace detection + rewrite, .docx, reports) · aigc-down-skill (Chinese thesis AIGC reduction) · humanizer-zh (remove AI traces, ZH/EN) · writing-anti-ai (Wikipedia "Signs of AI writing" patterns).
Review / Revise — nature-reviewer (3 reviewer reports + cross synthesis) · nature-response (point-by-point rebuttal letters) · review-response (analysis-to-rebuttal workflow) · _noncommercial/academic-paper-reviewer ⚠NC (5 simulated reviewers) · _noncommercial/academic-pipeline ⚠NC (end-to-end orchestrator).
Installation
git clone https://github.com/<USER>/claude-code-research-skills-bundle.git
cd claude-code-research-skills-bundle
# Reproduce the vendoring: bash scripts/vendor.sh
- (a) Per-project —
cp -R skills/<skill> /path/to/project/.claude/skills/ - (b) Global —
cp -R skills/<skill> ~/.claude/skills/ - (c) Claude Cowork (desktop) — point Cowork at this repo folder; Claude reads each
SKILL.mdand triggers automatically. - (d) claude.ai Project — upload each skill's
SKILL.mdonly. Single-file limit ≈ 200KB; putversion/last_updatedundermetadata:in the frontmatter.
[!CAUTION] Skills under
skills/_noncommercial/are CC-BY-NC 4.0: non-commercial use only, no relicensing. Do not install that directory in commercial settings — use the MIT alternatives instead. Full guide: docs/INSTALL.md.
Configuration Tips
Optional but powerful for long research tasks (at higher token cost):
- Agent Teams —
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1for parallel multi-agent skills. - Ralph Loop — iterate "write → self-review → revise" to convergence; set an explicit stop condition.
- Skip Permissions —
claude --dangerously-skip-permissions.
[!CAUTION] Skip Permissions warning: this flag bypasses all tool-use confirmations — Claude can read/write files and run commands without asking. Use only when you fully trust the working directory and run in an isolated environment (container / dedicated account / no sensitive data).
💡 Token note: multi-agent skills and Agent Teams / Ralph Loop consume many tokens. Start small, scale once the direction is confirmed, and prefer a Claude Pro/Max plan or a budget cap for long runs.
Licensing
| Scope | License |
|---|---|
| This repo's original files / scripts | MIT © 2026 <USER> |
Vendored skills under skills/ | Retain upstream license |
-
MIT (commercial use OK, keep license & attribution): all nature-skills (Yuan Yizhe); claude-scholar's nature-response / ml-paper-writing / review-response / writing-anti-ai (Gaorui Zhang); research-zh (Lan Zheng); aigc-down-skill (cangtianhuang); humanizer-zh (歸藏); aigc-detector (free-revalution).
-
CC-BY-NC 4.0 —
skills/_noncommercial/(⚠ no commercial use, no relicensing): deep-research / academic-paper / academic-paper-reviewer / academic-pipeline, by 吳政宜 / Cheng-I Wu. Required attribution:Based on Academic Research Skills by Cheng-I Wu https://github.com/Imbad0202/academic-research-skills
See CREDITS.md for the full mapping (author / URL / pinned SHA / license) and
NOTICE for a summary. When redistributing, keep each skill folder's LICENSE /
NOTICE / ATTRIBUTION.
Credits & Contributing
Thanks to all upstream skill authors (see CREDITS.md). This repo only does integration and Chinese cataloging — no skill content or license is modified.
Issues/PRs welcome: report broken links, propose new research skills (with a clear license),
or contribute omics-analysis. Add skills via scripts/vendor.sh (pinned SHA) and ensure
scripts/check-licenses.sh passes.