Academic Claude Skills · 學術研究 Claude 技能包

🌐 Language / 語言：繁體中文 · English

demo — Academic Claude Skills 研究流程示範(讀文獻/因果識別/投稿前稽核)

一組給商管、財金、社會科學研究者的 Claude Skills：從選題、找資料、讀文獻，到分析、寫作、投稿與複製包封存——量化／質化／實驗／混合方法全典範，涵蓋研究全流程。站在多位開源作者的肩膀上（見致謝）。
A curated bundle of Claude Skills for research in business, finance, and the social sciences — from framing a question, finding data and reading the literature, through analysis, writing, submission, and replication-package archiving. Quantitative, qualitative, experimental, and mixed methods. Built on the shoulders of several open-source authors (see Credits).

繁體中文

📌 這是什麼 / 給誰用

一組可掛到 Claude 上的「技能（Skills）」。掛上後，當你在對話中談到相關任務，Claude 會自動載入對應技能、套用該領域的專業框架與紀律，不必每次重打長提示詞。

適合對象：商管／財金／社會科學的研究者（碩、博士生與研究人員）。原本依台灣學術脈絡設計（繁體中文、APA 7、口試與審查意見回覆格式），但多數技能在任何脈絡都適用。方法典範不限量化——質化、實驗、混合方法均有對應技能。

這個包在意什麼：不只把答案生出來，更在意答案站不站得住。數字要能指回出處、引用要能查證真偽、識別假設要先講清楚、不確定就標示不確定。技能寧可回報「查無」，也不生一個看起來合理的東西——因為看起來合理的錯誤最難被發現。

✨ 為什麼是這個包（而不是一堆零散提示詞）

腳本會真的跑，不只是給指引。 資料與圖表類技能附實測過的 Python／R 腳本——你電腦裝了 Python 或 R，就能實際連線抓書目、清資料、建資料表、出圖（例：literature-matrix-builder 連 CrossRef 抓 DOI 直接建 Excel 比較矩陣；r-spss-syntax-architect 產出可貼上就跑的 R／SPSS／Stata 語法；management-figure 直出 300dpi 出版級圖）。
不只本國資料，也做跨國。 除了台灣公開資料，global-opendata-scout 內建 World Bank／Eurostat／ILOSTAT／IMF／UN Data 的免金鑰、已實測端點與撈取腳本，並主動提醒跨國比較的陷阱（國家代碼三套不可混、幣別／PPP／基期、產業分類不可直接對應），讓你做他國與跨國分析不被本國資料綁死。
反幻覺，答案要站得住。 數字能指回出處、引用能查真偽、CrossRef 查無就報錯不憑記憶補書目——寧可回報查無，也不生一個看似合理的錯誤。
一條龍、由「研究大腦」分派。 從方法選擇→找資料→清理→分析→寫作→投稿→複製包封存，research-orchestrator 會判斷你在哪一階段、該叫哪幾支、什麼順序，不必自己記 35 支各做什麼。
四大方法範式齊全，還能交複製包。 量化／質化／實驗／混合都有對應技能；reproducibility-architect 直接把研究打包成可重跑的複製包（環境鎖定、授權資料的可重現困境、資料／程式碼／AI 使用三聲明、DOI 封存），對接頂刊資料編輯要求。

⭐ 給顆星 = 我們的 impact factor

如果這裡有哪支技能幫你擋掉一個幻覺引用、省下一下午的洗資料、或讓 reviewer 2 沒那麼可怕——按顆 ⭐ 就是學術版的「被引用」：下一個需要的研究者靠它找到這裡。不用送審、不用 revise & resubmit。就當它是開源研究工具版的「按讚、訂閱、開小鈴鐺」。 🔔

⚠️ 本專案與 Anthropic 無官方關聯。部分技能需搭配外部工具或**付費資料庫（如 TEJ）**才能發揮完整功能。

🆕 最新更新（v0.11.0 · 2026-08）

近期把重心放在資料涵蓋的廣度（跨國）、取得的穩健度與文件前處理——研究流程最前段、也最容易「髒進髒出」的幾塊。

v0.11.0 — 跨國／國際比較資料（global-opendata-scout）。 研究要做他國或跨國比較時，內建 World Bank／Eurostat／ILOSTAT／IMF／UN Data 的免金鑰、已實測端點與撈取腳本，另附「如何找到任一國家官方統計機構」的五步方法論。核心價值不在找到數字，而在主動點出跨國資料的可比性陷阱（國家代碼三套、幣別／PPP／基期、會計年度、產業分類 ISIC／NACE／NAICS 不可直接對應、涵蓋率遺漏造成選擇偏誤）。台灣資料不涵蓋（World Bank／OECD 皆無台灣），走官方來源。
v0.10.0 — 複雜揭露文件前處理（text-analytics-architect）。 語料若是版面複雜的揭露文件（10-K、年報、ESG 永續報告、掃描檔），pdftotext 直抽會把表格壓平、多欄交錯、頁尾混進正文——髒進髒出，污染後面的斷詞與情緒分析。新增 Step 0 指引：先用版面感知抽取轉成保留語意結構的乾淨文字（並驗抽取品質、保 source map、遮罩 PII）再分析。融合 KingsleyOWO/Semark（Apache 2.0）的語意化文件處理概念，只取概念不取依賴。
v0.9.0 — 動態網站抓取升級階梯（public-disclosure-scout）。 requests 抓不到 JS 動態頁時的由輕到重升級階梯（先找背後 API → 官方批次 → 無頭瀏覽器）。融合 unclecode/crawl4ai（Apache 2.0）「把網頁轉成 LLM 可讀結構」概念，只取概念不取依賴，並明訂 🚫 禁用其 stealth 反偵測功能——學術研究抓不到就升級到官方管道／申請／人工，不是躲過偵測。
v0.8.0 — 讀文獻線（literature-matrix-builder + bilingual-paper-reader）。 把幾十篇 PDF 變成有結構的東西:文獻比較矩陣(PDF→DOI→CrossRef 免金鑰→APA 7→20 欄 Excel,CrossRef 查無 DOI 直接報錯不憑記憶補書目),與單篇論文雙欄精讀(逐段中譯+五色標記+離線閱讀器)。兩支天然接力,精讀結論回填矩陣綜整欄。

一貫紀律不變:融合外部開源都只取概念、不吞依賴、遇 copyleft 就規避、用了就在 NOTICE.md 與各技能 ATTRIBUTION.md 誠實致謝。

完整版本歷史見各 Releases。技能總數：35。
安裝:整個 repo 的 ZIP 無法直接當單一技能上傳;請到 dist/ 下載你要的個別 .zip(見下方安裝說明)。

🧭 運作原則（三條底線）

資料由你自己抓。 資料類技能一律假設你（或你的機構）擁有合法訂閱／授權帳號，由你自己登入下載。技能只教「在哪找、怎麼判斷、怎麼分析」，不代抓資料、不散布任何資料庫的專屬目錄。
教方法，重識別假設。 找到資料後，技能會建議最恰當的研究設計與估計方法（先講清楚識別假設會不會成立，再談跑哪個模型），並產出可重現的 R／SPSS／Stata 語法。
資料庫中立、歡迎擴充。 框架適用於任何資料庫（TEJ 只是內建範例）。歡迎社群依 docs/ADD_A_DATABASE.md 新增 WRDS/Compustat、CSMAR、World Bank 等 profile。

🗂 技能總覽（依研究流程分組）

研究階段	技能	一句話功能	觸發詞（示例）	授權
① 選題・文獻・資料	`research-orchestrator`	研究大腦總管，替你分派合適的子技能	不知道從何著手、綜合任務	原創
	`research-method-selector`	方法論適配：判量化/質化/實驗/混合＋Q1 過程套模；含新手小白引導模式	該用什麼方法、不知道要研究什麼	原創
	`phd-researcher`	文獻分析、方法論逆向、研究缺口、PRISMA、後設分析	文獻分析、系統性回顧、meta-analysis	混合 🔒
	`literature-matrix-builder`	文獻語料庫與比較矩陣：PDF→DOI→CrossRef→APA 7→Excel 橫向比較表	文獻矩陣、文獻整理、建文獻庫、APA7、DOI 查書目	原創
	`bilingual-paper-reader`	單篇論文雙欄精讀：逐段中譯＋五色預先標記＋離線閱讀器（標記可持久化/匯出）	精讀論文、論文翻譯、中英對照、畫重點做筆記	原創
	`tej-data-scout`	資料可行性偵察＋研究設計/估計方法建議（以 TEJ 為範例，資料庫中立）	選題、資料可行性、這題能不能做	原創
	`public-disclosure-scout`	免費官方公開揭露偵察（MOPS 重大訊息/年報/裁罰）＋事件研究事件源整備	公開資訊觀測站、MOPS、免費資料、事件源	原創
	`global-opendata-scout`	跨國/國際公開統計偵察：World Bank/Eurostat/ILOSTAT/IMF/UN Data 免金鑰端點＋撈取腳本＋可比性陷阱	跨國、國際比較、World Bank、他國統計、cross-country	原創
	`multi-source-data-integrator`	多源嚴謹結合：實體解析、跨源值調解、來源譜系、三角驗證、合併損耗對帳	多源結合、跨源、統編對接、三角驗證	原創
	`tej-variable-mapper`	把 Compustat／CRSP 變數定義映射到 TEJ 對應欄位	變數對應、TEJ 欄位、Compustat	原創
	`tej-data-wrangler`	TEJ 原始 Excel/CSV 清理、遺漏值分析、格式標準化	資料清理、遺漏值、格式標準化	原創
② 分析・語法・量表	`survey-research-architect`	問卷研究全流程：設計、先驗檢定力抽樣、發放回收、CMV 攻防	問卷設計、樣本數、CMV	原創
	`interview-method-designer`	深度訪談設計：三層大綱、理論抽樣與飽和、倫理知情同意	訪談大綱、訪幾個人、飽和	原創
	`experiment-design-architect`	實驗設計：組間/組內、counterbalancing、情境實驗、操弄檢核	實驗設計、vignette、操弄檢核	原創
	`r-spss-syntax-architect`	依假說生成可重現的 R／SPSS／Stata 語法（含 SEM/PLS 軌）	R 語法、SPSS 語法、PLS-SEM	原創
	`causal-inference-architect`	因果識別策略：現代交錯 DiD、IV、RDD、合成控制、事件研究圖	DiD、內生性、識別策略	原創
	`text-analytics-architect`	文字資料變研究變數：主題模型、情緒語調、LLM 標註信效度	文字探勘、LDA、LLM 標註	原創
	`ob-hrm-scale-adaptor`	量表跨文化改編（合規版）＋測量恆等性檢定語法	量表改編、測量恆等性、lavaan	原創
	`qualitative-thematic-coder`	Braun & Clarke 主題分析，深度訪談逐字稿編碼	主題分析、逐字稿編碼、質性	原創
	`management-figure`	出版級統計圖：倒 U 轉折點、係數森林圖、交互作用、邊際效果	出版級圖、forest plot、交互作用圖	📎 MIT
③ 寫作・潤飾	`academic-journal-polisher`	台灣學術環境文句潤飾，杜絕 AI 慣用語	潤飾、去 AI 味、學術文句	原創
	`q1-journal-polisher`	Q1–Q4 國際期刊英文潤飾＋APA 7＋模擬審查	期刊潤飾、投稿前、APA 7	原創
	`nstc-grant-writer`	國科會計畫申請書寫作：結構化撰寫＋審查人視角自評	國科會、NSTC、計畫書、預期成果	原創
	`response-letter-craftsman`	投稿修訂（R&R）逐點回覆信＋Response to Reviewers	審查回覆、response letter、R&R、逐點回覆	原創
④ 投稿前品管	`thesis-consistency-audit`	六維度一致性稽核：假設↔迴歸表、樣本數、文字↔表格、引用	一致性稽核、論文對帳、投稿前檢查	🔒 CC BY-NC-SA
	`reproducibility-architect`	複製包架構：可重現專案結構、環境鎖定、授權資料可重現困境、資料/程式碼/AI 使用聲明、DOI 封存	複製包、可重現、資料可用性聲明、AI 使用揭露	原創
	`q1-journal-reviewer`	模擬 ABS 3/4 匿名審查委員的批判	審稿、peer review、審查意見	原創
	`citation-verifier`	揪出 AI 捏造假文獻、孤兒引用、引用不貼合主張	引用查核、假文獻、孤兒引用	原創
⑤ 口試・簡報	`academic-pptx`	學術簡報內容與結構標準：行動式標題、論證式編排	口試簡報、conference talk	📎 MIT
	`academic-slides`	Beamer 風格、零依賴單檔 HTML 學術簡報（含 KaTeX）	學術投影片、Beamer、HTML 簡報	📎 MIT
	`academic-deck-animator`	簡報動畫引擎：HTML+Canvas 粒子／原生 PPTX 進場動畫、逐步揭示	PPT動畫、粒子特效、動態簡報、進場動畫	原創
	`academic-poster`	研討會學術海報：A0/A1、傳統三欄或 Better Poster 版式	海報、poster、A0、研討會海報	原創
	`defense-qa-coach`	口試答辯教練：委員提問題庫＋擬答框架＋模擬攻防	口試、答辯、模擬口試、追問	原創
⑥ 博士修業（模板）	`phd-milestone-tracker`	博士修業里程碑與 deadline 追蹤（規則為範例，請換成你系所的）	修業里程碑、資格考期限、畢業時程	原創
	`qual-exam-coach`	學科資格考備考教練（可依你的考科調整）	資格考、備考、記憶卡、模擬考	原創

授權圖例：原創＝作者本人著作（採根 LICENSE）；📎＝改作／收錄他人開源作品（保留原授權與姓名標示）；🔒＝含非商業（NC）條款。詳見 NOTICE.md。

關於引用查核：另有優秀開源工具 PHY041/claude-skill-citation-checker（比對 CrossRef／Semantic Scholar／OpenAlex）。因其上游未附授權（預設保留一切權利），本 repo 不重製其程式碼，建議直接前往取用，與本包 citation-verifier 搭配。

🚀 安裝

⚠️ 不能把整個 repo 的 ZIP(Code → Download ZIP)當成一個技能上傳——那裡面有 35 個 SKILL.md、深層巢狀,Claude 一次只吃一個技能。也不能只給網址讓 claude.ai 自己抓;技能上傳是「上傳檔案」。請用個別的 .zip。

claude.ai(逐支安裝)

到 dist/ 點你要的技能 .zip(例 literature-matrix-builder.zip)→ 右側 Download。
claude.ai → 頭像 → Settings → Capabilities,開啟 Code execution(含腳本的技能需要)。
Settings → Skills → Add / Upload → 選剛下載的 .zip → 完成。要幾支重複幾次(平台一次一支)。上傳後跨裝置自動同步。

Claude Code:clone 後把 skills/<名稱>/ 資料夾複製到 ~/.claude/skills/(不需打包)。詳見 dist/README.md 與 docs/INSTALL.md。

其他 AI 代理(Codex 等):技能的「內容」(SKILL.md 指引+scripts+references)是可攜的 markdown 與 Python;技能的「自動載入」機制是 Claude 專屬。在 Codex/其他代理要用,見 docs/USE_WITH_OTHER_AGENTS.md。

💡 使用範例

範例 1 — 發想期問「這題能不能做」＋方法建議

你：我想做「董事會性別多元化對創新產出（專利數）的影響」，TEJ 做得起來嗎？
Claude（tej-data-scout）：拆構念、給「變數 × 資料對照表」，標示董監資料【直接】、專利數【外部】（建議 TIPO）；判定為公司-年 panel，因專利為計數且過度離散 → 建議負二項固定效果；再交棒 /r-spss-syntax-architect 產語法。

範例 2 — 投稿前一致性稽核

你：這是我的迴歸章節（.docx），投稿前幫我抓內部矛盾。
Claude（thesis-consistency-audit）：逐一比對假設↔迴歸表、樣本數、文字↔表格數字、APA 引用，列出會被審查委員圈起來的矛盾點與修正建議。

範例 3 — 模擬 Q1 期刊審查

你：用 ABS 3* 審查委員的角度狠一點審這篇 introduction 與 hypotheses。
Claude（q1-journal-reviewer）：以匿名審查口吻指出理論貢獻、假設推導斷點、identification 威脅，並給 Major/Minor Revision 的具體要求。

📜 授權

本 repo 是合輯（collection），採逐資料夾授權：原創技能採根 LICENSE；改作／收錄他人作品的技能，各資料夾內保留原作者 LICENSE 與 ATTRIBUTION.md。完整對照見 NOTICE.md。🔒 標示 Non-Commercial 的技能僅供非商業學術用途。

🙏 致謝

Zara Zhang — academic-slides（MIT）
Yuan Yizhe（Yuan1z0825） — nature-skills，management-figure 的出版級繪圖底子（MIT）
Cheng-I Wu（Imbad0202） — academic-research-skills，phd-researcher 的系統性回顧／後設分析模組（CC BY-NC 4.0）
qqfly1to19 — awesome_proofreading_auto，thesis-consistency-audit 的稽核框架理念（CC BY-NC-SA 4.0）
PHY041 — claude-skill-citation-checker，推薦搭配的引用查核工具（本 repo 未重製其程式碼）
academic-pptx 內容準則參考 Barbara Minto《Pyramid Principle》、Naegle (2021) PLOS Comput Biol 等公開學術實務。

若你是上述任一作者、對收錄或標示有任何意見，歡迎開 issue，我會立即配合調整。

🤝 貢獻 / ⚖️ 免責

歡迎 issue 與 PR，特別是新增資料庫 profile（見 docs/ADD_A_DATABASE.md）。請先讀 CONTRIBUTING.md。本專案與 Anthropic 無官方關聯；技能輸出僅供研究輔助，最終學術判斷與責任在使用者本人。phd-milestone-tracker／qual-exam-coach 內的規則為範例模板，務必以你所屬系所公告為準。部分技能需付費資料庫（TEJ）。

English

📌 What it is / Who it's for

A bundle of Claude Skills covering the full research workflow. Once installed, Claude auto-loads the relevant skill when your conversation touches a matching task, applying that field's framework and discipline — no need to re-type long prompts.

Audience: researchers in business, finance, and the social sciences (master's/PhD students and faculty). Built around the Taiwanese academic context (Traditional Chinese, APA 7, oral-defense and reviewer-response conventions), though most skills work anywhere. Not limited to quantitative work — qualitative, experimental, and mixed-methods skills are included.

What this bundle optimizes for: not just producing an answer, but whether the answer holds up. Numbers should trace back to a source; citations should be checkable; identification assumptions get stated before models get run; uncertainty gets labelled as uncertainty. These skills would rather report "not found" than return something that merely looks plausible — because a plausible-looking error is the hardest kind to catch.

✨ Why this bundle (not a pile of loose prompts)

The scripts actually run — not just guidance. Data and figure skills ship tested Python/R scripts: if your machine has Python or R, they genuinely fetch bibliographic data, clean data, build tables and render figures (e.g. literature-matrix-builder pulls DOIs from CrossRef straight into an Excel comparison matrix; r-spss-syntax-architect emits paste-and-run R/SPSS/Stata syntax; management-figure outputs 300 dpi publication-grade figures).
Not just domestic data — cross-country too. Beyond Taiwan's public data, global-opendata-scout ships key-free, tested endpoints and fetch scripts for World Bank / Eurostat / ILOSTAT / IMF / UN Data, and proactively flags cross-country comparability traps (three country-code schemes that mustn't be mixed, currency/PPP/base-year, non-mappable industry classifications) — so international and cross-country analysis isn't tied to home-country data.
Anti-hallucination — answers that hold up. Numbers trace back to a source; citations are checkable; a CrossRef miss fails loudly instead of filling a bibliography from memory — it would rather report "not found" than return a plausible-looking error.
End-to-end, dispatched by a "research brain." From method selection → finding data → cleaning → analysis → writing → submission → replication-package archiving, research-orchestrator works out which stage you're in and which skills to call in what order — you don't have to memorize what all 35 do.
All four method paradigms, plus a replication package. Quantitative, qualitative, experimental and mixed methods are all covered; reproducibility-architect packages the study into a re-runnable replication bundle (environment locking, restricted-data reproducibility, data/code/AI-use statements, DOI archiving) that meets top-journal data-editor requirements.

⭐ A star is our impact factor

If a skill here caught a hallucinated citation, saved you an afternoon of data-wrangling, or made Reviewer 2 a little less terrifying — a ⭐ is the academic equivalent of a citation: it's how the next researcher finds this. No peer review, no revise-and-resubmit required. Think of it as the "like, subscribe, and hit the bell" of open-source research tools. 🔔

⚠️ Not affiliated with Anthropic. Some skills require external tools or a paid database (e.g., TEJ) for full functionality.

🆕 What's new (v0.11.0 · 2026-08)

Recent work focused on data coverage (cross-country), data-acquisition robustness, and document preprocessing — the earliest stages of research, and the ones most prone to "garbage in, garbage out."

v0.11.0 — cross-country / international-comparison data (global-opendata-scout). For studies that need other-country or cross-country comparisons, it ships key-free, tested endpoints and fetch scripts for World Bank / Eurostat / ILOSTAT / IMF / UN Data, plus a five-step method for locating any country's official statistics office. Its core value isn't finding numbers but proactively flagging cross-country comparability traps (three country-code schemes, currency/PPP/base-year, fiscal-year differences, non-mappable industry classifications ISIC/NACE/NAICS, coverage gaps causing selection bias). Taiwan isn't covered (neither World Bank nor OECD includes it) — use official domestic sources.
v0.10.0 — complex-disclosure preprocessing (text-analytics-architect). When the corpus is layout-heavy disclosure (10-Ks, annual reports, ESG/sustainability reports, scans), naive pdftotext flattens tables, scrambles columns and mixes footers into the body — garbage in, garbage out, poisoning downstream tokenization and sentiment. A new Step 0 says: run layout-aware extraction into clean, structure-preserving text first (verify extraction quality, keep a source map, mask PII), then analyze. Adopts the concept from KingsleyOWO/Semark (Apache 2.0) — concept only, no dependency.
v0.9.0 — dynamic-scraping escalation ladder (public-disclosure-scout). When requests can't reach a JS-rendered page, a lightest-to-heaviest escalation ladder (find the underlying API first → official batch files → headless browser only if needed). Adopts the "turn web pages into LLM-readable structure" concept from unclecode/crawl4ai (Apache 2.0) — concept only, no dependency — and explicitly forbids its stealth / anti-detection features: scholarly scraping that can't reach a page escalates to official channels / a request / manual work, not to evading detection.
v0.8.0 — the literature-reading line (literature-matrix-builder + bilingual-paper-reader). Turn dozens of PDFs into something structured: a comparison matrix (PDF→DOI→CrossRef, no key→APA 7→20-column Excel; CrossRef miss fails loudly, never fills bibliography from memory) and side-by-side close reading of a single paper (paragraph translation + five-colour marking + offline reader). The two hand off naturally.

The standing discipline holds: external open source is adopted as concept only, never as a bundled dependency; copyleft is avoided; and every borrowing is credited honestly in NOTICE.md and each skill's ATTRIBUTION.md.

Full version history in Releases. Total skills: 35.
Install: the whole-repo ZIP is not a single installable skill — grab the individual .zip you want from dist/ (see Install below).

🧭 Operating principles (three ground rules)

You fetch your own data. Every data skill assumes you (or your institution) hold a legitimate subscription/license and download the data yourself. Skills only teach where to look, how to judge feasibility, and how to analyze — they never fetch data for you and never redistribute any database's proprietary catalog.
Teach method, foreground identification. After data is located, skills recommend the appropriate research design and estimator (is the identifying assumption plausible first, model choice second) and generate reproducible R/SPSS/Stata syntax.
Database-agnostic and extensible. The framework applies to any database (TEJ is just the built-in example). Contribute profiles for WRDS/Compustat, CSMAR, World Bank, etc. via docs/ADD_A_DATABASE.md.

🗂 Skills overview (grouped by research stage)

Stage	Skill	What it does	License
① Ideation・Literature・Data	`research-orchestrator`	A router "research brain" that dispatches the right sub-skill	Original
	`research-method-selector`	Methodological-fit advisor (quant/qual/experiment/mixed) + Q1 process templates + beginner guidance mode	Original
	`phd-researcher`	Literature analysis, methodology reverse-engineering, research gaps, PRISMA systematic reviews, meta-analysis	Mixed 🔒
	`literature-matrix-builder`	Literature corpus + comparison matrix: PDF→DOI→CrossRef→APA 7→Excel synthesis table	Original
	`bilingual-paper-reader`	Side-by-side close reading of one paper: paragraph-aligned translation + five-colour pre-marking + offline reader (persistent highlights, note export)	Original
	`tej-data-scout`	Data-feasibility scouting + research-design/estimator advice (TEJ as example; database-agnostic)	Original
	`public-disclosure-scout`	Free official public-disclosure scouting (MOPS filings/annual reports/sanctions) + event-study event source	Original
	`global-opendata-scout`	Cross-country/international open-stats scout: key-free World Bank/Eurostat/ILOSTAT/IMF/UN Data endpoints + fetch scripts + comparability traps	Original
	`multi-source-data-integrator`	Rigorous multi-source integration: entity resolution, cross-source reconciliation, data lineage, triangulation, merge-loss accounting	Original
	`tej-variable-mapper`	Maps Compustat/CRSP variable definitions to TEJ fields	Original
	`tej-data-wrangler`	Cleans raw TEJ Excel/CSV: missing values, outliers, formatting	Original
② Analysis・Syntax・Scales	`survey-research-architect`	End-to-end survey research: design, a-priori power & sampling, fielding plan, CMV defenses	Original
	`interview-method-designer`	In-depth interview design: layered protocols, theoretical sampling & saturation, ethics	Original
	`experiment-design-architect`	Experimental design: between/within, counterbalancing, vignettes, manipulation checks	Original
	`r-spss-syntax-architect`	Generates reproducible R/SPSS/Stata syntax from hypotheses (incl. SEM/PLS lane)	Original
	`causal-inference-architect`	Causal identification: modern staggered DiD, IV, RDD, synthetic control, event-study plots	Original
	`text-analytics-architect`	Text-as-data: topic models, tone/sentiment, LLM-annotation validity discipline	Original
	`reproducibility-architect`	Replication packages: reproducible project structure, environment locking, restricted-data reproducibility, data/code/AI-use statements, DOI archiving	Original
	`ob-hrm-scale-adaptor`	Cross-cultural scale adaptation (copyright-compliant) + measurement-invariance syntax	Original
	`qualitative-thematic-coder`	Braun & Clarke thematic analysis for interview transcripts	Original
	`management-figure`	Publication-grade figures: inverted-U turning points, coefficient forest plots, interaction & marginal-effects plots	📎 MIT
③ Writing・Polishing	`academic-journal-polisher`	Prose polishing for Taiwanese academic writing; removes AI-tells	Original
	`q1-journal-polisher`	Q1–Q4 journal English polishing + APA 7 + mock review	Original
	`nstc-grant-writer`	NSTC (Taiwan) grant-proposal writing + reviewer-lens self-assessment	Original
	`response-letter-craftsman`	Point-by-point R&R response letters + Response to Reviewers	Original
④ Pre-submission QA	`thesis-consistency-audit`	Six-dimension consistency audit: hypotheses↔tables, sample sizes, text↔tables, citations	🔒 CC BY-NC-SA
	`q1-journal-reviewer`	Simulates an ABS 3/4 anonymous reviewer's critique	Original
	`citation-verifier`	Catches AI-fabricated references, orphan citations, claim–citation mismatches	Original
⑤ Defense・Slides	`academic-pptx`	Academic slide content & structure standards: action titles, argument-driven decks	📎 MIT
	`academic-slides`	Beamer-style, zero-dependency single-file HTML academic slides (KaTeX)	📎 MIT
	`academic-deck-animator`	Presentation animation engine: HTML+Canvas particles / native-PPTX entrance animations	Original
	`academic-poster`	Conference posters: A0/A1, classic 3-column or Better-Poster layouts	Original
	`defense-qa-coach`	Defense Q&A coach: committee question bank + answer frameworks + mock drills	Original
⑥ PhD milestones (templates)	`phd-milestone-tracker`	PhD milestone & deadline tracker (rules are a template — replace with your program's)	Original
	`qual-exam-coach`	Qualifying-exam prep coach (adaptable to your subjects)	Original

Legend: Original = authored by this project (root LICENSE); 📎 = adapted from / includes third-party open-source work (original license & attribution preserved in-folder); 🔒 = includes a Non-Commercial (NC) clause. See NOTICE.md.

On citation checking: there's an excellent open-source tool, PHY041/claude-skill-citation-checker (checks CrossRef/Semantic Scholar/OpenAlex). Because its upstream ships no license (all rights reserved), this repo does not reproduce its code — please get it from the original repo and pair it with this bundle's citation-verifier.

🚀 Install

⚠️ The whole-repo ZIP (Code → Download ZIP) is not a single installable skill — it holds 35 nested SKILL.md files, and Claude ingests one skill at a time. You also can't just give claude.ai a URL; skill upload is a file upload. Use the individual .zip files.

claude.ai (one skill at a time)

In dist/, click the skill .zip you want (e.g. literature-matrix-builder.zip) → Download on the right.
claude.ai → avatar → Settings → Capabilities, enable Code execution (needed for skills with scripts).
Settings → Skills → Add / Upload → pick the downloaded .zip → done. Repeat per skill (the platform takes one at a time). Uploads sync across devices.

Claude Code: clone, then copy the skills/<name>/ folder into ~/.claude/skills/ (no packaging needed). See dist/README.md and docs/INSTALL.md.

Other AI agents (Codex, etc.): a skill's content (SKILL.md guidance + scripts + references) is portable Markdown and Python; the auto-loading mechanism is Claude-specific. To use these with Codex or another agent, see docs/USE_WITH_OTHER_AGENTS.md.

💡 Usage examples

1 — "Is this topic feasible?" + method advice

You: I want to study board gender diversity → innovation output (patent counts). Can TEJ support this?
Claude (tej-data-scout): decomposes the constructs, gives a variable×data table (board data = direct; patent counts = external, suggest TIPO), identifies it as a firm-year panel, and — since patents are over-dispersed counts — recommends a negative-binomial fixed-effects model, then hands off to /r-spss-syntax-architect for runnable syntax.

2 — Pre-submission consistency audit

You: Here's my results chapter (.docx). Find internal contradictions before I submit.
Claude (thesis-consistency-audit): cross-checks hypotheses↔tables, sample-size flow, text↔table numbers, and APA citations; lists the contradictions a reviewer would circle, with fixes.

3 — Mock Q1 journal review

You: Review this introduction and hypotheses as a tough ABS 3* reviewer.
Claude (q1-journal-reviewer): in an anonymous-reviewer voice, flags gaps in theoretical contribution, breaks in hypothesis logic, and identification threats, with Major/Minor-Revision-level requests.

📜 License

This repo is a collection under per-directory licensing: original skills follow the root LICENSE; adapted/included third-party skills keep their own LICENSE and ATTRIBUTION.md in-folder. Full map in NOTICE.md. Skills marked 🔒 Non-Commercial are for non-commercial academic use only.

🙏 Credits

Deep thanks to the open-source authors this work stands on:

Zara Zhang — academic-slides (MIT)
Yuan Yizhe (Yuan1z0825) — nature-skills, the publication-grade plotting foundation of management-figure (MIT)
Cheng-I Wu (Imbad0202) — academic-research-skills, the systematic-review / meta-analysis module of phd-researcher (CC BY-NC 4.0)
qqfly1to19 — awesome_proofreading_auto, the audit-framework concept behind thesis-consistency-audit (CC BY-NC-SA 4.0)
PHY041 — claude-skill-citation-checker, a recommended companion tool (its code is not reproduced here)
academic-pptx's content guidelines draw on Barbara Minto's Pyramid Principle and Naegle (2021, PLOS Comput Biol).

If you are any of these authors and have any concern about inclusion or attribution, please open an issue — I will adjust immediately.

🤝 Contributing / ⚖️ Disclaimer

Issues and PRs welcome — especially new database profiles (docs/ADD_A_DATABASE.md). Please read CONTRIBUTING.md. Not affiliated with Anthropic; skill outputs are research aids — final academic judgment and responsibility rest with the user. The phd-milestone-tracker / qual-exam-coach rules are templates — verify against your own institution. Some skills require a paid database (TEJ).

claude-academic-skills