Research

無網環境的引用『內部一致性稽核＋幻覺風險評估』。做三件事：(1) 內文↔參考文獻清單雙向對帳（孤兒引用、冗餘文獻）；(2) APA 7 格式診斷；(3) 幻覺風險訊號標記（頁碼範圍異常、DOI 格式錯、作者-年份-期刊組合可疑）——只標『疑似、需工具驗證』，不宣稱已查證文獻真實存在。適用中英文論文的引用稽核。觸發詞：引用檢查、參考文獻對帳、孤兒引用、幻覺文獻、假文獻、chimeric、拼接文獻、APA 7 檢查、DOI 檢查、文獻一致性、citation check。與 check-citations 劃界：本 skill 是無網、做內部一致性與風險評估，無法確認文獻真偽；要對接 CrossRef/Semantic Scholar/OpenAlex 做真實性查驗，改用 check-citations（Claude Code 環境須加 anthropic-skills: 前綴）。與 thesis-consistency-audit 劃界：那個做全論文六維度數字對帳，本 skill 專攻引用與參考文獻。

因果推論架構師(頂刊識別策略,2026 前緣標準):現代差異中之差異 DiD——交錯採用(staggered adoption)下 TWFE 的偏誤、Goodman-Bacon 分解、Callaway & Sant'Anna / Sun & Abraham 等異質穩健估計量、事件研究圖與平行趨勢診斷(含 Rambachan & Roth 誠實區間);工具變數 IV(弱工具診斷、排除限制論證);斷點迴歸 RDD;合成控制 SCM;安慰劑檢定矩陣。輸出:識別策略選擇書+對應 R 語法(fixest/did/rdrobust)+審稿人攻防表。何時用:政策/制度/揭露的採用效果、任何想宣稱因果的檔案資料研究、審稿人質疑內生性與識別。觸發詞:DiD、差異中之差異、雙重差分、交錯採用、staggered、平行趨勢、事件研究圖、event study、事件源、重大訊息事件、MOPS 事件、工具變數、IV、弱工具、2SLS、斷點迴歸、RDD、合成控制、synthetic control、因果推論、識別策略、內生性、自然實驗、政策評估、處理效應、Callaway、Goodman-Bacon、TWF...

Create Beamer-inspired academic HTML presentations from scratch or by converting PowerPoint files. Use when the user wants to build slides for a conference talk, lecture, seminar, or thesis defense. Supports theorem environments, KaTeX equations, algorithm pseudocode, and citations. Helps academics discover their preferred visual theme through Beamer-style previews rather than abstract choices.

Use this skill whenever the user wants to create or improve a presentation for an academic context — conference papers, seminar talks, thesis defenses, grant briefings, lab meetings, invited lectures, or any presentation where the audience will evaluate reasoning and evidence. Triggers include: 'conference talk', 'seminar slides', 'thesis defense', 'research presentation', 'academic deck', 'academic presentation'. Also triggers when the user asks to 'make slides' in combination with academic ...

Validate bibliography entries against citations in all lecture files. Find missing entries and unused references.

OCRs scanned or image-only corpora with vision-language models in three phases. Evaluate compares candidate systems against stratified human ground truth and selects one by measured CER or WER. Run builds the production pipeline, including model selection, image handling, prompts, architecture, batching, accuracy evaluation, and reproducibility. Clean corrects raw OCR with LLM and rule-based passes, quality diagnostics, multilingual handling, and span-level provenance. Use when choosing an OC...

Specifies and diagnoses structural topic models for survey and experimental text. Covers STM, LDA, and BERTopic choice, preprocessing consequences, prevalence and content formulas with spectral initialization and recorded seed, topic-count selection across semantic coherence, exclusivity and FREX, held-out likelihood, and residuals, interpretation and validation against representative documents, robustness checks, and DA-RT-compliant reporting. Use when the user has open-ended responses or an...

Designs and validates LLM-based text classification for research data. Covers codebook construction, learning-regime choice, model selection and reproducibility, prompt construction, pilot validation against human coding with kappa and F1, hybrid human-LLM workflows, and reporting. Includes a resumable batch pipeline with a rule-based baseline for repeated free text against a closed codebook, including occupation, institution, registry text, and open-text survey responses, with frozen inputs,...

Designs and formats publication-quality tables for social-science manuscripts. Covers column order, row grouping and panels, decimal precision, standard-error and confidence-interval conventions, self-contained titles and notes, and code-generated output. Use when the user builds or revises a regression, balance, descriptive-statistics, summary, or robustness table, mentions stargazer, modelsummary, gt, huxtable, kable, pystout, esttab, or booktabs, or asks whether a result belongs in a table...

Designs survey instruments. Covers item-specific wording that avoids acquiescence, double-barreling, and leading language, scale construction including points, labels, polarity, feeling thermometers, Cronbach’s alpha, and McDonald’s omega, flow and ordering, treatment placement, buffer items, cognitive interviews, soft launches, burden, attention checks, speeding rules, sensitive questions, social-desirability bias, and comprehension gates. Use when the user writes or revises survey questions...

Audits fielded survey-response data for registered elements, data quality, bot and AI-automation screening, and sample integrity. Produces an appendix-ready quality report.

Spawns full Claude Code peer sessions in separate terminal panes and git worktrees, real sessions rather than subagents, each with a directed task and contract brief, monitored and merged by the spawning lead. Detects the environment and uses herdr first, then tmux, then a native claude background agent. Use when work must persist or run beside the current session, needs a separate worktree or permission settings, should remain user-steerable in a visible pane, or when the user asks to spawn,...

Plans a research project as a decision map that persists across sessions. The destination is a defensible, pre-registerable design reached by resolving four kinds of decision tickets one at a time. Adapted for research from Matt Pocock’s wayfinder. The map lives as repository Markdown, each session claims one frontier ticket, literature sweeps run as parallel research tickets, and the resolved map compiles into a pre-analysis plan. Use at project start, when a design has more open decisions t...

Scaffolds or audits an entire research-project repository organized around its source library. Use when the user starts, structures, organizes, or reviews a project, including "set up a research repo", "how should I structure/organize this project", "initialize my sources folder", "new paper or literature-review project", "audit my repo structure", "is my sources folder set up right", or "check my project layout". Builds the source spine, original PDFs, tracked Markdown conversions, drop zone...

Interviews a researcher in rounds until a research idea, design, or draft has no unexamined decision. Numbers every question, explains "why this matters", recommends an answer, fetches facts rather than asking for them, and records settled decisions. Three stages cover idea, a topic or hunch to a falsifiable question and contribution claim, design, a question to estimand, identification, sample and power, measurement, pre-registration, analysis plan, and venue, and defend, reviewer objections...

Scaffolds or audits a social-science replication package and evaluates the manuscript and archived research objects against FAIR principles. Scaffold mode writes folder structure, README, master.R, figure-table crosswalk, codebook template, LICENSE placeholder, .gitignore, and a pre-release checklist. Audit mode grades the package and audits data, code, materials, prompts, preregistrations, DOIs, metadata, licenses, access restrictions, and availability statements. Use when setting up or repa...

Organizes and formats an author’s response to peer review. Extracts every distinct point from referee reports and the editor’s letter, tags severity and type, maps dependencies to sequence revisions, flags potential pushback, and builds a numbered comment → response → location table with substantive responses left to the author. It formats and checks the response without writing scientific content. Use when the user has reports, a revise-and-resubmit decision, or an editor’s letter and says "...

Operates or audits a live Qualtrics survey through the v3 APIs without disrupting fielding. Covers publish gating, quotas, flow routing, embedded data, panel-vendor redirects, read-back verification, and read-only pre-fielding audits. Use when publishing or patching an instrument, when a quota counts without blocking, when configuring panel-vendor redirects or flow gates, or when auditing before launch for consent-before-anything gates, force-response completeness, quota and redirect checks, ...

Launches and configures the standalone presubmit CLI, an API-driven adversarial peer-review pipeline with 30-plus stages, including Red Team finders, Blue Team defence, verification cascade, legal pass, and copyedit, producing one consolidated report on disk. Verifies installation and API key, chooses the output location and smoke, standard, or custom mode, launches the run, and reports the result path. Use when the user asks to run presubmit, wants a deep unattended pre-submission audit, or ...

Writes a pre-analysis plan before data collection. Covers registry selection, OSF, AsPredicted, AEA, EGAP, PAP structure, an analytical strategy specified through model and decision rule, preregistered analysis code using simulated data, contingencies for attrition, failed manipulations, and exclusions, deviation documentation, and timeline. Operationalizes the pre-data-collection side of DA-RT. Use when the user asks to write or review a pre-registration or PAP, asks which registry to use, w...

Typesets a working paper or journal submission in house-style LaTeX from Markdown, Word (.docx), TeX, ODT, RTF, or HTML. Converts with pandoc, wraps the document in an EB Garamond template, builds the PDF with latexmk, and prepares journal-specific spacing, page limits, anonymization, disclosures, and citation style. Use for "format/typeset/convert my paper to LaTeX", "make a working paper", or "prepare this for submission to <journal>".

Provides a pre-submission manuscript audit covering argument, numerical consistency, references and DOIs, writing, figures, CONSORT flow, pre-registration, and the replication archive. Use before paper submission or when a draft needs an adversarial, quote-grounded review with a journal-readiness checklist. Add `--codex` to run the specification independently on Claude and Codex, GPT-6 Astra, and cross-check findings.

Runs a multi-model orchestration workflow led by the session’s strongest available model, Fable 5.1 or Claude Opus 5. Delegates mechanical work to a fast Sonnet subagent, wide reasoning to high-effort Opus subagents, and high-stakes or fresh-perspective work to Codex or a different-vendor GPT-6 Astra peer (`gpt-6-astra` by default). Detects the lead from session context, while `--lead fable` and `--lead opus` override it. Under Fable, judgment remains with the lead. Under Opus, parallel Agent...

Drafts or audits a scientific paper’s introduction and framing. Covers the motivating real-world "why", evidence audit, theory-to-falsifiable-if-then funnel with a named estimand, bridges across experiments, and a contribution list matched to the design. Use when the user asks to write, restructure, or critique an introduction, motivate or frame a paper, repair a gap-in-the-literature opening, sharpen or calibrate a contribution claim, or connect experiments into one argument. Sentence-level ...