Analyze & Visualize

Data Analysis & Statistics

• GEZHI - by SiyaoZheng · GPL-3.0 · net.Agent skills for social-science research in R and Python that persist dataset/codebook context across sessions, with a benchmarks section. Good for reproducible quant social-science analysis; GPL-3.0 and stats-workflow-focused, not writing.
• social-data-analysis - by nealcaren · MIT · net.An 18-skill sociology methods pack covering quantitative and qualitative social-data analysis, authored by a computational sociologist. Scope and provenance are strong; component depth unverified (content not inspected).
• claude-statistical-analysis-skill - by terryfyl · MIT · net.Statistical-consultant skill that diagnoses assumptions first, auto-selects methods (t-test to SEM/HLM/IRT/meta-analysis) and emits an APA-7 table, 300dpi figure and results paragraph. Assumption-first is its edge; advanced methods need the R Docker.
• stata-ai-fusion - by haoyu-haoyu · MIT.Lets an agent drive a real Stata session (run commands, inspect data, extract estimation results, capture graphs) via an MCP server plus a Stata skill knowledge base. Fills the AI-Stata gap for econ/epi/biostat; requires a licensed Stata install.
• stata-skill - by Dylan Moore · No License.The eval-backed reference that stops Claude writing subtly-wrong Stata — the silent traps that run clean but mislead.
• econometrics-skill - by xiaomihu1992 · MIT.Applied causal-inference skill: 17 estimators (OLS, PSM/IPW/AIPW, IV/2SLS, DiD/event-study, sharp/fuzzy RDD) with a method-selection decision tree, diagnostics and reporting templates. Solid for tabular econometrics; assumes clean panel data.
• fiftyone-skills - by voxel51 · Apache-2.0 · net.Official FiftyOne skill suite: expert workflows for computer-vision dataset curation, evaluation and model analysis via AI assistants, with a companion MCP server. Strong for CV/ML dataset research; tied to the FiftyOne tool and CV-specific.
• materials-simulation-skills - by heshamfs · Apache-2.0 · net.23-skill computational-materials-science suite: numerical-stability analysis, finite-element meshing, DFT/MD setup, Monte-Carlo/phase-field modeling, spectral methods, time-integration, and uncertainty quantification. Strong methodological breadth.
• matlab-agentic-toolkit - by matlab · No License · net.Official MathWorks toolkit of 61 agent skills for MATLAB/Simulink engineering and scientific work: curve fitting, system identification, timeseries forecasting, signal detrending/filtering, parameter optimization, unit conversion, and data cleaning.
• scienceclaw - by lamm-mit · Apache-2.0 · net.Computational science-and-engineering skill suite (340+ skills) from MIT's LAMM lab: materials science, molecular dynamics, finite-element analysis, symbolic math and multiscale modeling.
• daaf - by daaf-contribution-community · LGPL-3.0 · net hooks.Instruction framework making Claude Code behave like a rigorous, reproducible quantitative researcher in any domain, with auditable steps, guardrails, and human-in-the-loop verification. Heavyweight and opinionated; core decisions stay with the user.

Figures & Visualization

• paper-craft-skills - by zsyggg · No License.From an arXiv link to publication-style method figures, AIGC slide decks, and deep-dive explainer articles in one command, no API keys. Strongest at turning paper internals into visuals; output style is opinionated.
• engineering-figure-agent - by heyu-233 · MIT · net.Generates publication-style engineering and computer-science figures (diagram and plot modes) with bilingual EN/ZH labeling and editable output, tuned for Chinese academic papers. Narrow but does the figure job well; not a general plotting library.
• gen-pseudocode-skill - by huiyuli-2000 · MIT.Reconstructs submit-ready algorithm2e LaTeX pseudocode from a paper's methodology, code, and notes, with a code-to-math map, per-venue style, complexity analysis, and a compile check. Strong for ML/SCI algorithm exhibits; works at the method level.
• paperbanana-skill - by plutolei · MIT · net.Multi-agent pipeline (Retriever→Planner→Stylist) for academic diagrams, methodology figures, statistical plots and slide decks from text/data, plus a figure-vs-reference evaluator. Eight providers; mature (v4.x). Heavyweight for one-off diagrams.
• research-paper-figure-skill-factory - by c-narcissus · MIT-0.A figure-skill factory: distills one class of paper figure (framework, mechanism, taxonomy, results) into a reusable diagram-generation skill with multi-candidate rounds. Process-heavy and meta; overkill for a single one-off figure.
• skill-research-figure - by chingswy · No License.Draws professional academic figures for papers — LaTeX/TikZ diagrams and hand-drawn-style SVG. Tightly scoped to scholarly figure generation; needs a LICENSE added. Clean static scan; a solid pick for paper figures, especially TikZ.
• thesis-figure-skill - by 0xe1337 · MIT.Turns paper text into publication-grade figures: LaTeX/TikZ (pixel-precise, embeddable) or draw.io XML (route maps, slides). Targets high density + one-pass compile; bilingual. Best for structured CS/ML schematics, not data-driven statistical plots.
• nice-figures - by mapika · MIT.Matplotlib figures in a soft-pastel research-blog register: smoothed trends with confidence bands, rounded bars, log-log scaling-law scatter, white background, plus 16 recipes. Great for ML/alignment write-ups; one opinionated aesthetic.
• stata-graphics-skill - by youngfujun · No License.Knowledge base for AI-assisted Stata graphing aimed at economics empirical research. Narrow but well-targeted for econ publication figures; the main gap is a missing LICENSE. A good companion for Stata-based analysis workflows.
• AgentFigureGallery - by Dsadd4 · MIT · net.A taste layer, not a plotting library — curated references so the agent stops guessing what a Nature figure looks like.
• paper-style - by freemty · MIT.Five low-saturation color themes giving a paper one coherent identity across LaTeX text and matplotlib figures (documentclass to savefig), with a preview PDF. Great for consistent palettes; purely a styling layer, not a plotting or layout tool.
• tufte-data-viz - by caylent · MIT.Applies Tufte's data-viz principles (high data-ink ratio, no chartjunk, no pie charts) across six charting libraries plus accessibility rules. Excellent for clean, honest charts; deliberately opinionated, so it overrides styling choices.

Discover & Collect

Literature Search & Discovery

• arxiv-skills - by ultimatile · MIT · net.Pair of arXiv skills: convert a paper (LaTeX-source-first, PDF fallback) into implementation-ready Markdown with preserved math, plus a lightweight arXiv API DOI/title-search lookup. Conversion quality hinges on pandoc.
• cnki-skills - by cookjohn · No License · net.Drives CNKI (China's main academic database) from Claude Code via Chrome DevTools MCP: search, journal browse, citation analysis, PDF download, BibTeX/Zotero export. Go-to for Chinese-language lit work; needs browser-MCP setup and ships no license.
• gs-skills - by cookjohn · MIT · net.A genuinely useful Google Scholar toolkit for the CLI: keyword/advanced search, citation-chain tracking, full-text link resolution and one-step Zotero export. Relies on Chrome DevTools MCP and scrapes the DOM, so it can break when Scholar changes.
• scientific-skills - by yorkeccak · MIT · net.Natural-language literature search over 13 sources (PubMed, arXiv, bioRxiv/medRxiv, ChEMBL, DrugBank, Open Targets, trials, patents) via Valyu semantic search. Strong biomedical/drug-discovery coverage; depends on the Valyu API, not self-contained.
• ai-skill-scholar - by Sébastien Dubois (dsebastien) · MIT · net.Three composable OpenAlex skills (venue search, citation-graph walk, review orchestrator) — stdlib-only, no pip, no key.
• paper-search-pro - by O0000-code · Apache-2.0 · net.OpenAlex-led five-source search (PubMed, arXiv, Semantic Scholar, CrossRef) across four depth tiers; goes past a hit list to a saturation curve, a stop decision, and an interactive HTML report. No LLM key; one free OpenAlex key to start.
• scholar-kit - by lottshin · MIT · net.Chinese-first literature toolkit: search CNKI/OpenAlex/Semantic Scholar/arXiv/NSSD, enrich via Crossref, resolve OA via Unpaywall, batch-download (incl. CNKI PDFs), export GB/T 7714/BibTeX/RIS/APA. Strong CNKI; scraping is fragile + ToS-sensitive.
• literature-review-skill - by yanzhanlin · MIT.End-to-end lit-review suite (search, acquire, deep-read, write, dissertation chapter) built on PRISMA-S, PICO/SPIDER and named review-methodology frameworks. Strong grounding and reproducible search logs; Chinese-first, modest in stars.
• litreviewskill - by zsun79 · No License · net.End-to-end literature-review workflow: drafts keywords, builds an OpenAlex seed set, expands by backward/forward citations, screens by title/abstract until saturation, then ranks and reads up to 30 full texts into a concept matrix.
• oneshot-academic-research-skill - by orhoncan · MIT.Iterative deep-literature-research skill: 5-15 search cycles gathering 12-50+ sources with gap identification, source-diversity tracking and APA7 footnotes, to Obsidian or PDF. Turkish/English auto-detect; depth depends on the underlying web tools.
• literature-harvest - by zhongzhx · MIT · net.Bulk keyword-driven literature harvesting across PubMed, Europe PMC, Crossref and OpenAlex: builds a candidate table, downloads legally-accessible full-text PDFs with an HTML-to-PDF second pass, and dedups.
• lune-research - by Retrograde Labs · MIT · net.Restricts an agent's literature search to full text from a fixed set of peer-reviewed venues and returns a verbatim supporting quote for each verified claim. Strong on ML and security; indexes nothing else, so not a general search.
• asta-skill - by agents365-ai · MIT.Instruction-pack skill that drives Ai2's Asta MCP (Semantic Scholar): keyword/ID/author search, citation traversal, batch lookup and ~500-word snippet retrieval, with safe defaults. Clean routing; needs the Asta MCP + free API key.
• semanticscholar-skill - by agents365-ai · MIT · net.Wraps the Semantic Scholar Graph API for paper search, citation-graph traversal, and author lookup across 200M+ papers, rate-limited for multi-agent use. A reliable discovery building block; intentionally narrow (S2 only), so pair it with others.
• scholar-deep-research - by agents365-ai · MIT · net.Script-driven 8-phase literature-review pipeline over 7 federated sources with enforced citation anchoring, dedup, transparent ranking, citation-chasing and a mandatory self-critique gate.
• paper-fetch - by agents365-ai · MIT · net.Resolves a DOI (or batch) to a downloadable PDF via a 7-source fallback chain (Unpaywall, Semantic Scholar, arXiv, PMC, bioRxiv, publisher, then Sci-Hub) with per-source reporting. Clean and zero-deps; the Sci-Hub fallback is a caveat.

Reference & Citation Management

• apa-referencing-skill - by Chuah Kee Man (keemanxp) · No License · net.The most complete APA-7th formatter: 28 source types, down to tweets and datasets. Ships no open-source license.
• chinese-reference-formatter-skill - by zechang-xiong · MIT.Formats Chinese academic references to GB/T 7714 and completes BibTeX from Chinese/English titles. Fills a real gap for CN scholars that Western citation tools ignore. Narrow by design; bundles helper scripts and an agent.
• citation-assistant - by zhangny301 · No License · net.Automated academic-citation skill over the Semantic Scholar API: resolve, verify, and format citations into a manuscript. Useful, focused citation helper; single-skill and tied to S2 coverage, and its license needs confirming.
• wenxian - by njzjz · LGPL-3.0 · net.Generates BibTeX from a DOI, PMID, arXiv ID, or paper title by querying CrossRef, PubMed, arXiv, Semantic Scholar and ChemRxiv. Mature, tested, PyPI-published; the agent skill is a thin wrapper over a solid CLI. LGPL-3.0.
• fetch-bib - by MilkClouds · No License · net.Human-in-the-loop BibTeX fetcher pulling every field from the authoritative publisher (ACL/PMLR/arXiv/NeurIPS, DBLP fallback) not the LLM, with provenance per entry and a stop-to-ask on ambiguous venues. Great for de-hallucinating .bib; no license.

Read & Understand

PDF / OCR / Document Parsing

• auto-paper-harvester - by jxtse · No License · net.Batch-downloads paper PDFs by DOI, cascading publisher TDM APIs (Wiley/Elsevier/Springer) to OA sources (Crossref/OpenAlex/Unpaywall) to an optional browser fallback on an institutional session. Good for bulk retrieval; ships a hardcoded credential.
• DOCX (Anthropic document skill) - by Anthropic · Proprietary.Anthropic's Word skill whose academic edge is first-class tracked changes and comments — the best off-the-shelf option for revising a marked-up manuscript.
• mineru-skill - by nebutra · MIT · net.Parses PDFs, Office docs and images into clean Markdown with LaTeX formulas, tables and OCR via MinerU. A solid OCR/extraction entry point for getting papers into editable text.
• PDF (Anthropic document skill) - by Anthropic · Proprietary.Anthropic's own PDF skill — a task router over pypdf/pdfplumber/reportlab/qpdf for routine merge/split/extract/forms. The reliable default, not a high-accuracy OCR engine.
• PPTX (Anthropic document skill) - by Anthropic · Proprietary.Anthropic's PowerPoint skill that builds decks and pulls text from any .pptx — an authoring tool for paper-to-talk, not layout-faithful slide OCR.
• XLSX (Anthropic document skill) - by Anthropic · Proprietary.Anthropic's spreadsheet skill that, unusually, recalculates formulas rather than reading cached values — good for cleaning messy supplementary-data tables.

Reading, Summarization & Q&A

• dailypaper-skills - by huangkiki · Apache-2.0 · net.A daily paper-reading pipeline built from Agent Skills: arXiv search, PDF parsing, key-point extraction, Zotero sync, and structured notes plus a daily digest. Cohesive and popular; Chinese-first docs, so non-CJK users lean on the code.
• scholar-skill - by eesjgong · MIT · net.Obsidian-centric academic-reading suite: deep-reads papers and links them into an evolving personal knowledge base with reflection prompts. A strong pick for Obsidian users; it reads and organizes rather than discovering papers.
• benchmark-research-skill - by eternalwavee · MIT · net.Extracts benchmarks, datasets, metrics and baselines from a paper, or surveys what benchmarks fit a research direction; arXiv-source-first with Obsidian notes. Uses an unscoped Bash(*) allowlist — broad shell access.
• claude-paper - by alaliqing · MIT · net hooks.Paper-study plugin that turns a PDF or arXiv link into a learning environment: parsed text, adaptive summaries/Q&A, runnable code demos and a KaTeX web viewer. Rich but heavy (Node/Nuxt/poppler deps); uses an unscoped Bash allowlist.
• deeppapernote - by 917dhj · MIT · net.Deep-reads a single paper and produces high-quality, structured Obsidian-style research notes (claims, methods, takeaways) usable across Claude Code/Codex/Cursor. Great for building a notes vault; one-paper-at-a-time, not corpus-scale.
• paper-analyst - by flyer-li · MIT.Analyzes a paper in five depth modes (quick to presentation-with-figures) with anti-hallucination source tagging and method extraction tuned to paper type. Good for structured reading/summaries; single-skill, no external data sources of its own.
• paper-reader-heilmeier - by realzyzhang · MIT.Reads STEM papers through Heilmeier's Catechism, surfacing the core problem, approach, cost and impact for a fast big-picture take. A focused, opinionated comprehension aid; great for triaging papers, less so for deep line-by-line reading.
• paper-reading-skill - by kingslayer-bot · MIT.Deep paper-reading skill with two modes: paragraph-level close reading with agentic Q&A, and a 20+-dimension analysis run by parallel agents, emitting Obsidian-ready notes. Strong for deconstructing single papers; not a search or citation tool.
• Claude Deep Research Skill - by 199 Longevity (199-biotechnologies) · No License · net.The most engineering-heavy deep-research skill — verifies each claim against its source via CrossRef and persists citations across compaction. No license file.
• Deep Research Skill - by Lan Zheng · MIT · net.Two-phase deep research that gates on your approval of the outline before any search runs — the human-in-the-loop answer to fire-and-forget agents.
• llm-wiki - by nvk · MIT · net.LLM-compiled knowledge bases for agents: parallel multi-agent research, thesis-driven investigation, source ingestion, then a queryable wiki. Good for building a durable, cited knowledge base from many sources.
• NotebookLM Research Skill - by Claude-World · MIT · net.Drives Google NotebookLM from Claude for cited answers and free artifacts (podcast, mind map, study guide) — an unofficial cookie-based client that can break.
• notebooklm-skill - by pleaseprompto · MIT.Bridges Claude Code to Google NotebookLM so you get source-grounded, citation-backed answers drawn only from your uploaded documents. Excellent for grounded Q&A over a personal corpus; depends on a Google NotebookLM account and the notebooklm-mcp.
• arxiv-paper-zh - by zeya-labs · MIT · net.Translates arXiv LaTeX papers into Chinese while preserving formatting (formulas, figures, citations), from an arXiv ID or local source. The format-preserving approach beats plain-text translation; scope is narrow (arXiv LaTeX, EN to zh).
• cheatsheet-generator-skill - by evan715823 · MIT · net.Turns course slides and PDFs into dense, exam-ready LaTeX cheatsheets. Narrow and student-focused (study/teaching rather than original research), but well-targeted and popular; good for condensing course material, not for literature work.
• lecture-to-notes - by ysyecust · NOASSERTION.Turns YouTube/Bilibili lecture videos into structured LaTeX/PDF notes via transcript extraction, key-frame detection, and smart-cropped screenshots. Novel and useful for students; quality depends on transcript availability, and the license is murky.
• paper2code - by prathamlearnstocode · MIT · net.arXiv ID in, citation-anchored Python implementation out — each module tagged to the paper section it implements, ambiguities flagged rather than guessed. Great for paper reproduction; not a general code generator.

Review & Publish

Peer Review & Response

• AI-research-feedback - by Claes Bäckman · MIT.Pre-submission referee simulator tuned only for AER/QJE/JF-tier economics — six parallel agents, not general review.
• archora-skills - by Richard Kim (Archora) · MIT.Simulates an editor-in-chief plus three reviewers; vendor-affiliated (Archora) but runs standalone as plain markdown.
• polisci-review - by cmertdalli · MIT.Pre-submission audit for political-science manuscripts: 9 modules (contribution, theory, measurement, identification, transparency, journal fit) with journal-aware personas and 8 verified journal-policy profiles. Rigorous and discipline-specific.
• academic-review-skill - by pengkangzhen · MIT.Reviews OR/management-science manuscripts for academic merit rather than implementation, detecting domain-specific red flags (infeasible quantities, trivial VSS%) and deliberately phrasing critiques as queries to avoid killing genuine findings.
• ai-peer-review-skill - by alexwortega · MIT.Drop in a paper PDF, get N independent structured reviews plus a synthesized meta-review and a concern matrix CSV. Adapts poldrack/ai-peer-review to free Claude subagents; defaults to a hard alignment-style critic.
• manuscript-review-skill - by shaowen-ye · MIT.Pre-submission manuscript review by six role-specialized AI reviewers (architecture, theory, methods, etc.) that emits a bilingual EN-ZH color-annotated .docx with a 1-10 scoring matrix and prioritized fixes, calibrated to a target journal.
• mean-reviewer-skill - by xz-liu · No License.Roleplays the worst-case adversarial peer reviewer: an inflated wall of weaknesses, framework attacks, a score fixed at reject, and a rebuttal phase conceding nothing. Half polemic on LLM reviews, half a pre-submission stress test; not a gentle tool.
• diogenes - by diogenes-project · GPL-3.0 · net hooks.Applies a unified, anti-sycophantic evidence methodology - ICD 203, GRADE, PRISMA, Cochrane RoB and others - as a 14-step appraisal for systematic reviews, evidence grading and claim verification. Rigorous; ships hooks.
• slr-prisma - by keemanxp · NOASSERTION.A systematic-literature-review scaffold following the PRISMA 2020 framework (protocol, search, screening, flow diagram, synthesis). Solid methodology for rigorous reviews; process-guidance rather than automated search, and its license is unclear.

Submission, Formatting & Conversion

• paperfit - by openraiser · MIT · net.LaTeX typesetting suite that compiles, renders pages to images, then visually diagnoses and fixes layout problems (overfull boxes, float placement, spacing, tables). The vision-in-the-loop loop is its real edge; needs a working local LaTeX toolchain.
• claude-skill-overleaf - by junhahyung · MIT.Reads and writes Overleaf projects from Claude Code via the Overleaf git bridge: edit, commit, and push LaTeX to the web editor with secure token handling and no force-push. Includes a rebuttal scaffold. A clean way to bring an agent into Overleaf.
• latex-document-skill - by ndpvt-web · MIT.The most template-heavy LaTeX skill here — 27 templates spanning papers, forms and tables; breadth over a single niche.
• beamer-academic - by faust-donf · MIT.Turns a thesis or paper into ready-to-present academic Beamer slides with a real layout library and an interactive editing loop. Bilingual, no LaTeX expertise required; scope is defense/conference decks, not general slide design.
• beamer-skill - by Noi1r · MIT.The strongest pick for academic LaTeX slides, with opinionated pedagogy: no overlays, density caps, visual overflow detection.
• beamer-skill - by jaxonjp · MIT.Create/compile/review/polish academic Beamer decks with TikZ and pedagogical audits and quality gates; explicitly defers PPTX to other skills. A derivative work, extended; LaTeX-only by design.
• powerpoint-skill - by noi1r · MIT.Creates visually rich PowerPoint decks with native OMML math and LaTeX support, well suited to research talks and lectures that need real equations in slides. Not academic-only, but the math support sets it apart.
• thesis-defense-pptx-skill - by zouchenzhen · Apache-2.0.Builds an editable defense deck from your thesis PDF/LaTeX while strictly reusing a university PPTX template; inventories cited figures and runs overflow/layout checks. Windows+PowerPoint COM for the full quality gates.
• docx-skill-4-cn-paper - by gostyan · MIT.Applies standard Chinese-paper Word formatting - fonts, sizes, spacing, margins, heading levels, figure/table numbering, references - for course papers, math-modeling competitions and theses, incl. md-to-docx. Narrow but does its job well.
• Manuscript Submission Formatting Agent - by Chen Zhu (朱晨) · No License · hooks.Reformats a finished manuscript to a new journal's author guidelines for post-rejection resubmission, never touching the results.
• word-format-skill - by yeap531 · No License.Replicates a reference Word document's typography onto new content via an HTML bridge — useful for matching a journal or institutional .docx template. Strong for submission/formatting chores.
• paper2patent - by 7tocr · MIT.Turns a finished paper into a China invention-patent application draft (claims layout, specification norms, technical-effect argumentation) via structured prompts, patent-figure generation, and skills. A unique paper-to-IP bridge for China patents.
• any2pdf - by lovstudio · MIT.Markdown to PDF with no LaTeX; its real edge is CJK rendering where Pandoc and wkhtmltopdf break.
• Markdown Exporter - by bowenliang123 · Apache-2.0.The broadest, most battle-tested format coverage here (DOCX/PDF/PPTX/XLSX/EPUB), running locally with no outbound calls.
• latex-arxiv-SKILL - by appautomaton · No License · net.Issue-driven Codex skill for ML/AI arXiv review papers: scaffolds the LaTeX project and verifies every BibTeX entry end-to-end. The BibTeX verification lifts it above generic templates; aimed narrowly at arXiv-style ML reviews; ships no license.

Suites, Systems & Meta

Autonomous Research Systems

• ARIS (Auto-Research-In-Sleep) - by Yang Ruofeng · MIT · net hooks.An autonomous research loop whose signature is a cross-model jury — Claude, GPT and Gemini critique each other so no model rubber-stamps itself.
• autodidact-skill - by damonchen-anan · MIT.Autonomous topic-to-wiki research agent extending Karpathy's LLM-Wiki: auto-fetches web sources, compiles an interlinked wiki, and self-audits coverage gaps each round. Good for unattended literature scaffolding; not a verified-citation tool.
• autoresearch-skill - by wjgoarxiv · MIT · net.Turns a natural-language research.md goal into a Karpathy-style experiment-evaluate-iterate loop, generalized beyond ML to any task with a mechanical evaluator, with auto rollback, audit trail, safety budgets. Domain-general; thin on literature work.
• de-anthropocentric-research-engine - by yogsoth-ai · Apache-2.0.Huge pure-markdown library of autonomous-research SOPs (~770 skills) arranged in a Campaign/Strategy/Tactic/SOP hierarchy covering direction-setting, literature acquisition, gap discovery, hypothesis formation and adversarial stress-testing.
• nano-scientist - by ai4scientist · No License · net hooks.Autonomous research agent: turns a topic into a full draft paper via four budget-capped loops (deep-search, experiment, improvement, review) with CrossRef citation recovery and a multi-model review gate.
• rstack - by sunnnybala · MIT · net.Research-automation chain: /research runs lit-review→novelty-check→experiment→analyze-results→write-paper toward a submittable draft in one session; each skill also works standalone. Ambitious end-to-end; quality hinges on the experiment step.
• academic-research-agent-skill - by ngtiendong · MIT.Human-in-the-loop research workflow for MS/PhD in CS/AI/math/eng: scopes questions, grounds a lit review in sources, runs a novelty gate, plans experiments, simulates reviewers, verifies claims. Disciplined collaboration, not a hand-off.
• anaxa - by citrus-bit · MIT · net.Auditable, pausable research-agent workbench: drives a topic through lit search, evidence binding, sandboxed experiments, draft and citation audit to a LaTeX/PDF bundle with human gates. Heavy full-stack app, not a drop-in skill.
• phd-skills - by fcakyon · MIT · net hooks.PhD/ML research guardrails: reproduce arXiv papers, debug runs evidence-first, compare experiments at the same epoch, audit dataset bias, run pre-flight launch checks. Catches costly AI research mistakes; uses a Bash wildcard plus opt-in alerts.
• research-units-pipeline-skills - by willoscar · No License · net.File-first research harness turning open-ended goals into protocolized, resumable pipelines (lit survey, paper review, evidence synthesis) with per-stage acceptance gates and durable artifacts. Heavy machinery; steep to learn; no LICENSE file yet.
• vibe-science - by th3vib3coder · Apache-2.0 · net hooks.Integrity-first research runtime for Claude Code: a plugin that enforces checks and persists state, plus a methodology skill built on falsification, adversarial review, and confounder discipline. Aims to make AI science hard to fake; heavy exec.
• ai-research-army - by terryfyl · Apache-2.0 · net.Nine-agent pipeline taking clinical data to a submission-ready medical manuscript — requirement scoping, data profiling, stats, figures, literature, drafting, submission packaging. Ambitious end-to-end; some modules are a partial public release.
• claude-workflow-for-econ-phd - by saptarsibhowmick · MIT · net hooks.Fork-ready scaffold for multi-paper empirical-economics PhD work (lit review to submission): 9 skills (missing-lit, referee2, blindspot), a Zotero PDF-chunking literature pipeline, file-safety and IRB handling. A scaffold to adopt, not a drop-in.
• neurico - by chicagohai · Apache-2.0 · net.Give it a structured hypothesis; it runs the experiment loop and emits code, figures, and a LaTeX paper. Lab-built and genuinely scholarly, though heavy (git+docker) and only as good as the domain harness behind it.
• scienceclaw - by beita6969 · MIT · net.Large self-evolving research-agent platform (OpenClaw-based): 285 skills across 28+ disciplines, persistent memory, API-grounded citations. Popular for end-to-end research agents; bundles a full messaging/browser/sandbox OS, so read before trusting.
• shidi-skill - by IcyCreamDAS · MIT · net.Bilingual research workflow for agent IDEs: multi-angle literature review with per-angle files, anchored experiment design with a caveat list, figures, paper reading; returns files plus a cross-verification brief for a second model. Zero deps.
• create-research-experiment-repo - by Shiba-hua · MIT · net.Scaffolds repos for delegated experiments (Chinese-first): gets report questions, figure specs, directory tree and roles approved first, so each claim traces to raw runs. Executors: agents, lab staff or instruments; runs nothing itself.

Awesome Lists & Ecosystem Maps

• Awesome AI for Economists - by Lu Han · CC0-1.0.A discipline-focused map of AI tools for economics research and teaching, from the OpenEcon team.
• Awesome Scientific Skills - by Intern Science · MIT.The closest active peer — a curated awesome-list of agent skills for science. Cross-reference, don't race.
• Awesome-AI-Scientists - by FIB Lab, Tsinghua University · MIT.The reading list behind Tsinghua FIB Lab's AI-Scientists survey — a paper-centric map of the literature.
• Awesome AI for Science - by ai4s-research · MIT.A broad AI4Science map of tools, papers and datasets — wider and more model-oriented than a skills list.

Discipline-Specific Packs

• alterlab-academic-skills - by alterlab-ieu · MIT · net.186+ domain-organized academic skills for faculty/researchers, wrapping real scientific databases and libraries (ENA, biopython, cellxgene, cobrapy, ESM, etc.).
• beril-research-observatory - by kbaseincubator · AGPL-3.0 · net.KBase AI co-scientist over the BER microbial-ecology data lakehouse: 16 skills plus reusable patterns for pangenomics, fitness, metagenomics and biochemistry. Strong for KBase research; large executable surface and AGPL, so read before trusting.
• bioSkills - by GPTomics · MIT.The most benchmark-honest bioinformatics pack — idiomatic samtools/DESeq2/Seurat code, with published Bio-Task Bench scores — now archived upstream.
• clawbio - by clawbio · MIT · net.Bioinformatics-native agent skill library: local-first, reproducible genomics/QC/GWAS workflows built on OpenClaw, with CI and a cited release. Strong domain coverage; large executable surface typical of a real bioinformatics stack.
• dr-cook - by wen-chen · MIT.Modular research-lifecycle suite (literature, gap analysis, writing, peer review, bioinformatics, grants) with a routing layer and depth in Traditional Chinese Medicine. Strong for TCM/biomed; modest adoption, do-everything breadth.
• encode-toolkit - by ammawla · AGPL-3.0 · net.Genomics suite: searches ENCODE, cross-references 14 databases (GWAS, ClinVar, GTEx, JASPAR), runs 7 pipelines, and writes provenance-tracked methods. Heavily tested; AGPL and genomics-specific, so niche elsewhere.
• labclaw - by wu-yc · No License · net.Large lab-science skill library (240 SKILL.md across biology, pharma, medicine, literature, vision) wrapping real tools — ESM, AlphaFold DB, biopython, single-cell stacks — for dry-lab reasoning and protocol composition.
• medical-research-skills - by aipoch · MIT.A 500+ skill medical-research suite: RNA-seq/scRNA/spatial analysis, ADMET, evidence mapping, and a full manuscript-writing track. Broad and uneven by nature, but the depth in biomedical data analysis is unmatched here.
• medsci-agent - by omar-a-hassan · MIT · net.Biomedical-research suite of 17 skills: multi-source literature search, ClinicalTrials.gov, drug interactions, and a genomics stack (gene lookup, variant annotation via ClinVar/dbSNP/gnomAD, pathway analysis). Deep domain coverage.
• medsci-skills - by aperivue · MIT · net.A 59-skill medical-research suite from a physician-researcher: PubMed search, study design, IRB protocols, biostatistics, reporting checks against 49 EQUATOR guidelines, figures, manuscript drafting. Biased toward radiology and diagnostic accuracy.
• neuroclaw - by cuhk-aim-group · MIT · net.Neuroscience and medical-imaging skill pack (87 skills) from CUHK's AIM group: connectome analysis, fMRI/EEG pipelines, neuroimaging statistics and clinical-data workflows.
• OpenClaw-Medical-Skills - by FreedomIntelligence · No License · net.The largest open medical-research library (hundreds of skills) — broad coverage, but giant, unauditable and unlicensed.
• protein-design-skills - by Adaptyv Bio · MIT.Narrow, deep protein design that auto-chains BoltzGen, Chai and QC into one pipeline; expects real compute backends.
• SciAgent-Skills - by jaechang-hits · CC-BY-4.0.The largest open life-sciences library (199 skills), reporting a 65%-to-92% BixBench jump; low-risk but too large to fully audit.
• skillfoundry - by ma-compbio-lab · Apache-2.0 · net.Computational-biology discipline pack: 260+ skills wrapping standard bioinformatics workflows (RNA-seq with DESeq2/Salmon, GATK variant calling, ChIP/ATAC-seq, single-cell QC and integration, spatial transcriptomics, MaxQuant proteomics).
• automcm-pro - by realseaberry · MIT · net.End-to-end math-modeling-competition assistant (CUMCM/MCM): drives literature search, derivation, code, LaTeX and paper writing with forced code self-verification. Strong for modeling contests; competition-specific and Chinese-first.
• mathmodel-skill - by yushui2022 · No License · net.Agent-native math-modeling-contest workflow (10 skills): parse the problem, pick a model route, generate and run code, gate on real evidence, then write and Word-format a full paper with format checks. Built for CUMCM-style contests; uses eval/exec.
• scp - by internscience · MIT · net.200+ computational-science skills (physics, chemistry, materials, numerical methods, HPC) wrapping real simulation tools like ASE. A discipline pack for bench/sim science rather than literature work; broad and cleanly built.
• agent-research-skills - by lingzhi227 · No License · net.31 full-lifecycle skills distilled from 17 LLM-agent repos — useful breadth, but unlicensed and rough-edged.
• awesome-gaussian-skills - by jaccen · Apache-2.0 · net.Discipline pack for 3D Gaussian Splatting / NeRF research: a 600+ method catalog with an interactive explorer plus skills for paper tracking, CUDA-kernel review, and spatial-intelligence Q&A. Strong for graphics researchers; uses a Bash wildcard.
• remote-sensing-research-radar - by limi124 · No License.Remote-sensing research-frontier radar: scans arXiv/Papers-with-Code/GitHub/HF and venue pages for geospatial-AI and transferable-CV work, ranks by novelty/reproducibility/fit, drafts related-work tables. Niche but well-scoped; no LICENSE.
• academic-research-skills (economics) - by franklee16 · No License.An economics/social-science pack useful for empirical economists, but lightly curated (near-duplicate skills) and unlicensed.
• ai-agent-research-starter-kit - by drchronx · NOASSERTION · net hooks.Chinese social-science research teaching pack with unusually deep methodology skills: causal inference (DID/RDD/IV/PSM), scale development, theory building, and EEG/ERP work, plus lit search and stats. Methods depth is its value; very large.
• education-agent-skills - by Gareth Manning · CC-BY-SA-4.0.The largest education pack — 165 evidence-based pedagogical skills across 20 domains; local install free, hosted MCP gated.
• open-science-skills - by scdenney · NOASSERTION.Social-science methods suite from real methodological texts: conjoint design/cleaning/diagnostics, list-experiments, cross-national design, methods-reporting, FAIR/figure-table audits. Source-grounded; broad, so audit individual skills.
• ai-anthropology-toolkit - by mattartzanthro · NOASSERTION.Qualitative-research skills for anthropology — peer review, manuscript evaluation, revision strategy — that treat epistemic stance as a design parameter rather than an afterthought. Scholarly and self-aware; confirm the license terms.
• claude-skills-journalism - by Joe Amditis · MIT · net hooks.A journalism pack whose academic core is its research-toolkit: legal paywall-bypass via Unpaywall, web archiving, change-monitoring.
• tw-research-skills - by fw1201 · No License.A full Taiwanese-academic lifecycle pack (proposals to submission), strong on qualitative methods (grounded theory, content analysis with Kappa) and TSSCI/NSC norms. Localized but broadly useful; confirm the license.
• genealogy-research - by sliday · MIT.Applies the Genealogical Proof Standard to family-history research — multilingual handwritten-record reading, evidence-level tracking, negative-result logging, GEDCOM and an Obsidian knowledge base. Specialist, but methodologically serious.
• oh-my-hermes-for-legal-researcher - by charliehotel · Apache-2.0 · net.US legal-research methodology ported from Anthropic's claude-for-legal: structured Federal Register digests, web-searched opinions tagged (verify), optional free CourtListener API for verified citations. No paid databases; US-only, candid on limits.
• awesome-rosetta-skills - by xjtulyc · NOASSERTION · net hooks.Multi-discipline research-skills library: 169 skills across 24 fields (physics to linguistics to public health), each wiring domain data sources and methods. Unmatched breadth for non-CS disciplines; depth varies, license is NOASSERTION.

Skill Suites

• academic-research-skills - in-the-loop pipeline built to fight AI fabrication that deliberately refuses to ghost-write; non-commercial license | net hooks |
• academic-research-skills-codex - native Academic Research Skills: deep research, paper drafting, manuscript review, research-to-paper pipelines, citation/integrity checks. Broad end-to-end; vendored as one giant skill (audit accordingly); non-commercial license | net hooks |
• AcademicForge
• AI Research Skills (Orchestra) - skill curriculum for the engineering half of ML research, taking an agent from idea to trained model — not wet-lab | net |
• anthropics/life-sciences - sciences marketplace — a vetted directory of ~20 partner data integrations, not a skill collection | net |
• aut_sci_write - lifecycle suite: literature search across arXiv/PubMed/Web of Science, PDF and figure extraction, review/meta-analysis writing, Zotero sync and slides. Broad and well-adopted; a do-everything suite, not a best-in-one-stage tool | net |
• claude-academic-research - built suite spanning the research lifecycle: PRISMA review, peer review, manuscript revision, empirical-integrity stat checks, and Zotero/MCP-grounded citations that refuse hallucinated references. Strong on rigor; needs Zotero+MCP setup | net |
• claude-scholar - gated Claim Promotion Gate; uniquely ships its own PreToolUse security-guard hook | net hooks |
• claude-scholar - skill academic toolkit: OpenAlex/arXiv search, DOI-to-BibTeX, citation verification, LaTeX cleanup, arXiv packaging, pre-submission checks, math verification, manuscript/figure critique. Honest that it is early-stage; lists its data egress | net |
• claude-scientific-writer - skill scientific-writing suite: literature review, citation management, peer review, figures, posters, slides and manuscript drafting. The most popular here and broad, but jack-of-all-trades; ships an unscoped Bash | net |
• codex-paper-skills - to-claim consistency, citation auditing, proof checking and rebuttal, with an evidence-first agent protocol. Strong breadth; the missing LICENSE is its main gap | net |
• google-deepmind/science-skills
• mgmt-paper-skills - research workbench: idea-to-review pipeline with SPSS/Stata/Python analysis, causal methods (DiD/RDD/IV/SC/PSM), qualitative coding and journal-figure profiles (UTD24). Deep for econ/management; Chinese-data oriented, needs a stats MCP | net |
• nature-skills - skill bundle strongest on Nature-style prose and figures, weakest as a data toolkit | net |
• open-scholar-skill - science paper suite for top journals: analytics with publication-ready regression tables/figures (NHANES/IPUMS/GSS), literature synthesis, causal design, citation. Broad end-to-end; non-commercial license + a self-citation nudge | net |
• papercash - to-end Chinese-student paper workflow over 8 free sources (S2, arXiv, Crossref, PubMed, CNKI, Wanfang, Baidu/Google Scholar): search, review, polish, plagiarism precheck, AI-rate reduction, GB/T-7714 refs. Broad but shallow per stage | net hooks |
• posit-dev/skills
• q-skills - skill academic toolkit for empirical research: literature review, R and Stata analysis, reproducible-research scaffolding, survey design, grant writing, peer review, and citations. Strong for quantitative social science; breadth over depth | — |
• research-agora - tracking and benchmark angle is its edge; breadth means moderate depth | net hooks |
• researchclaw - to-end OpenClaw research companion (six capabilities incl. arXiv search, reading, writing) with no API keys and a live demo, from a USTC lab. Broad workflow coverage; a stray SKILL.md.bak suggests light repo hygiene | — |
• scholaraio - in-one research suite (47 skills): literature search, reference management, PDF parsing and writing support, built to compose. A reasonable general backbone of moderate depth; note it ships an SSH-backup feature and a hook to review | net hooks |
• sciclaw - scientist' research suite (23 skills) wrapped in a lightweight Go runtime that injects lifecycle hooks, a manuscript spine and run-logging so every session is reproducible, logged and citeable | net hooks |
• scienceclaw - to-end research suite (275 skills, 9 agents, 77 databases) spanning literature search to publication. Ambitious breadth in one repo, but quality is uneven and it ships a chat-bot channel, hooks and a tools wildcard worth reviewing | net hooks |
• scientific-agent-skills - science pack — deep for life-science labs, near-useless for humanities | net |
• scientific-research-skills - first research skills for agents: literature search, three-depth paper reading, full-text harvest, related-work survey, Zotero management and publication-figure generation | net |
• voidful/academic-skills - school pipeline (CS/ML flavor) across Claude Code, Codex and Gemini | — |
• xueshuzhi-skills - coverage Chinese academic suite: topic selection, research design, CNKI/foreign literature search and verification, thematic coding, review and dissertation drafting, plus a 12-checkpoint TA pipeline assembling a Word draft. Breadth over depth | net |
• flonat-research - researcher infrastructure: 50 skills plus agents, hooks and rules for bibliography validation, LaTeX health-checks, experiment and causal design, and reproducible project setup. Strong end-to-end; ships hooks and an unscoped Bash | net hooks |

Write & Refine

Academic Writing & Drafting

• academic-skills - by chtc66 · MIT · net.Everyday-researcher workflow suite (AI/NLP-leaning, Chinese-first): paper deep-notes, survey writing, arXiv monitoring, review/rebuttal drafting, experiment-log summaries, benchmark extraction, research-gap finding and weekly lab updates.
• academic-writing - by alessandrocaforio · MIT.Multi-agent writing system that decomposes an empirical thesis into per-section skills (intro, methods, results, discussion, abstract) plus literature review, analysis, and submission. The section-level structure is its strength; nominally econ.
• academic-writing-skills - by wenyuchiou · MIT.Field-agnostic paper-writing workflow: findings-first drafting, de-AI/overclaim detection, claim-to-evidence checks, reviewer-response tables, and pre-submission checklists, with per-paper journal overrides. Ships evals — rare and welcome.
• enhanced-mathmodel-codex-skills - by xzwwwwww · No License · net.Full-pipeline skills for mathematical-modeling competitions, automating the workflow through to a complete modeling paper (Codex-oriented). Niche but genuinely scholarly and specific; ships no license.
• nature-paper-skills - by boom5426 · MIT · net.A journal-first manuscript pipeline for Nature-style submissions — citation-integrity, claim/evidence discipline, submission preflight and rebuttal. Opinionated and narrow by design; strong for high-end bio/medical writing.
• nature-writing-skill - by syntaxsmith · No License.Nature-family (NMI/NC/NCS/Nature) paper-writing craft for Chinese AI/ML researchers, built from sentence-level extraction of 44 open-access papers with a grep-traceable corpus and a sister figure skill. Niche to Nature-style AI methods papers.
• paperorchestra - by ar9av · NOASSERTION · net.Portable skill pack implementing Google's PaperOrchestra: turns agent experiment logs into idea+log inputs, then outline→writing→refinement→plotting with autoraters to a LaTeX paper. Best when experiments already exist, not literature-first.
• paperspine - by wubing2023 · MIT · net.Ten-skill academic-writing suite covering discovery, literature synthesis, argument-building, manuscript revision, citations, figures, reviewer responses, and submission. Strong end-to-end coverage, bilingual EN/zh; orchestration-heavy.
• paper-orchestra-for-claude-code - by sunjongos · MIT · net.An ambitious 12-agent Korean+English SCI-paper pipeline, but its SKILL.md leans on grand "surpasses everything" marketing — read warily.
• paper-writing-skill - by snl-ucsb · MIT.Encodes one systems lab's battle-tested paper-writing methodology (brainstorm to compress) with rhetorical-move templates and a figure-synthesis guide. Strongest for CS/systems venues; opinionated voice may not transfer to every field.
• trivium - by MetaQiu · MIT · net.Drafts each paragraph by Claude/Codex/Gemini consensus, betting three-model-approved text also satisfies LLM reviewers — needs all three CLIs.
• ageaf - by onireimu · MIT · net.Brings AI agents (Claude, OpenAI and others) directly into Overleaf for LaTeX paper writing and editing. A focused, genuinely academic integration for the dominant collaborative LaTeX editor.
• awesome-latex-skills - by calix-l · MIT · net.LaTeX paper-writing suite: fix compile errors, polish academic prose, format for venue templates, read papers, and recover lost source. Tight and practical for manuscript work; prompt-pack style, so depth depends on the agent.
• latex-paper-skills - by yunshenwuchuxun · MIT · net.Modular LaTeX paper-writing bundle for ML/AI: topic-to-compiled-PDF with verified BibTeX citations, gated workflows, a results-backfill step and a prose 'rhythm refiner', plus multi-agent (Claude+Gemini) collaboration.
• academic-writing-agents - by andrehuang · MIT.Claude Code plugin running 12 specialist agents (logic, technical, prose, bibliography, LaTeX reviewers) in parallel over a draft, diagnosing then fixing with human approval. Strong for structured manuscript review; not a from-scratch writer.
• stats-paper-writing-agent-skills - by fuhaoda · No License.A statistics-writing LaTeX workbench: drafts abstracts/keywords, audits a manuscript for structure, references, notation and reproducibility cues, writes referee reports and point-by-point response letters, and ships check_tex.py/check_bib.py.
• chinese-thesis-workbench-skill - by zyhsechub · MIT.Turns a student project (source code, database, screenshots, references) into a traceable, resumable, reviewable undergraduate thesis following a university template and sample-paper style. Great for CN capstones; not a general academic-writing tool.
• humanities-thesis-skill - by ganzhi-black · MIT · net.A Chinese-first humanities/social-science thesis copilot that stresses thinking the argument through and auditing every paragraph over generating text. Strong anti-fabrication discipline.
• physics-lab-report-skill - by ydh0411 · MIT.Generates pre-lab and formal university-physics lab reports in spec-compliant LaTeX, with real data-processing built in: uncertainty propagation, the method of successive differences, least-squares fitting and uncertainty analysis.
• skill-thesis-writer - by yanlin-cheng · MulanPSL-2.0.A Chinese-academic thesis-writing skill for undergraduate and graduate work across engineering, psychology, education and management, enforcing GB/T 7714-2015 reference formatting with built-in data-analysis and citation management.
• academic-paper-skills - by Li Shixiong (lishix520) · MIT.A rare humanities-focused drafting pipeline for philosophy and preprint venues, with citation-backed gap analysis — now unmaintained.
• lit-review - by bethww · MIT.Guides a literature review as an argument across seven phases (question, method, synthesis) instead of summarizing sources. Bilingual EN/zh; explicitly will not write the paper for you.
• litllm - by litllm · Apache-2.0 · net.RAG-based literature-review assistant (TMLR 2025) that drafts structured related-work sections from a paper set, as a web app or Claude Code skill. Strong for related-work synthesis; narrow (related-work, not full-paper) and demo-hosted.
• claude-code-my-workflow - by Pedro H. C. Sant'Anna (pedrohcgs) · MIT · net hooks bypass.An econometrician's fork-ready, 36-command Claude Code setup from a PhD course — but ships with bypass-permissions on by default.
• paper-discipline-skills - by lambenthan · MIT.12 zh research-writing 'discipline' skills from a real book of mistakes: backup-before-Word, terminology protection, pilot-before-batch, logical-consistency, parallel-audit. Each has a rationalization table to stop skipping. Guardrails, not content.
• econ-writing-skill - by Lu Han (hanlulong) · MIT.The best-sourced discipline writing skill here, distilling 50+ named economics guides into enforceable rules.
• english-writing - by yzy1996 · No License.Academic-English polishing skill backed by a curated per-section phrasebook (abstract, intro, related work, method, experiment, rebuttal) of good usages mined from real papers. Great phrasing help for non-native CS/ML writers; no LICENSE file.
• journal-adapt-writing-skill - by wantongc · MIT.Goes beyond generic style rules by building a corpus-grounded 'dynamic skill' from a target journal's own papers, then revising section by section. The differentiator is journal adaptation; needs you to supply the reference corpus.
• research-paper-writing - by Xudong Cai (Master-cai) · MIT.Packages Peng Sida's well-known research-writing notes into paragraph-level rewriting discipline for ML/CV/NLP papers.