Academic skill pack for Claude Code and other agents

⚡ Try It — One Line, the Right Skill

You don't need to know skill names. Describe the task and alterlab-skill-finder (the front door) picks the skill for you — and tells you which one, and why.

Route — one task → one skill:

💬 "Use AlterLab skills to review my methods section." 🧭 → alterlab-paper-reviewer (methodology-focus)

💬 "Which AlterLab skill finds and fact-checks papers on CRISPR off-target effects?" 🧭 → alterlab-deep-research + alterlab-pubmed

alterflow — a goal that spans stages → a clarified, multi-agent workflow:

💬 "alterflow — investigate the link between sleep and memory, and turn it into a publishable paper." 🧭 → asks 2–4 scoping questions first, then plans and runs:

flowchart LR
    A["⌨️ alterflow"] --> Q{"clarify<br/>2–4 questions"}
    Q --> P["plan + confirm<br/>(with rough cost)"]
    P --> R["🔬 deep-research"]
    R --> W["📝 paper-writer"]
    W --> V["🔍 paper-reviewer"]
    V --> F["📄 final paper"]

Questions before execution — it never starts a 20-agent run blind, and simple tasks stay single-skill. Built on Anthropic's routing and orchestrator-workers patterns.

🚀 What's New in v2.6.0

• 🧭 New front door — alterlab-skill-finder — most users don't know 239 skill names by heart, so this router turns "use AlterLab skills for this" into the right skill(s). It classifies the task, maps it across all 17 domains, names the concrete skill(s) it picked (and why), and applies them — no name-memorization required. → Core Pipeline
• ⚡ alterflow keyword → clarify-first multi-agent orchestration — say alterflow (aliases alterresearch / ultralab) and the router first asks a few scoping questions, then selects the skills the goal needs and launches a dynamic multi-agent workflow composing them (via alterlab-research-pipeline, alterlab-ssci-orchestrator, or alterlab-workflow-orchestration). Questions before execution — it never starts blind.
• 🗂️ Always-current skill index — a generated skill_index.md lists every skill grouped by domain with a one-line "use when"; a new CI test (test_skill_index.py) fails the build if a skill is added or renamed and the index is not regenerated (scripts/gen_skill_index.py --check).
• 📈 239 skills across 17 domains, 239 / 239 with executable evals; audit clean (0 errors, 0 warnings), full test suite green.

Previously — What's New in v2.5.0

• ⚖️ Social-Science Workflow made even-handed (11 → 17 skills) — the domain was quantitative-positivist by construction; v2.5 adds the qualitative and survey methods that were missing, plus depth. → Social-Science Workflow
• 📊 alterlab-survey-analysis (correctness gap) — design-based inference for complex-sample surveys (GSS/ANES/ESS/DHS): declare weights/strata/PSU/FPC before estimating, Taylor + replicate-weight SEs, post-stratification/raking, design-adjusted GLMs. Analyzing weighted surveys unweighted is a real error (falsely narrow CIs); this fixes it. samplics/svy or R survey+srvyr.
• 🗣️ alterlab-qualitative-analysis + alterlab-ssci-reflexivity-gate (balance fix) — intercoder reliability done right (Krippendorff's α primary, + a stdlib bootstrap-CI calculator; chi-square/Cronbach/Pearson refused), branching to consensus-and-reflexivity for reflexive TA; and the qualitative analog of the measurement gate (positionality + Lincoln & Guba trustworthiness, fail-closed).
• 📈 Quantitative depth — alterlab-multilevel-models (mixed-effects with the under-reported reporting checklist), alterlab-meta-analysis (pooling, I²/τ², Egger/trim-and-fill, PRISMA), alterlab-missing-data (MCAR/MAR/MNAR + MICE with Rubin's-rules pooling; the sklearn single-imputation trap flagged).
• 🔬 Every library web-verified & version-pinned (samplics 0.6 / svy 0.18, krippendorff 0.8, statsmodels MixedLM, metafor 5.0, R mice 3.19, lme4 2.0). The orchestrator now paradigm-selects the measurement-stage gate (quant vs qual) and routes the new modules; confusion_matrix.py reports zero new gaps.
• 📈 238 skills across 17 domains, 238 / 238 with executable evals; audit clean (0 errors, 0 warnings), full test suite green.

Previously — What's New in v2.4.0

• 🧭 New 17th domain — Social-Science Workflow — a stage-gated methods spine of 11 skills: an orchestrator, four discipline-enforcing validity gates a study passes through before it is analyzed or written — design (pin the identifying assumption before any causal claim), measurement (reliability is not validity — omega, not alpha alone), sampling (power / precision / saturation, never a rule of thumb), inference (the claim may not exceed its design, sample, or uncertainty) — and six pluggable analysis modules the orchestrator dispatches to.
• 🔬 Six analysis modules, every library web-verified to its current API — alterlab-causal-inference (DiD/IV/RDD/panel via statsmodels, linearmodels, pyfixest, DoWhy, EconML, rdrobust), alterlab-sem-psychometrics (semopy/factor_analyzer + a stdlib omega calculator), alterlab-qca (shells to R's QCA package — no maintained Python library exists), alterlab-sna (networkx/igraph + R ERGM), alterlab-abm-mesa (the current Mesa 3 AgentSet API, not the removed schedulers), and alterlab-text-as-data (BERTopic/LDA/embeddings). No invented flags; version pins in each SKILL.
• 🧪 Executable, not just prose — stdlib helpers (design_router.py, sample_size.py, claim_audit.py, passport.py, estimator_router.py, omega.py, text_method_router.py), loaded-on-demand references, and vignette-in / routing-out evals with near-miss negatives. confusion_matrix.py reports zero new routing gaps.
• 🔗 Composes, does not duplicate — the gates and modules route to the existing alterlab-qualitative-methods, alterlab-survey-design, alterlab-statistical-analysis, alterlab-statsmodels, alterlab-networkx, and alterlab-test-selection-guard skills rather than reimplementing them, threading a Design Passport from design → measurement → sampling → analysis → inference.
• 📈 232 skills across 17 domains, 232 / 232 with executable evals; audit clean (0 errors, 0 warnings), full test suite green.

Previously — What's New in v2.3.0

• 🛡️ Trust manifest you can verify before installing — a generated, CI-attested SECURITY_SCAN.md: the full outbound-host allowlist (66 domains, each categorized) plus an attestation of no shell-pipe, no eval-on-input, no hardcoded secrets. → SECURITY_SCAN.md
• 🎯 Activation + collision benchmarks — run_evals.py --activation measures auto-selection rate against Anthropic's 90% bar, and scripts/confusion_matrix.py statically ranks cross-firing risk between sibling skills. A nightly rotating 1/7 behavioral shard LLM-judges the whole corpus each week. → docs/evals.md
• 🧭 Sharper routing — 42 confusable descriptions now carry explicit prefer alterlab-x routing (quantum, protein/structure DBs, cheminformatics toolkits, single-cell, pathology, plotting, RL).
• 🧩 Agents & teams, formalized — docs/agents-and-teams.md names the 35 subagents and the 4 teams they compose (Research, Writing, Reviewer Panel, Pipeline Orchestration).
• 🔌 Portability map — docs/portability.md states exactly what transfers to Codex / Cursor / Gemini vs. Claude-Code-specific, and check_claudeai_compat.py guards the claude.ai uploader whitelist.
• ✂️ Every body under the soft cap — the last 17 over-500-line skills split into references/; the body-length ratchet is now empty. Coverage holds at 210 / 210. → docs/evals.md

Previously — What's New in v2.2.0

• 🧠 Research-backed elevation of 191 skills — a corpus-wide accuracy and depth pass over every pre-existing skill: corrected regulation/API details, slimmed oversized bodies into references/, and version-bumped touched skills.
• 🧩 New core skill: Workflow Orchestration (alterlab-workflow-orchestration) — composes AlterLab skills into multi-agent workflows (subagent fan-out, pipelines, judge panels, adversarial verification), lifting the corpus to 210 skills. → Core Pipeline
• 🇹🇷 Turkish-academia flagship (12 skills) — the deepest Turkish-academia suite in any skills library: DergiPark, TR Dizin, YÖK Tez, YÖK Akademik, YÖKATLAS, TÜBİTAK 1001/1002-A proposals, doçentlik eligibility, akademik teşvik scoring, Turkish APA-7 style, TR research ethics, KVKK data-management plans, and TÜBİTAK Aperta — each wired to a verified national-platform endpoint. → Turkish Academia
• 🎓 Faculty Life (6 skills) — syllabus AI policy, post-award grant reporting, recommendation letters, AACSB/ABET accreditation (AoL), REDCap/CDISC capture, and preprint deposition. → Faculty Life
• 🧪 Methodology (3 skills) — superpowers-style rigor (Iron Laws, Excuse-vs-Reality tables) applied to research methods: pre-registration discipline, statistical-test-selection guard, and results-reporting transparency. → Methodology
• 🧬 Bioinformatics grew 25 → 30 — production pipelines: nf-core/sarek variant calling, QIIME 2 amplicon, salmon/kallisto RNA-seq quant, NCBI BLAST+, and Squidpy spatial transcriptomics. → Bioinformatics

Previously — What's New in v2.0

• ✅ Full-corpus executable evals — schema + trigger-coverage are CI-gated on every PR; the behavioral pass (claude CLI + LLM judge) runs on demand via workflow_dispatch. → docs/evals.md
• 🔎 Deterministic citation-existence verifier that catches hallucinated references against Crossref / OpenAlex / Semantic Scholar / arXiv — no LLM guesswork in the gate. → Headline feature
• 📦 Per-domain claude.ai bundles on every Release — each dist/<domain>.zip clears the 200-file / 30 MB caps. → Releases
• 🌐 Live searchable catalog to browse every skill in the browser. → Browse the Catalog
• 🤖 35 registered subagents + slash commands + a bundled academic MCP wiring PubMed / OpenAlex / Crossref / Zotero. → MCP & Slash Commands
• 📜 Honest provenance — upstream K-Dense fork credited in THIRD_PARTY_NOTICES.md and a machine-readable CITATION.cff. → Cite This Repository

📋 Table of Contents

Click to expand / collapse

• ⚡ Try It — One Line, the Right Skill
• 🚀 What's New in v2.6.0
• 🎯 What Is This?
• 🆕 v2.0 Highlights
• ✨ Key Features
• 🗂️ Domain Overview
• 🔎 Browse the Catalog
• 🚀 Quick Start
• ⚡ Core Pipeline — 10 Skills
• 📚 All 239 Skills
• 🏗️ Project Structure
• ⚙️ How Skills Work
• 💡 Usage Examples
• ⚖️ How It Compares
• 🤝 Contributing
• 📖 Cite This Repository
• 📜 License
• 🙏 Credits

🎯 What Is This?

A comprehensive suite of 239 purpose-built Claude AI skills for faculty members, academicians, and researchers — organized into 17 domain categories spanning the full academic research lifecycle.

Each skill transforms Claude into a domain-specific expert assistant tailored to academic research, scientific computing, and scholarly publishing workflows.

[!TIP] How it works: Each skill is a structured SKILL.md file. Install the collection as a Claude Code plugin, or copy a skill directory into ~/.claude/skills/ (or a project's .claude/skills/) — and Claude becomes your research expert, with real scientific frameworks, professional output templates, and deep domain knowledge. A loose .md file does not auto-load.

🆕 v2.0 Highlights

v2.0 is the "Trust & Reach" release — it turns the loudest objection to academic skills ("just prompts, no evals") into the headline differentiator.

	What	Detail
✅	Executable evals across the corpus	239 / 239 skills ship evals/evals.json on the canonical agentskills.io schema, validated in CI. See docs/evals.md.
🔎	Citation-existence verifier	New core skill alterlab-citation-verifier — a deterministic gate that checks every reference actually exists against four keyless scholarly APIs (Crossref, OpenAlex, Semantic Scholar, arXiv), flags retractions, and maps claim-faithfulness verdicts. See below.
📦	Per-domain bundles for claude.ai	dist/<domain>.zip bundles (one per domain) attached to each Release — each clears claude.ai's 200-file / 30 MB caps and vendors the shared handoff contracts.
🔌	Bundled academic MCP	The core and databases plugins ship a .mcp.json wiring PubMed / OpenAlex / Crossref / Zotero, with a documented requests/ fallback when no MCP is available.
⌨️	Slash commands	/lit-review, /review-paper, /cite-check, /research-pipeline — drive the core pipeline directly.
🗂️	Generated catalog	A machine-readable skills.json is generated from skill frontmatter; a CI gate keeps every count in this README honest.
📜	Honest provenance	Upstream K-Dense fork credited in THIRD_PARTY_NOTICES.md + PROVENANCE.md / CITATION.cff.

🔎 Headline feature — alterlab-citation-verifier

Around 147 k hallucinated citations surfaced in published work in 2025. Prompt-only "fact-checking" misses roughly a third of them. alterlab-citation-verifier is a deterministic core capability that, for every entry in a bibliography:

• cross-checks existence against Crossref + OpenAlex + Semantic Scholar + arXiv (all keyless),
• resolves DOIs / arXiv IDs and matches title + author with a Levenshtein threshold,
• flags Retraction Watch entries,
• maps each verdict to the suite's TF / PAC / IH / PH / SH faithfulness taxonomy,

and degrades gracefully to plain requests / WebSearch when no MCP or network is present. It rewires the pipeline's bibliography and integrity agents to call it instead of generic web search.

🧭 Headline feature — alterlab-skill-finder (the front door)

No one memorizes 239 skill names. alterlab-skill-finder is the front door: say "use AlterLab skills for this" and it classifies your task, maps it across all 17 domains, and applies the right skill(s) — telling you which it picked and why. For a goal that spans stages, the alterflow keyword turns it into a launcher:

• CLARIFY — asks 2–4 scoping questions before doing anything,
• SELECT & PLAN — picks the skills the goal needs and lays out a dynamic multi-agent workflow, scaling the agent count to complexity,
• CONFIRM — shows the plan and its rough cost, then executes,
• COMPOSE — reuses the existing orchestrators (alterlab-research-pipeline, alterlab-ssci-orchestrator, alterlab-workflow-orchestration) rather than reinventing them.

Built on Anthropic's documented routing and orchestrator-workers patterns — questions before execution, simplest path by default.

✨ Key Features

	Feature	Description
🧭	One-Command Front Door	Just say "use AlterLab skills" — alterlab-skill-finder picks the right skill(s) for you; the alterflow keyword clarifies your goal, then launches a multi-agent workflow
🔬	Research-Ready	Skills built on real scientific methods, databases, and professional frameworks used by working researchers
🤖	Multi-Agent Pipelines	Core skills chain together: Research → Write → Review → Publish in a seamless workflow
📊	39 Database Integration Skills	Instant access to PubMed, ChEMBL, UniProt, ClinicalTrials.gov, COSMIC, and more
🧬	Deep Domain Coverage	From single-cell RNA-seq analysis to quantum computing, from clinical trials to digital humanities
📝	Publication-Quality Output	LaTeX papers, conference posters, grant proposals, scientific visualizations — all formatted to professional standards
🔄	Mix & Match	Combine multiple skills in one Claude Project for a multi-expert research team

🗂️ Domain Overview

	Domain	Skills	Focus Areas
🔄	Core Pipeline	10	Multi-agent research → write → review → publish pipeline + citation verifier + teaching + thesis + workflow orchestration + skill-finder router (the alterflow launcher)
🗄️	Databases	39	Connectors to scientific databases — PubMed, ChEMBL, UniProt, ClinicalTrials.gov, COSMIC, and more
🧬	Bioinformatics	30	Genomics, proteomics, molecular biology — Scanpy, BioPython, ESM, single-cell analysis, nf-core/sarek, QIIME 2, RNA-seq quant
⚗️	Cheminformatics	12	Chemistry and drug discovery — RDKit, molecular dynamics, docking, ADMET
🏥	Clinical Research	7	Clinical decision support, treatment planning, medical imaging, regulatory
📊	Data Science	22	ML/statistics — scikit-learn, PyTorch Lightning, SHAP, transformers
📈	Visualization	8	Scientific plotting — Matplotlib, Seaborn, Plotly, schematics, infographics
✍️	Writing Tools	13	Scientific writing, citations, grants, posters, academic career
🔧	Lab Integrations	9	Laboratory platforms — Benchling, DNAnexus, Opentrons, Protocols.io
🌍	Domain-Specific	17	Quantum computing, geospatial, materials science, social science methods, digital hum