aperivue/medsci-skills
44 skills2k installs10.4k starsGitHub
Install
npx skills add https://github.com/aperivue/medsci-skillsSkills in this repo
1Meta AnalysisMeta-analysis is a Claude skill that runs the full systematic review and meta-analysis pipeline for medical research. It covers PROSPERO protocol registration, search strategy, screening, data extraction, risk-of-bias assessment, statistical synthesis, and PRISMA-compliant reporting. It supports diagnostic test accuracy meta-analyses with bivariate/HSROC models and intervention meta-analyses with random-effects models.57installs2Analyze StatsAnalyze Stats generates reproducible Python or R code for statistical analyses in medical research, producing publication-ready tables and figures that follow journal standards. A researcher uses it to auto-detect the analysis type, propose a plan, and generate code for diagnostic accuracy, agreement, meta-analysis, survival, survey-weighted, regression, propensity-score, and repeated-measures designs. It includes a PHI check and journal-specific table profiles.52installs3Check ReportingCheck-Reporting audits a medical manuscript against the appropriate reporting guideline and produces an item-by-item compliance report suitable for journal submission. It supports 36 guidelines including STROBE, CONSORT, STARD, TRIPOD, PRISMA, CLAIM, and risk-of-bias tools like QUADAS-2, RoB 2, and PROBAST. A researcher uses it to mark each checklist item PRESENT, MISSING, or PARTIAL before submitting a paper.52installs4Make FiguresMake-figures is a Claude skill that generates publication-ready figures and visual abstracts for medical research papers. It produces statistical plots (ROC, forest, calibration, Kaplan-Meier, Bland-Altman, confusion matrices) and flow diagrams (CONSORT, STARD, PRISMA) that meet journal specifications for dimensions, resolution, fonts, and color accessibility. It checks for PHI and follows journal AI-image policies before generating.52installs5Find JournalFind-journal is a journal recommendation engine for medical manuscripts. Given an abstract, key findings, and study type, a researcher uses it to match against a curated public profile library plus optional private profiles and get the top 5 ranked journals with scope-fit rationale and AI-disclosure policy. It carries no cached impact-factor or APC data; users verify current metrics at journal sites.51installs6Fill Icmje CoiFill-icmje-coi batch-generates per-author ICMJE Conflict of Interest Disclosure .docx forms for manuscript submission. A researcher uses it when a paper with several co-authors and no real financial conflicts is being submitted to an ICMJE journal. It clones a synthetic all-None seed per author with the date, name, and title replaced, avoiding manual Word clicking.48installs7Peer ReviewPeer-review is a Claude skill that helps a medical researcher write structured peer reviews for journals. It analyzes the manuscript systematically, runs a task-formulation audit and data-leakage and conditioning checks, applies a methodology issue checklist, and produces a journal-formatted review draft with a constructive tone. It is for reviewing others' manuscripts, not one's own paper.48installs8Find Cohort GapFind-cohort-gap discovers novel, publishable research topics from a longitudinal cohort database. A researcher uses it to profile cohort variables, match PI expertise, scan literature saturation, and produce ranked topic proposals with gap evidence. Unlike PICO or Elicit tools that work from literature to gaps, it works from the data outward.47installs9Self ReviewThis skill runs a pre-submission self-review of a researcher's own manuscript, applying the same critical lens as a peer reviewer. It checks the paper across 10 categories, flags issues as Fatal or Fixable, and outputs an actionable list of anticipated reviewer comments with fix suggestions. A researcher uses it before journal submission to strengthen the manuscript before real reviewers see it.47installs10Academic AioAcademic AIO reviews titles, abstracts, structured summary boxes, manuscripts, README/CITATION.cff, and model cards so a medical AI paper is surfaced and cited accurately by AI search engines and RAG literature tools. A researcher uses it to apply generative-engine-optimization principles alongside reporting-guideline requirements. It outputs a visible pass/fail checklist with concrete edit suggestions rather than editing silently.46installs11Add JournalAdd Journal extracts a journal's metadata from its author guidelines and generates two canonical reference profiles: a detailed write-paper profile and a compact find-journal profile. A medical researcher uses it to extend the MedSci Skills journal database with a new venue. It enforces a duplicate check and public-versus-private tier verification before writing profiles.46installs12Author StrategyAuthor Strategy fetches an author's PubMed publication portfolio, classifies each paper's study type and author position, and produces a CSV dataset, seven visualizations, and a strategy report to reverse-engineer their research strategy. A researcher uses it to surface topic clusters, growth trajectory, and replicable publication patterns. An optional gated step classifies the author's trajectory into career archetypes after a disambiguation review.46installs13Calc Sample SizeCalc-Sample-Size walks a medical researcher through test selection with a decision tree, then generates reproducible R and Python code for sample-size and power calculations. It supports diagnostic accuracy, agreement, proportions, continuous outcomes, survival, ANOVA, logistic regression, and non-inferiority/equivalence designs. It also interprets effect sizes clinically and produces IRB-ready justification text for the study protocol.46installs14HumanizeHumanize detects and removes AI writing patterns from academic manuscripts and response-to-reviewers letters. A medical researcher uses it to scan for 24 common AI-generated text patterns, then rewrite flagged passages to read as naturally human-written while preserving every technical claim, number, and citation. It runs a scan, report, fix, and verify loop targeting a low AI-pattern density.46installs15Ma ScoutMa-scout is a Claude skill for discovering and vetting meta-analysis topics in medical research. It runs either professor-first (build an author profile from PubMed, then find gaps) or topic-first (start from a clinical question and find gaps). It uses the PubMed E-utilities API and MeSH clustering to produce a ranked, feasibility-assessed topic list before any protocol is written.46installs16OrchestrateOrchestrate is a Claude skill that routes research requests across the medsci-skills bundle. When a user describes a research goal without naming a specific skill, or when a task spans multiple skills, it classifies the request and delegates to the right skill or chains several in order. It plans and routes rather than doing the work itself.46installs17Search LitThis skill runs literature searches and manages citations for medical research. It queries PubMed, Semantic Scholar, and preprint servers, deduplicates and verifies each reference against a live API, and generates BibTeX entries. A researcher uses it to build a verified reference library for a manuscript section without fabricated citations.46installs18Verify RefsThis skill audits a manuscript's or bibliography's references against PubMed, CrossRef, and OpenAlex to catch fabricated or mismatched citations before submission. It cross-checks author names index-by-index, flags duplicate PMIDs and DOIs, and writes a single reference_audit.json without modifying the bibliography. A researcher runs it before submission, especially on manuscripts inherited from coauthors or produced with AI assistance.46installs19Write PaperThis skill runs the full pipeline for writing a medical or scientific manuscript, moving through an 8-phase IMRAD workflow from outline to submission-ready draft. It supports many paper types, loads journal profiles and paper-type templates, and selects the matching reporting guideline. A researcher uses it to draft a publication-quality manuscript with the right structure and disclosures.46installs20Batch CohortBatch Cohort generates many analysis scripts from a single validated methodology template, each differing only in the exposure/outcome variable combination. A researcher uses it to systematically explore multiple exposure-to-outcome pairs on one database (KNHANES, NHANES, NHIS, or any cleaned cohort) and collect results into a summary matrix. It identifies template slot variables, adjusts covariates per combination, and generates a master runner script.45installs21Design StudyDesign-study is a study-design and validity review for radiology and medical AI research. A researcher runs it when the study question is known but the analysis plan is still fluid to catch leakage, analysis-unit mismatches, and comparator problems before drafting. It outputs a structured review with strengths, major validity risks, minimal fixes, and a recommended reporting guideline.45installs22Fill ProtocolFill-protocol populates institutional Word form templates for IRB protocols, ethics applications, and grant proposals without breaking the original formatting. A researcher uses it to render drafted content into an institutional template by opening the file and replacing cell or paragraph text in place. It pairs with write-protocol, which drafts the scientific content.45installs23Fulltext RetrievalFulltext-retrieval batch-downloads open-access full-text PDFs from a DOI list using legitimate open-access APIs (Unpaywall, PMC, OpenAlex, Crossref). A researcher uses it to gather papers for a meta-analysis or literature review and optionally convert the PDFs to Markdown for token-efficient repeated LLM analysis. It only retrieves open-access articles; paywalled papers need institutional access.45installs24Grant BuilderGrant-builder supports competitive proposal writing for radiology and medical AI projects, including national R&D grants, consortia, and Korean government grants. A researcher uses it to structure significance, innovation, approach, milestones, and consortium roles while keeping claims evidence-based and executable. It decodes the funding call, frames the problem, and builds the proposal spine section by section.45installs25Intake ProjectIntake-project is a Claude skill that turns a new or disorganized medical research folder into a structured project state. It classifies the project type, domain, and target output, reconstructs what exists versus what is missing, and scaffolds lightweight project memory files. It enforces a canonical manuscript and submission folder layout so later skills can work safely.45installs26Manage RefsManage-refs is a Claude skill that routes reference-handling work for a medical manuscript. It validates citation keys against refs.bib, renders the manuscript with journal-specific CSL via pandoc citeproc, runs manuscript-to-DOCX cross-reference QC, converts numeric markers to citekeys, and injects Zotero CWYW field codes. It writes references, while a separate audit skill only reads.45installs27Present PaperPresent-paper is a Claude skill that prepares academic presentations from a research paper. It runs a five-phase pipeline: analyze the source paper, find supporting references, draft an audience-adapted speaker script, generate or augment a PPTX with speaker notes, and prepare Q&A. It supports journal club, grand rounds, seminar, conference, and lecture formats.45installs28Render Pdf DocRender-pdf-doc is a Claude skill that renders academic Markdown documents to publication-quality PDF via pandoc and xelatex. It targets non-bibliography artifacts such as research proposals, IRB cover letters, briefings, and anchor docs. It auto-infers pipe-table column widths, applies CJK-aware font fallback for Korean, and scans for glyphs the font would silently drop. It is not for manuscripts with a bibliography.45installs29ReviseThis skill parses peer-reviewer decision letters and generates a formal Response to Reviewers document with tracked manuscript changes. It numbers and classifies every comment, coordinates new analyses or figures with sibling skills, and produces the editor cover letter. A researcher uses it during a journal revision round to answer reviewers point by point without introducing numerical errors.45installs30Write ProtocolThis skill drafts a research protocol for IRB or ethics-committee submission. It fully writes four core sections (Background, Study Design, Sample Size, Statistical Plan) and provides six skeleton sections with TODO markers for institution-specific content. A researcher uses it to produce the scientific core of a protocol before adapting it to their institution's IRB form.45installs31Clean DataClean Data is an interactive profiling and flagging assistant for clinical CSV/Excel datasets that runs a three-stage workflow: profile the data, flag potential issues, then generate cleaning code. A researcher uses it to surface missing values, outliers, duplicates, type mismatches, structural zeros, and reverse-coded scale items with approval gates at each step. It generates code and reports but does not auto-clean data, since every decision requires confirmation.44installs32Cross NationalCross-National runs an end-to-end cross-national comparison study using parallel nationally representative surveys such as KNHANES, NHANES, and CHNS. A researcher uses it to harmonize variables across surveys, run country-stratified survey-weighted analyses, and produce side-by-side comparison tables for a clinical question. It supports 2-country and 3-country designs and enforces that data is never pooled across countries.44installs33DeidentifyDeidentify de-identifies clinical research data before LLM-assisted analysis using a standalone Python CLI that runs locally without network or AI calls. A medical researcher runs it to strip PHI from CSV, TSV, or Excel datasets using country-specific locale pattern packs. The LLM only guides and verifies; it never sees the raw patient data.44installs34Lit SyncLit-sync is a Claude skill that synchronizes research references from .bib files into a Zotero library and Obsidian literature notes. It parses BibTeX entries, deduplicates and adds them to a Zotero collection, refreshes the manuscript refs.bib snapshot, and creates per-reference notes. When enough literature notes accumulate it extracts cross-cutting concept notes.44installs35Publish SkillPublish-skill is a Claude skill that converts a personal agent skill into a distributable, open-source-ready package. It runs a seven-phase pipeline: identify the source, apply a skill-worthiness gate, check originality, run a zero-tolerance PII de-identification audit, generalize language, verify license compatibility, and package for commit. It stops publication when a skill fails the worthiness or originality gates.44installs36Replicate StudyThis skill helps a medical researcher replicate a published cohort study's methodology on a different database. It reads the source paper, maps each exposure/outcome/covariate to the target database through a harmonization table, generates reproducible analysis code, and produces a report documenting every deviation. A researcher uses it to take a validated methodology from one dataset and apply it to another, such as a cross-national replication.44installs37Sync SubmissionThis skill keeps a canonical manuscript and its journal-specific submission packages from drifting apart. It audits, builds, and freezes submission packages via deterministic scripts, records whether each package is current, stale, or frozen, and runs a pre-flight gate that halts the freeze on unresolved errors. A researcher uses it right before submitting a journal package or when retargeting to another journal.44installs38Define VariablesDefine-variables turns a data dictionary and a research question into a citation-backed table of exposure, outcome, and covariate definitions with cutoffs and DB variable mappings. A medical researcher uses it after study design and before writing the protocol Methods section. It forces a literature-first operationalization so reviewers do not reject on construct validity.43installs39Setup MedsciThis skill is a read-only diagnostic checklist for the MedSci Skills runtime. It detects the OS, checks for Python, R, Node, Git, Claude Code CLI, and configured MCP servers with minimum versions, and prints a pass/fail table pointing to the setup doc for any missing piece. A researcher runs it to confirm their environment is ready before using the other MedSci skills.40installs40Polish LanguagePolish-language is a Claude skill that lints academic English consistency and does ESL clarity polish for medical manuscripts. A deterministic linter flags seven families of defects (abbreviations, US/UK spelling, en-dash ranges, p-value case, hyphenation, small numbers, unit spacing) with line numbers, then applies user-approved, style-only fixes. It never changes numbers, citations, or scientific meaning.38installs41Design Ai BenchmarkingDesign-ai-benchmarking reviews and designs studies that benchmark AI systems against a human-expert panel as the reference. A researcher uses it before data collection to lock the evaluation question, arm definitions, rubric, calibration probes, and judge strategy. It guards against unfair comparisons, tautological rubric items, and uninterpretable reliability.37installs42Generate CodebookGenerate-codebook turns a raw tabular dataset into a citable data dictionary. A researcher runs the bundled deterministic profiler to record each variable's role, type, level frequencies, ranges, and missingness, emitting codebook.md and codebook.json. It describes what is in the data but flags coded values whose meaning is unknown as [NEEDS DICTIONARY] rather than inventing them, feeding define-variables.37installs43Review PaperThis skill scaffolds and drafts a medical or medical-AI literature review as a narrative, scoping, or systematic review. It builds a 7-part skeleton, forces a scope and non-overlap statement, generates summary-table stubs, and wires in the correct reporting guideline. A researcher uses it to structure a review article and then hand off to the QC skills before submission.37installs44Version DatasetThis skill puts a research dataset under version control by fingerprinting it, detecting drift, and locking a reproducible version. It builds a deterministic content-hash manifest, verifies later copies against it, and diffs two manifests to show what changed. A researcher uses it to prove an analysis ran on the intended data and to make any silent data change loud before re-running.37installs