Medical Research Skills for Claude Code, Codex, Cursor, and Copilot

Single entry point for the full bundle. Classifies your request and routes to the right skill — or chains multiple skills end-to-end via Full Pipeline Mode.

• ✓Automatic intent classification
• ✓End-to-end pipeline: skills chain automatically
• ✓Project context detection

Read-only setup diagnostics for MedSci Skills. Checks local Python/R/Node/Claude Code and MCP availability, then points users to the right onboarding guide without installing or changing anything.

• ✓Read-only environment check
• ✓Python/R/Node/Claude Code readiness report
• ✓Links missing components to setup docs

PubMed + Semantic Scholar + bioRxiv search with anti-hallucination citation verification. Token-efficient error handling.

• ✓Every reference verified via API
• ✓BibTeX export with verified / verified_by / verified_on fields
• ✓No fabricated citations

Pre-submission reference audit for .md / .docx / .bib / .tsv inputs. Extracts references, verifies DOI/PMID via CrossRef + PubMed, and writes qc/reference_audit.json as the sole source of truth. Blocks fabricated references before submission. Gate 5 PMID/DOI duplicate detection.

• ✓DOI + PMID verification via CrossRef & PubMed
• ✓Writes only qc/reference_audit.json; never edits your references or .bib
• ✓PostToolUse hook auto-runs on manuscript saves (auto / warn / enforce modes)
• ✓Duplicate detection (PMID/DOI)

Batch open-access PDF downloader. Unpaywall → PMC → OpenAlex → CrossRef pipeline. OA-only — no paywall bypass.

• ✓DOI list or TSV input
• ✓PDF validation built-in
• ✓Manual-needed list for paywalled papers

Manuscript compliance audit against reporting guidelines and risk-of-bias tools with item-by-item assessment and section boundary checks.

• ✓STROBE, STARD, STARD-AI, PRISMA, CONSORT, TRIPOD+AI, CARE, MI-CLEAR-LLM
• ✓QUADAS-3, QUADAS-2, RoB 2, ROBINS-I, PROBAST, NOS
• ✓Results/Discussion boundary check

Identifies analysis unit, cohort logic, data leakage risks, and validation strategy.

• ✓Leakage risk detection
• ✓Comparator design review
• ✓Reporting guideline matching

Reproducible Python/R code for 22 analysis types including regression, propensity score, and repeated measures. Calibration mandatory for prediction models.

• ✓Regression (logistic/linear), propensity score (PSM/IPTW), repeated measures (GEE/mixed)
• ✓DTA, survival, demographics, agreement
• ✓Publication-ready tables with journal-specific formatting

Journal-spec figures and visual abstracts: ROC curves, forest plots, flow diagrams, Kaplan-Meier, and more.

• ✓300 DPI, colorblind-safe
• ✓PRISMA / CONSORT / STARD flows
• ✓Visual/graphical abstracts (python-pptx template)

Structures significance, innovation, approach, milestones, and consortium roles.

• ✓Korean government grant mode (three-attachment format)
• ✓Specific aims page
• ✓Budget justification

Classifies new research projects, summarizes state, and recommends next steps.

• ✓Project type classification
• ✓Missing input detection
• ✓Lightweight scaffolding

Academic presentation prep: paper analysis, speaker scripts, slide note injection, Q&A.

• ✓Audience-adapted scripts
• ✓PPTX note injection utility
• ✓Multi-perspective Q&A prep

Convert personal Claude Code skills into distributable, open-source-ready packages.

• ✓Automated PII audit
• ✓License compatibility check
• ✓Generalization pipeline

Full 10-phase systematic review and meta-analysis pipeline. DTA (bivariate/HSROC) and intervention MA with PRISMA-DTA compliance. Dual-extractor + cohort overlap detection.

• ✓Protocol to manuscript pipeline
• ✓Bivariate & HSROC models
• ✓PRISMA-DTA compliant
• ✓Dual-extractor SR-MA
• ✓Cohort overlap detection

Full 8-phase IMRAD pipeline from outline to submission-ready manuscript with built-in critic-fixer loops and section boundary enforcement.

• ✓Anti-interpretation guardrails in Results
• ✓Interactive Discussion with anchor papers
• ✓AI writing pattern detection

Pre-submission check across 12 categories with research-type branching (AI, observational, educational, meta-analysis, case report, surgical).

• ✓Major / Minor severity classification
• ✓Calibration mandatory for prediction models
• ✓R0 numbering for /revise pipeline

Parses reviewer decision letters, classifies comments as MAJOR/MINOR/REBUTTAL, and generates point-by-point responses.

• ✓Structured response templates
• ✓Change log with page/line refs
• ✓Cover letter generation

Scaffold research projects, track writing progress, generate submission checklists and backwards timelines.

• ✓Project scaffolding with --zotero-collection auto-create (pyzotero)
• ✓Phase progress tracking
• ✓Pre-submission checklist

SSOT-to-submission drift audit and journal package helper. Treats submission/{journal}/ as derived output, records source hashes in .journal_meta.json, and blocks freezing drifted packages.

• ✓SHA-256 source-hash drift detection
• ✓Per-journal build / audit / freeze commands
• ✓Blocks submission when SSOT has diverged

Interactive sample size calculator with decision-tree guided test selection. Covers 17 designs including Cox regression EPV.

• ✓DTA, t-test, ANOVA, chi-square, McNemar, survival
• ✓Logistic regression & Cox EPV (Peduzzi rule)
• ✓IRB-ready justification text

Standardize, validate, and transform raw research datasets. Handles missing data, outlier detection, and variable recoding.

• ✓Missing data summary & imputation
• ✓Outlier detection (IQR, Z-score)
• ✓Codebook generation

Match your manuscript to target journals by scope, impact factor, turnaround time, and open-access policy.

• ✓Scope & audience matching
• ✓Acceptance-readiness pre-flight and reject-fallback cascade
• ✓OA fee & turnaround estimates

Draft an IRB or ethics research protocol: Background, Study Design, Sample Size, and Statistical Plan in full prose, with institution-specific sections left as TODO skeletons.

• ✓Four core sections written in full prose
• ✓Institution-specific sections marked TODO, never invented
• ✓Hands the institution's Word form to /fill-protocol

Add new journal profiles to the database. Extracts metadata from author guidelines and generates dual profiles (write-paper + find-journal) with quality gates.

• ✓Author guideline extraction
• ✓Dual profile generation
• ✓Batch mode for multiple journals

De-identify clinical research data before LLM-assisted analysis. Standalone Python CLI with 11 country locale packs. No LLM involved.

• ✓PHI detection via regex + heuristics
• ✓Interactive terminal review
• ✓Date shifting & pseudonymization

Research gap finder for longitudinal cohort databases. 6-phase pipeline (cohort intake → PI profiling → intersection matrix → literature saturation scan → 6-Pattern scoring → feasibility gate) that outputs ranked topic proposals with comparison tables and one-pagers.

• ✓Works with any cohort: NHIS, UK Biobank, institutional EMR, health checkup registries
• ✓6-Pattern gap scoring with feasibility gate
• ✓Ranked proposals with one-pager output

Replicate an existing cohort study on a different database. Extracts methodology from a source paper, maps variables via harmonization table, generates analysis code, and produces a replication difference report.

• ✓5-phase workflow: source analysis → variable mapping → code generation → difference report → verification
• ✓Built-in KNHANES/NHANES harmonization table
• ✓Validated on Joo 2026 cross-national replication

End-to-end cross-national comparison study using parallel survey data. Variable harmonization, parallel weighted analysis (no data pooling), and country-stratified comparison tables.

• ✓KNHANES + NHANES + CHNS variable coding references
• ✓Country-specific survey design with no data pooling
• ✓Supports 2-country and 3-country designs

Generate N analysis scripts from one validated template × multiple exposure/outcome combinations. The "80-person team" pattern: same method, swap variables only.

• ✓Self-adjustment prevention & EPV checks
• ✓Bonferroni correction & summary heatmaps
• ✓Validated on KNHANES and on cross-national combinations

Fill Korean IRB and institutional Word form templates (.doc/.docx) while preserving original layout, cell structure, and formatting. Extracts content from your protocol draft and maps it to labeled form cells.

• ✓Preserves original form structure — no layout breakage
• ✓Cell label matching for Korean IRB forms
• ✓LibreOffice converts a legacy .doc template to .docx

Batch-fill ICMJE Conflict of Interest Disclosure forms (.docx) for co-authors from a pre-filled seed document. Replaces date, author name, and manuscript title while keeping the 13-item disclosure block and checkbox states intact.

• ✓Ships with a PII-free synthetic seed (safe to redistribute)
• ✓Literal XML replacement inside Word Content Controls (SDTs)
• ✓Post-run verification: 14 ☒, 13 "None", zero placeholder leakage

Detect and remove AI writing patterns from academic manuscripts. Scans for 27 patterns — significance inflation, AI vocabulary, copula avoidance — and rewrites flagged passages while preserving technical accuracy.

• ✓27-pattern detection with per-instance flagging
• ✓Density target: <2.0 AI instances per 1,000 words
• ✓Rewrites preserve all numbers, p-values, and clinical terms

PubMed author profile analysis. Fetches publication history via E-utilities, classifies study types (GBD, SR/MA, NHIS, AI/ML, etc.), generates 7 visualizations, and produces a strategy report with replication opportunities.

• ✓E-utilities batch fetch with deduplication
• ✓Study type classifier: GBD, SR/MA, NHIS, AI/ML, cross-national, case series
• ✓Replication opportunity scoring with co-author gap analysis

Structured peer review drafting for medical imaging journals. Systematic manuscript analysis, journal-specific formatting (RYAI, INSI, EURE, AJR, KJR), length targets (700–1,000 words by default), and pre-submission QC checklist. Phase 2A SR-MA probe extension (P1–P19).

• ✓Major / Minor / Reject recommendation with rationale
• ✓Journal-specific format presets for 5 radiology journals
• ✓Pre-submission QC: word count, tone, ethical red flags
• ✓SR-MA Phase 2A (P1–P19)

Meta-analysis topic discovery and feasibility assessment. Two modes: Professor-first (profile → pillar analysis → MA gaps) or Topic-first (question → landscape scan → co-author matching). Multi-source validation with realistic k estimation.

• ✓Professor-first and Topic-first dual modes
• ✓Multi-source validation: PubMed, PROSPERO, bioRxiv
• ✓Realistic k estimation with 15–30% registration discount

Sync research references from .bib files to Zotero library + Obsidian literature notes. Concept extraction from 10+ accumulated notes with cross-cutting theme discovery. Works after /search-lit or standalone.

• ✓Zotero import via Better BibTeX key matching
• ✓Obsidian literature note generation with wikilinks
• ✓Cross-cutting concept extraction from 10+ notes

Turns a folder of research PDFs into Obsidian literature notes, one per paper, plus cross-linked concept notes. Existing notes are never overwritten.

• ✓One templated literature note per PDF
• ✓Cross-linked atomic concept notes synthesized across papers
• ✓Numbers, authors, and dates transcribed from the PDF text; existing notes never overwritten

Visibility checklist for medical AI papers against AI search engines (Perplexity, ChatGPT web, Elicit, Consensus, SciSpace) and RAG literature tools. Integrates TRIPOD+AI, CLAIM, STARD-AI, TRIPOD-LLM, DECIDE-AI with generative-engine-optimization (GEO) principles.

• ✓Covers title, abstract, Key Points, Plain-Language Summary, preprints, README, CITATION.cff, Zenodo, Hugging Face cards
• ✓Defends against LLM citation fabrication
• ✓Visible PASS / PARTIAL / FAIL checklist — never applies silent edits

Citekey ↔ .bib validation, journal-CSL pandoc rendering, manuscript ↔ rendered DOCX cross-reference QC, [N] ↔ [@key] marker conversion, and native Zotero CWYW field-code injection for co-author Word workflows. Sole writer of manuscript_final.docx and qc/xref_audit.json. Hybrid 3-phase strategy (pandoc draft → CWYW transition → Zotero CWYW for circulation/revision/submission).

• ✓check_citation_keys.py + check_xref.py --strict as the submission gate
• ✓Bundled journal CSLs: ER, Radiology, AJR, CVIR, KJR, Vancouver — cascade-rejection re-render in seconds
• ✓Native CSL-JSON CWYW injection (handles webpage / report / non-journal item types correctly)

Render non-bibliography academic markdown (proposal, briefing handout, anchor doc, IRB cover, reference table) to publication-quality PDF via pandoc + xelatex. Boundary opposite of manage-refs — non-bib, layout-focused.

• ✓CJK font fallback: Apple SD Gothic Neo (macOS), Noto Sans CJK KR (Linux)
• ✓Content-proportional pipe-table column widths — first column shrinks to fit, data columns share the remainder
• ✓Frontmatter redact_internal flag for circulation PDFs (no version numbers / changelog / PI attribution leakage)

Offer your local changes back to the project — a journal profile you added, a checklist item you fixed, a skill you adapted to your department — as a pull request or an issue, without ever typing a git command.

• ✓Diffs your copy against the installed version
• ✓Scans for patient data and identifiers before anything leaves
• ✓Shows you every line and sends nothing until you confirm

Literature-grounded variable operationalization for observational research. Turns a data dictionary plus research question into a citation-backed table of exposure / outcome / covariate definitions, cutoffs, and DB-variable mappings. Tier 0 dictionary-first rule prevents ad-hoc phenotype definitions that invite reviewer rejection.

• ✓Dictionary-first Tier 0 enforcement (verbatim citation of codebook entries)
• ✓Bridges /search-lit output into /write-protocol Methods
• ✓Per-variable citation provenance (source / row / verbatim mapping)

Design and validity review for studies benchmarking AI systems against a human-expert panel. Covers arm definitions, multi-dimensional rubrics with calibration probes, reviewer panels, inter-rater targets, and LLM-as-judge vs human adjudication.

• ✓AI-vs-expert evaluation design
• ✓Calibration probes & inter-rater targets
• ✓LLM-as-judge vs human adjudication

Generate a citable data dictionary / codebook from a tabular dataset (CSV, Excel, Parquet, Stata, SAS). Profiles every variable — role, type, level frequencies, range, missingness — and flags coded variables with unknown meanings as [NEEDS DICTIONARY] rather than guessing.

• ✓Per-variable profiling → codebook.md + .json
• ✓Flags unknown codes, never guesses
• ✓Feeds /define-variables (dictionary-first)

Dataset version control for reproducibility. Builds a deterministic content-hash manifest (file SHA-256 + schema + per-column value hashes), verifies later copies to detect drift, and diffs two manifests — proof an analysis ran on the intended data.

• ✓Content-hash manifest (SHA-256 + schema)
• ✓Drift detection: schema / rows / values
• ✓Reproducibility-lock datasets & demos

Scaffold and draft a literature review — narrative (SANRA), scoping (PRISMA-ScR + JBI), or systematic (PRISMA 2020). Builds a 7-part skeleton with a required scope and non-overlap block, then hands off to self-review, check-reporting, and verify-refs.

• ✓Narrative / scoping / systematic modes
• ✓Required scope and non-overlap framing
• ✓Never invents citations — verify-refs handoff

Academic English consistency linting plus non-native (ESL) clarity polish. A deterministic linter flags abbreviation, US/UK spelling, en-dash range, P/p case, hyphenation, number-style, and unit-spacing issues; a gated pass then fixes wording without touching numbers, citations, or scientific meaning.

• ✓8-family deterministic consistency report
• ✓Style-only — never alters numbers or citations
• ✓Distinct from humanize (AI-tell removal)

Paper-grounded architecture choice for medical imaging, then a check of the actual repo or checkpoint: licence, version pin, weight provenance and benchmark overlap.

• ✓Task → architecture decision tree, each archetype tied to its source paper
• ✓Licence, intended use, and pretrained-weight provenance on the record
• ✓Catches an evaluation arm sitting on the model's own development benchmark
• ✓Vets an artifact; never downloads, runs, or benchmarks one

Generate a reproducible, runnable PyTorch training repo for a medical-imaging task — segmentation, classification, detection, synthesis, self-supervised pretraining, or fine-tuning a pretrained backbone — with a patient-level seed-locked split, train/evaluate scripts, and a Methods stub. Integrates MONAI / nnU-Net, never reimplements them.

• ✓Six task templates, leakage-safe by construction
• ✓Auditable patient-level seed-locked split
• ✓Training-hygiene AST gate + network-free build test

Validation design, task-correct held-out metrics (Dice + HD95, AUROC + AUPRC, FROC, calibration), uncertainty/OOD and explainability for trained imaging models, each with a gate.

• ✓Split-leakage gate proves patient disjointness
• ✓Task-correct metrics with bootstrap CIs, computed only from executed code
• ✓Flags unmeasured uncertainty coverage and an unvalidated OOD guard
• ✓Grad-CAM sanity checks and quantitative localisation vs ground truth

Generate the documentation an engineer-built model must carry — a Model Card (Mitchell et al. 2019), a Datasheet for its dataset (Gebru et al. 2021), and a METRIC data-quality pass — filled only from user-supplied facts, then verify every required section is present with a completeness gate.

• ✓Model Card + Datasheet templates
• ✓Facts in, never fabricated; gate checks completeness
• ✓METRIC-informed data-quality pass

Model-agnostic evaluation harness for an LLM or MLLM on a clinical task — report generation, visual question answering, clinical text extraction — covering the adjudicated reference, clinical-efficacy metrics (RadGraph-F1 / CheXbert-F1), faithfulness, contamination, prompt sensitivity, and a reader study.

• ✓Clinical-efficacy metrics beyond BLEU/ROUGE
• ✓Contamination + prompt-sensitivity checks
• ✓Reader study + MLLM reviewer probe

Profiles a DICOM/NIfTI dataset (spacing, orientation, intensity, labels) against the plan, then designs and audits preprocessing and augmentation for leakage.

• ✓Catches a "test set" that carries no ground truth
• ✓Flags label/image grid mismatch and stray label indices
• ✓Data-stage leakage gate (normaliser fit on train only)
• ✓Integrates MONAI / TorchIO transforms

Produce or audit a radiomics / tabular clinical-ML study — features to a classical learner to a clinical outcome — with a nested-CV, feature-stability, and calibration gate. No GPU.

• ✓Nested cross-validation (tuning never on reported folds)
• ✓Feature-stability (ICC) filtering + calibration
• ✓LASSO / SVM / random forest / XGBoost, learner-agnostic