MedSci Skills

55 skills that actually work. Built by a physician-researcher, tested on real publications.

MedSci Skills is an end-to-end research tool for physician and medical-engineering researchers — design → scaffold → validate → publish — for the clinical manuscript and the medical-AI model behind it. Its moat is the compliance layer — 46 reporting guidelines and risk-of-bias tools, reference/citation verification, and deterministic integrity gates before peer review — now extended by a model-engineering lane that scaffolds reproducible, leakage-safe training repos and audits model validation. Clinical AI model research engineering is in scope; a general AI-scientist platform is not. It competes on clinical submission reliability, not skill count.

Topic Discovery → Literature Search → Full-Text Retrieval → Study Design → Sample Size → Protocol → De-identification → Data Cleaning → Statistics → Figures → Writing → Humanize → Compliance → Journal Selection → Peer Review → Revision → Presentation

Created & maintained by Yoojin Nam, MD Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center, Seoul, Republic of Korea

What is MedSci Skills?

MedSci Skills is an open-source Agent Skills collection for clinical research — the manuscript and the medical-AI model alike — designed to be driven directly by AI coding agents (Claude Code, Codex, Cursor, and GitHub Copilot). It helps physician-researchers and biomedical/medical-engineering investigators move from literature search, study design, statistics, and figures to reporting-guideline compliance, citation/reference auditing, numerical-consistency checks, and response-to-reviewer workflows — combining agentic writing with deterministic integrity gates for submission-grade biomedical research. As of v5.0 it adds a model-engineering lane: choose a paper-grounded architecture, scaffold a reproducible, leakage-safe PyTorch training repo, and validate, document, and evaluate a medical-imaging or LLM/MLLM model so the work reaches a paper — it ships a minimal runnable default model for a forward-pass smoke test and integrates MONAI / nnU-Net / timm / torchvision for production-grade models, rather than reimplementing the ecosystem. Clinical AI model research engineering is in scope; it is not a diagnostic tool, an autonomous author, or a general AI-scientist platform, and every output requires human-expert verification. New here? See the 3 workflows below, the FAQ, the research connectors it calls (keyless public APIs — nothing to set up in the common case), and the scope boundary.

Quick Start

No terminal? Use the classroom installer ZIP — download, unzip, double-click the installer, then restart your agent app (see Installation).

Have a terminal? Fastest path — one command, nothing to clone:

npx medsci-skills install        # copies every skill into your agent's folder

Recommended (especially for clinicians): add --enable-update-notify so Claude Code shows a one-line "update available" notice when a new version ships — otherwise you stay on the version you installed and are never told. (No terminal at all? The classroom installer below turns this on for you.)

npx medsci-skills install --enable-update-notify        # install + in-app update reminders

Have git? Install every skill in three commands:

git clone https://github.com/Aperivue/medsci-skills.git
mkdir -p ~/.claude/skills
cp -r medsci-skills/skills/* ~/.claude/skills/

Restart Claude Code, then start with /orchestrate — it classifies your request and routes you to the right skill. Full install options (Codex, Cursor, individual skills) are in Installation.

Install as a Claude Code plugin

Prefer plugins? One line adds the marketplace; /plugin then lets you browse nine category plugins and enable the ones you want:

/plugin marketplace add Aperivue/medsci-skills
/plugin            # browse nine category plugins; enable the ones you want

| Plugin | Covers | |--------|--------| | medsci-literature | Literature search, full-text retrieval, Zotero sync, reference-integrity audits | | medsci-data | Study design, variable operationalization, sample size, data cleaning, de-identification, codebooks, dataset versioning | | medsci-modeling | Architecture selection, reproducible model-scaffold repos, model-validation audits, Model Card/Datasheet, model & LLM/MLLM evaluation | | medsci-analysis | Statistics, figures, batch/cross-national/replication analysis, meta-analysis | | medsci-writing | IMRAD & protocol drafting, AI-pattern removal, AI-search optimization, reviewer responses | | medsci-review | Self-review, peer review, reporting-guideline compliance | | medsci-submission | Submission packaging, journal selection, ICMJE/IRB form filling, grant proposals | | medsci-project | Orchestration, project intake/management, gap & topic discovery, author strategy | | medsci-presentation | Presentations/PPTX, PDF/document rendering, environment setup, skill publishing |

Install a single category and invoke its skills under that namespace:

/plugin install medsci-analysis@medsci-skills
/medsci-analysis:analyze-stats

All eight plugins share the same repository source, so this groups and enables skills by category — it is not a partial download. The marketplace tracks main, so a plugin's version is its git commit.

Want just one capability? Two skills are also published as focused standalone repos (generated mirrors; this repo stays the source of truth), each installable on its own with /plugin marketplace add Aperivue/<repo>:

• Aperivue/verify-refs — catch fabricated/mismatched citations (PubMed + CrossRef).
• Aperivue/check-reporting — audit a manuscript against the bundled EQUATOR reporting guidelines and risk-of-bias tools.

Start here: 3 workflows

New users don't need all the skills at once. Most work starts as one of three workflows. Each runs through /orchestrate or by invoking the named skills in order; all outputs require human-expert review.

Workflow A — Manuscript pre-submission audit. Use when a manuscript is nearly ready and you want it checked before a reviewer sees it. Skills: /self-review → /check-reporting → /verify-refs → /sync-submission. In: your manuscript (+ refs.bib, tables/figures). Out: anticipated reviewer comments, an item-by-item reporting-guideline audit, a citation-integrity report, and a submission-package drift check. Safety: it flags issues; you fix and verify them.

Workflow B — Data to manuscript package. Use when you have a cleaned dataset and need a full draft. Skills: /clean-data → /analyze-stats → /make-figures → /write-paper → /check-reporting → /find-journal. In: a cleaned CSV/parquet

• a research question. Out: reproducible analysis code, publication-ready figures, an IMRaD draft, a reporting checklist, and a journal shortlist. Safety: statistics and claims must be verified against your data; the toolkit never fabricates numbers or references.

Workflow C — Systematic review / meta-analysis. Use when you are running an SR/MA. Skills: /meta-analysis (with /search-lit, /make-figures, /check-reporting). In: a research question + search strategy. Out: PROSPERO-style protocol scaffolding, screening/extraction structure, PRISMA-consistent counts and diagram, pooled-estimate figures, and a manuscript draft. Safety: screening and extraction decisions stay with the human review team.

Live Demos: Four Study Types, Four Full Pipelines

Four public datasets. Four study types. Demos 1–3 each produce a complete manuscript, publication-ready figures, and a reporting-compliance audit; Demo 4 runs the medical-AI model-engineering lane end to end (scaffold → gates → training → evaluation → interpretability).

| Demo | Dataset | Study Type | Compliance | |------|---------|------------|------------| | Demo 1: Wisconsin BC | sklearn built-in | Diagnostic accuracy | STARD 2015 | | Demo 2: BCG Vaccine | metafor::dat.bcg (13 RCTs) | Meta-analysis | PRISMA 2020 | | Demo 3: NHANES Obesity | CDC NHANES 2017-18 | Epidemiology (survey) | STROBE | | Demo 4: PneumoniaMNIST CNN | medmnist (CC BY 4.0) | Medical-AI model engineering (CNN) | CLAIM / TRIPOD+AI |

Demo 1: Diagnostic Accuracy — Wisconsin Breast Cancer

from sklearn.datasets import load_breast_cancer
data = load_breast_cancer()  # 569 samples, zero download

Output from orchestrate --e2e (see full demo):

| Output | Description | |--------|-------------| | Manuscript | IMRAD draft, ~1,800 words | | Title Page | STARD title page with key points | | DOCX | Submission-ready Word document | | ROC Curve | 3-model comparison with DeLong 95% CIs | | Confusion Matrices | Per-model confusion matrices at threshold 0.5 | | STARD Flow | D2-generated STARD 2015 flow diagram | | Reporting Checklist | STARD 2015 — 60.9% compliance (14/23 applicable) | | Self-Review | Initial 82 (REVISE) → 88 (PASS) after 1 fix iteration; final 0 major / 1 minor | | Pipeline Log | 7-step E2E execution trace |

Pipeline: analyze-stats → make-figures → write-paper → AI pattern scan → check-reporting (STARD) → self-review → DOCX build → present-paper

Demo 2: Meta-Analysis — BCG Vaccine Efficacy

library(metafor)
data(dat.bcg)  # 13 RCTs, 357,347 participants (Colditz et al. 1994)

Output from orchestrate --e2e (see full demo):

| Output | Description | |--------|-------------| | Manuscript | Pooled RR = 0.489 (95% CI: 0.344–0.696), ~2,200 words | | Title Page | PRISMA title page with key points | | DOCX | Submission-ready Word document | | Forest Plot | 13 studies, RE model (REML), 300 dpi | | Funnel Plot | Small-study / publication-bias visual | | PRISMA Flow | D2-generated PRISMA 2020 flow diagram | | Reporting Checklist | PRISMA 2020 — 57.1% (24/42) at check-reporting → 61.9% (26/42) after self-review fix | | Self-Review | Initial 78 → 82 (REVISE) after 1 fix iteration; 3 major / 4 minor (majors are out-of-scope RoB/GRADE/references) | | Pipeline Log | 7-step E2E execution trace |

Pipeline: analyze-stats (R metafor) → make-figures → write-paper → AI pattern scan → check-reporting (PRISMA 2020) → self-review → DOCX build → present-paper

Demo 3: Epidemiology — NHANES Obesity & Diabetes

# Pre-processed NHANES 2017-2018 CSV included
# 5,010 US adults after exclusions

Output from orchestrate --e2e (see full demo):

| Output | Description | |--------|-------------| | Manuscript | Adjusted OR = 3.03 (95% CI: 2.29–4.02), ~1,850 words | | Title Page | STROBE title page with key points | | DOCX | Submission-ready Word document | | OR Forest Plot | Adjusted odds ratios for 7 variables | | Study Flow | D2-generated participant flow diagram | | Reporting Checklist | STROBE — 83.3% compliance (25/30 applicable) | | Self-Review | ACCEPT-WITH-NOTES after 1 fix iteration; 0 genuine majors remaining | | Pipeline Log | 7-step E2E execution trace |

Pipeline: analyze-stats → make-figures → write-paper → AI pattern scan → check-reporting (STROBE) → self-review → DOCX build → present-paper

Project Folder Structure

Each demo (and real project) follows this role-based folder layout:

project/
├── data/                          # Input data
│   └── raw_data.csv
├── analysis/                      # /analyze-stats + /make-figures outputs
│   ├── tables/
│   ├── figures/
│   │   └── _figure_manifest.md
│   ├── _analysis_outputs.md
│   └── analyze.py
├── manuscript/                    # /write-paper outputs
│   ├── manuscript.md
│   ├── manuscript_final.docx
│   └── title_page.md
├── qc/                            # Quality verification
│   ├── reporting_checklist.md     # /check-reporting
│   ├── self_review.md             # /self-review
│   └── _pipeline_log.md
├── submission/                    # Post-journal-selection (manual trigger)
│   └── {journal_short}/
│       ├── cover_letter.md
│       ├── checklist.md
│       └── peer_review.md
└── presentation/
    └── presentation.pptx

The E2E pipeline (orchestrate --e2e) produces everything up to qc/. The submission/ directory is created after journal selection via /find-journal.

What's New

v5.21 — verification-layer batch, mostly promoted from real submission failures. A marked (tracked-changes) manuscript is now built by driving Word's Compare from the command line and proved by a round trip — accepting every revision must reproduce the revised paper exactly, rejecting every revision must reproduce the original — and the gate is move-aware, because Word encodes a relocated paragraph as w:moveFrom/w:moveTo and an insert-and-delete-only verifier calls a good file corrupt. Every detector's qc/*.json now names the detector that wrote it, so an artifact can be traced back to the check that produced it (a CI-enforced contract). Two /verify-refs precision defects: a Unicode hyphen in a surname fired MISMATCH — the verdict that means fabricated author — on a correct reference, and a Better BibTeX brace-protected surname was read as a corporate author, silently skipping the author check the tool exists to perform. New gates: publisher markup in a .bib title (<scp>WHO</scp> renders as garbage and no gate looked at the printed title), complete/quasi separation before a logistic model is fitted (a pathognomonic sign has an empty cell by construction; glm returns OR = 0.00, p = 0.99, and an AUC that gets reported), and a probe + gate for manuscripts claiming a system improved itself (which rung of the verification hierarchy said so?). /find-cohort-gap now accepts a local codebook — an institutional registry or EMR export, not only a named public cohort — enumerating variables verbatim with provenance rather than letting a model summarise them. Additive and backward-compatible; 55 skills / 46 guidelines / 61 integrity detectors / 23 domain-probe modules. (See the CHANGELOG for v5.0–v5.20.)

v5.20.1 — audit-driven coherence fixes. A real /orchestrate --e2e state-transition bug (the pipeline halted at step 3 because it required a DOCX that is only rendered at step 7), all 55 skills made routable from the single entry point with a CI reachability gate, and a README plugin-count that had drifted from the marketplace SSOT (now gated). No skill/detector change.

v5.20 — reviewer-arithmetic gates. Five deterministic self-review detectors promoted from a peer-review cycle, each recomputing what a manuscript already prints: check_table_percentages (an n (%) cell vs its column denominator), check_nested_group_comparison (a P value comparing an analysed subset against the parent cohort that contains it — nested, invalid), check_reported_p_from_counts (rebuilds each 2×2 row and recomputes Fisher / Pearson χ² ± Yates in pure stdlib, calibrating the family on rows that reproduce), check_dta_denominators (sensitivity/specificity denominators vs the reference-standard category counts, behind a matching grand total), and check_paired_difference_estimator (an odd-n integer-scale median cannot be non-integer; a zero-width CI; an unnamed estimator). Plus /peer-review request-type discipline (disclosure vs computation) and impossibility-claim verification. Additive and backward-compatible; 55 skills / 46 guidelines / 57 integrity detectors / 22 domain-probe modules. (See the CHANGELOG for v5.0–v5.19.)

v5.19 — reviewer-safety + reporting-checklist batch. A PDF hidden-text / prompt-injection guard for /peer-review — a PyMuPDF extractor plus a stdlib detector that catch a review-steering instruction hidden in a submitted PDF (white-on-white text, sub-visible fonts, off-page glyphs, invisible render mode, or a document-metadata field) before an LLM ingesting the text layer can be steered by it, and emit visible-only text to feed a model instead of the raw PDF; plus the TARGET (target-trial emulation; Cashin et al., JAMA 2025) and REMARK (prognostic tumour-marker; McShane et al.) reporting checklists. Additive and backward-compatible; 55 skills / 46 guidelines / 52 integrity detectors / 22 domain-probe modules. (See the CHANGELOG for v5.0–v5.18.)

v5.18 — reliability & workflow-integrity batch. A new deterministic response-claim gate for /revise and /peer-review (verifies a response letter's "we added / we now cite X" against the revised manuscript body — a claimed-but-absent edit is caught before it reaches a reviewer; detectors 50 → 51), a reframe / headline-change survivor scan (--retired-term / --old-value, finds stale framing or superseded values left in the body, supplement, and figure legends after a reframe), a pre-drafting backbone full-text readiness gate for /write-paper (the backbone article is used, not merely detected), a skill-registry consistency validator (capabilities.yml ⇄ skill.yml, CI-enforced), AI-tool citation-framing guidance for /academic-aio, and Demo 4 (PneumoniaMNIST CNN, the model-engineering lane end to end). Additive and backward-compatible; 55 skills / 46 guidelines / 51 integrity detectors / 22 domain-probe modules. (See the CHANGELOG for v5.0–v5.17.)

v5.17 — model-engineering produce-side depth, completion. Deployment safety + the final roadmap/candidate items: a new uncertainty-imaging skill + check_uncertainty_reporting gate (calibrated per-case uncertainty / OOD guard on a held-out set / abstention at a pre-specified target / calibration-under-shift — for a deployment-framed model), an MLOps wiring reference for model-scaffold (experiment tracking, config/data/environment versioning, CI-for-ML — pointing to W&B / MLflow / nnU-Net, reimplementing nothing), and an architecture-zoo graph-neural-net family card (GCN / GraphSAGE / GAT / GIN / BrainGNN for brain connectomes) that closes the last candidate gap. The six-item produce-side depth roadmap and its candidate list are now complete. Additive and backward-compatible; 55 skills / 46 guidelines / 50 integrity detectors / 22 domain-probe modules. (See the CHANGELOG for v5.0–v5.16.)

v5.16 — model-engineering produce-side depth, clinical fine-tuning focus (Items 3–4 of the produce-side depth roadmap). A new radiomics-ml skill + check_radiomics_ml gate for the most common solo-doable clinical-ML workflow — radiomics / tabular features → any classical learner (LASSO / SVM / random forest / XGBoost / …) → a clinical outcome, no GPU — with a learner-agnostic nested-CV / calibration / stability gate; plus a model-scaffold fine-tuning mode (--task finetune + --from-pretrained) that adapts a pretrained backbone on collected clinical data with a frozen→unfrozen schedule, discriminative learning rates, and a pretrained-weight provenance record (PRETRAINED.md + a PRETRAINED_PROVENANCE_MISSING verdict on the existing check_training_hygiene, plus a MedSAM-adaptation + train-only diffusion-augmentation guide). Additive and backward-compatible; 54 skills / 46 guidelines / 49 integrity detectors / 22 domain-probe modules. (See the CHANGELOG for v5.0–v5.15.)

v5.15 — model-engineering produce-side depth. Two new skills that produce what the review lane previously only audited: preprocess-imaging (DICOM/NIfTI data prep + a check_preprocessing_leakage gate that extends the split-leakage moat upstream to the data stage) and explainability (Grad-CAM / saliency held to the reviewer bar — Adebayo sanity checks, quantitative localisation vs ground truth, attribution-not-validation framing, via check_explainability_report). Plus a by-stage skill index, multi-host README/About (Claude Code · Codex · Cursor · Copilot), copy-paste citation ergonomics, and a real-project precision fix. Additive and backward-compatible; 53 skills / 46 guidelines / 48 integrity detectors / 22 domain-probe modules. (See the CHANGELOG for v5.0–v5.14.)

v4.10 — reviewer-coverage expansion reverse-engineered from high-IF, CC-BY papers (learn-only under the reverse_engineer/ license firewall), plus a clinician-friendly update path. Additive and backward-compatible; 45 skills / 46 guidelines / 36 detectors / 15 domain-probe modules (was 12):

• Three new reviewer domain-probe modules (/peer-review + /self-review, vendored byte-identical): Mendelian randomization (MR1–MR8 — IV assumptions, pleiotropy-robust sensitivity suite, Steiger, sample overlap, NLMR, drug-target colocalization), polygenic risk score (PG1–PG8 — ancestry portability, base/target leakage, incremental value over the clinical model, screening-vs-discrimination, calibration), and network meta-analysis (NM1–NM8 — transitivity, incoherence, SUCRA over-interpretation, CINeMA/GRADE-NMA, component-NMA additivity). Plus observational O17 (agnostic many-exposure-scan multiplicity: ExWAS/EWAS/MWAS).
• Two reporting-guideline checklists (36 → 38): STROBE-MR and PGS-RS / PRS-RS, with study-type routing. Four new /analyze-stats analysis guides (multiplicity, MR, PRS, NMA) and a /clean-data implausible-value + cross-field validity reference.
• Clinician-friendly update reminders — the classroom installers enable the in-app "update available" notice + one-click Desktop updater by default; the npx/manual paths print how to turn it on; the install guide recommends npx medsci-skills install --enable-update-notify.

v4.9 — analysis-integrity hardening promoted from real review cycles, plus journal-mechanics additions. Additive and backward-compatible; still 45 skills / 46 guidelines, analysis-integrity detectors 32 → 36:

• Four new gates — a duplicate-bibliography check (check_reference_duplication.py) for the hybrid [@key] + hand-typed ## References build that renders the list twice; a cross-script binning / composite-indicator consistency check (check_binning_consistency.py, BINNING_DRIFT / DERIVED_DEF_DRIFT) for a derived categorical or composite indicator defined inconsistently across analysis scripts; a float citation-order check (check_citation_order.py) for numbered Tables/Figures not first cited in ascending order per series; and an audit-dump leak gate (/sync-submission) that blocks a /check-reporting output mistakenly attached as a submission file.
• KJR technical-check conventions + percentage-decimal style, reader-allocation-under-burden and generative-image-as-study-object reporting (/design-ai-benchmarking, /check-reporting), and a Liver International CSL with that journal's submission mechanics (/manage-refs).

v4.8 is the review-harvest batch — deterministic detector hardening promoted from real-manuscript review cycles. Additive and backward-compatible; still 45 skills / 46 guidelines, analysis-integrity detectors 30 → 32:

• Two new gates — check_supplement_hygiene.py lints the rendered supplement / tables / caption files (not just the manuscript) for §-labels, placeholders, build markers, response-letter framing, and unresolved body↔supplement cross-references; check_null_calibration.py flags a headline negative/equivalence claim made without a minimum-detectable-effect / power / equivalence statement.
• Four detector false-positive fixes — gates no longer fire on a recommended colorblind-safe palette, author-footnote § daggers, a correctly-hedged disclaimer, or a tier-label digit; each with a regression fixture and three newly CI-wired test suites.
• Nine reviewer-side domain probes (SR/MA, observational, diagnostic, AI-overclaiming, survival) plus a /design-study design-stage ceiling gate for perceptual/reader-AI studies and a reusable confidence-weighted-rating→AUC monotonicity template.

v4.7 is the self-update foundation — physician-researchers stay current without GitHub, git, or a terminal. Additive and backward-compatible; still 45 skills / 46 guidelines / 30 detectors:

• Transactional, crash-recoverable installer. Each install runs through a durable journal state machine recovered on the next run (roll back / forward-clean / fail-closed), with per-target SHA-256 inventories — your modified or third-party skills are backed up and never clobbered or auto-deleted.
• One-click self-updater (~/.medsci-skills/updater/, install.py --check-update). Verifies the download against the github.com API digest and never extractall()s (per-entry rejection of traversal / symlink / duplicate / zip-bomb + an allowlist & per-file hash). The release pipeline injects a verified provenance.json, attests build provenance, runs on a protected release environment, and verifies each ZIP round-trips through the updater's own safe-extract before publishing.
• Opt-in update notice (off by default): install.py --enable-update-notify shows a one-line "update available" message at Claude Code session start — no telemetry, reads nothing about your session, installs nothing. --disable-update-notify / MEDSCI_NO_UPDATE_CHECK=1 turn it off. (Honest scope: the digest/attestation detect transport tampering, not a compromised publisher account — see SECURITY.md.)

v4.6 is a maintainability, governance, and review-depth release — still 45 skills / 46 guidelines; analysis-integrity detectors 28 → 30, domain probes 11 → 12:

• Fairness / equity / subgroup-performance probe (EQ0–EQ6) for AI/prediction/diagnostic studies that claim cross-population performance, plus two new detectors: an AI-disclosure + data/code-availability check (/sync-submission) and a structured-summary-box conformance check (/academic-aio).
• Governance + answer-engine layer: ROADMAP.md, MAINTAINERS.md, SECURITY.md, a maintainer workflow + release checklist, an AEO/GEO docs/faq.md, a "Start here: 3 workflows" + "Validation status" section in this README, and a new maturity field (official / experimental / community) on every skill.
• Token diet (pilot): write-paper Phase 7 integrity audits moved to a load-on-demand reference (~2,559 tokens saved per invocation). Positioning now leads with the compliance moat rather than skill count.

v4.5 deepens the review + submission surface with no new skill or reporting-guideline count (still 45 skills / 46 guidelines); analysis-integrity detectors 27 → 28:

• /clean-data + /analyze-stats — reverse-coded-item / negative-alpha detector. A multi-item Likert scale with a negatively-worded item must be recoded (min+max) − x before the scale total or Cronbach's alpha is computed; left un-recoded, the item correlates negatively with the rest of the scale and alpha collapses (often negative). A negative alpha is a coding bug, not a "multidimensional construct." New stdlib-only check_reverse_coding.py returns REVERSE_CODING_LIKELY / REVERSE_CODING_SUSPECT / OK from per-item item-rest correlations + raw alpha; the Likert summary template gains a --reverse-items recode flag.
• /peer-review + /self-review — SR/MA + DTA + prediction-model probe batch. sr_ma.md P12 risk-of-bias table row-sum ↔ traffic-light figure-matrix reconciliation and P13 included-study ↔ reference-list completeness; diagnostic_accuracy.md D7 index-test-as-enrollment-criterion circularity; clinical_prediction_model.md CP5 intended-use horizon leakage and CP6 development/CV vs held-out/external validation-nomenclature conflation. Vendored byte-identical into /self-review.
• /sync-submission — embedded absolute-path leak scan. A word/*.xml attribute (e.g. a pandoc-embedded image's <pic:cNvPr descr="…">) carrying an absolute home-dir path (/Users/…, /home/…) is a username leak invisible to a rendered-text scan; now flagged as docx_embedded_abs_path under check_asset_anonymization.py.

v4.4 adds reviewer/analysis depth with no new skill or reporting-guideline count (still 45 skills / 46 guidelines / 27 detectors):

• /author-strategy — trajectory-archetype classification (optional). Classifies a queried author's PubMed trajectory into abstract career archetypes (A1 infrastructure builder, A2 methodology rule-maker, A3 clinical→AI hybrid, A4 SR/MA volume engine, A5 large-consortium participation, A6 device/technique depth, + a computed composite) as an explainable, multi-label, confidence-scored heuristic — not an objective verdict. The rubric is a single canonical YAML (the narrative doc is generated from it); scores exclude unavailable signals (h-index/citation/venue-tier → [VERIFY], never fabricated); a disambiguation gate binds an approved corpus_manifest.json to the CSV (csv + PMID-set hashes) so a surname alone never classifies, and target-author attribution never borrows a co-author's ORCID/affiliation.
• /peer-review + /self-review — Image-Synthesis / cross-modality probe (IS1–IS4) for studies that synthesize one imaging modality from another and claim the output carries the target's information, plus a reviewer-side reference-integrity spot-check.
• /verify-refs — OpenAlex tertiary index recovers conference-proceedings / non-DOI citations (NeurIPS/ICLR/ACL) that fall through PubMed and CrossRef, the free analogue of a portal's second index.

v4.3 hardens the cross-sectional / observational cohort review surface end-to-end, much of it reverse-engineered from real CC-BY cohort papers (learn-only under the license firewall) — no new skill or reporting-guideline count (still 45 skills / 46 guidelines); analysis-integrity detectors 25 → 27:

• Observational probes O1 → O14 (/peer-review + /self-review, vendored) — over-adjustment / analysis-unit clustering / outcome construct-validity (O7–O9), overlapping-subset gradient (O10), complex-survey design & weighting for NHANES/KNHANES (O11), data-driven threshold / "inflection-point" mining (O12), cross-sectional mediation temporal-order & sequential-ignorability (O13), and interaction scale — additive RERI/AP/S vs multiplicative (O14). Plus a new clinical-prediction-model probe module CP1–CP4 and survival S9 (panel-data / multistate variance).
• Two new detectors (25 → 27) — check_wordcount_cap.py (the revision-inflation trap: body vs journal cap) and check_paren_spans.py (em-dash→paren conversions that wrap a whole sentence). Plus a check_confounding_completeness.py upgrade (DB-code↔prose alias map, SMD-from-mean±SD, exposure-defining-covariate exemption), a check_cohort_arithmetic.py ANALYSIS_UNIT_UNDISCLOSED check, a check_scope_coherence.py cross-sectional-yield lexicon, and a verify-refs corporate/collective-author render-abort fix.
• Analysis & submission tooling — /analyze-stats gains mediation and interaction & effect-modification guides; /sync-submission gains assemble_supplement.py (S{N} index↔file integrity) and a /revise body-word-count exit gate; /render-pdf-doc gains a scan_glyph_coverage.py xelatex silent-glyph-drop scan.

v4.2 builds out the case-report capability end-to-end, grounded in real CC-BY case reports (learn-only under the license firewall) — no new skill or reporting-guideline count (still 45 skills / 46 guidelines); journal profiles 68 → 73:

• Case-report + case-series writing — /write-paper gains a CARE narrative + 150-word-abstract case-report exemplar, a case-series paper type (methods-light mini-cohort, all-cases summary table, counts-not-rates), and adverse-event/pharmacovigilance (Naranjo/WHO-UMC causality) and diagnostic-pitfall/mimic subtypes.
• Radiology / imaging-led track — a dedicated exemplar_case_report_radiology.md (per-modality technique→findings→impression, structured-reporting lexicons BI-RADS/LI-RADS/PI-RADS/TI-RADS/Lung-RADS/O-RADS, quantitative threshold honesty, an interventional-radiology procedure/complication subtype, DICOM de-identification) plus a /make-figures annotated multimodality imaging-panel exemplar.
• Case-report reviewer probe — /peer-review + /self-review ship a vendored case-report domain probe CR1–CR9 (novelty/consent/n=1 causality, case-series design, adverse-event causality, imaging-led discipline).
• Where to submit — compact /find-journal profiles for Journal of Medical Case Reports, Cureus, Radiology Case Reports, BMJ Case Reports, and BJR Case Reports, and /check-reporting CARE notes for adverse-event and case-series subtypes.

v4.1 ships distribution levers and a submission pre-flight gate — analysis-integrity detectors 24 → 25 (still 43 skills):

• Claude Code plugin marketplace — /plugin marketplace add Aperivue/medsci-skills, then /plugin discovery of nine medsci-* category plugins generated from the catalog SSOT (.claude-plugin/marketplace.json).
• MedSci-Audit detector registry — the deterministic verification layer is now a named, enumerated, citable suite (MEDSCI_AUDIT.md + generated metadata/detectors_catalog.json, six audit families).
• Hero-skill standalone mirrors — scripts/sync_hero_skill.py mirrors a focused skill to its own star-funnel repo; first two live: Aperivue/verify-refs and Aperivue/check-reporting.
• Placeholder/marker gate — check_placeholders.py flags leftover [@NEW:] / [version] / TODO / template-URL markers before submission (the 25th detector).
• Submission pre-flight gate — preflight_gate.py bundles the existing detectors + /verify-refs into one halt-on-failure command (qc/preflight_gate_report.json, non-zero exit on any blocker) — the single last step before freeze.

v4.0 extends the project's own deterministic, no-drift SSOT discipline to the public storefront and finishes the detector backlog — bringing the analysis-integrity detector count in skills/ to 24 (still 43 skills):

• SSOT to the storefront — a generated, machine-readable metadata/skills_catalog.json (slug + research-lifecycle category + one-line description per skill) is now the source the aperivue.com storefront vendors, gated offline so the public site can never silently drift behind the repo (gen_skills_catalog_json.py --check).
• Asset/figure anonymization — /sync-submission scans figure-generating scripts, figure-PDF rendered text, and docx/PDF metadata authors for the institution/author leaks a body-text scan misses (check_asset_anonymization.py).
• Cross-artifact staleness — flags supplement values that disagree with the corrected body, and reporting checklists built against an older manuscript version (check_cross_artifact_stale.py; check_checklist_version.py with a target_manuscript/source_sha256 checklist contract).
• Survival reporting — /analyze-stats emits median survival with its 95% CI, a Cox events-per-variable gate, and cluster-robust SE for nested observation units.

v3.8.0 adds an evaluation/ harness suite that validates the instrument itself — deterministic detector recall on programmatically seeded defects (E1), fresh-clone manifest reproducibility (E4), claim audit-trail completeness (E5), host-portability and metadata-drift checks (E6/E7/E8), and a cost/time table (E3) — each writing a self-describing, reproducible run package. An LLM-comparator (E2) and a self-review convergence harness (E9) ship runnable but are NOT executed in this release. This release also reconciles the README Live-Demos numbers with the v3.7.0 clean-room demo artifacts. Catalog unchanged (still 43 skills, 21 detectors).

v3.7.0 adds three deterministic, stdlib-only detectors on top of the v3.6.0 panel-derived gates — bringing the analysis-integrity detector count in skills/ to 21 — without broadening the catalog (still 43 skills):

• Reference adequacy — /self-review and /write-paper now check that a draft cites enough references in the right sections and that every named method (a competing-risk model, multiple imputation, the E-value, an eGFR equation) carries a citation — the adequacy layer that complements /verify-refs's integrity layer (check_reference_adequacy.py).
• Panel lens-diversity — /self-review --panel post-processes its reviewers so the cost buys breadth, not a louder echo (check_panel_diversity.py).
• Generated-code quality — /analyze-stats lints emitted analysis scripts for reproducibility slop (missing seed, hard-coded data literals, absolute paths, in-place source overwrite) (check_generated_code.py).

Plus a publish-time skill-worthiness gate (/publish-skill) and public adoption/impact tracking (IMPACT.md).

v3.6.0 lands 18 gates from a 13-project panel self-review (158 traces → 12 recurring defect patterns), without broadening the catalog (still 43 skills). Six new stdlib detectors join the existing trio — deterministic where a grep is clean, prose/probe where the call needs a human:

• Cohort & pool arithmetic — /self-review recomputes incidence rates from events ÷ person-years, balances STROBE exclusion cascades, and checks ordinal tier/stratum partitions for disjointness (check_cohort_arithmetic.py); /meta-analysis locks patient/lesion aggregate totals and requires re-run evidence for any "fixed" audit note.
• Claim ↔ artifact ↔ scope — Methods ↔ Results ↔ disk coverage (a run-but-unreported analysis is flagged), an endpoint ↔ conclusion scope gate (a cross-sectional design cannot license a surveillance claim; a binary surrogate is not a care directive), and a reviewer-team 3-way that makes an LLM-as-reviewer fatal.
• Statistical & reporting contracts — a CI/estimand output contract (quantile/proportion/sHR must carry a 95% CI; Cox EPV gate; proportion-MA Egger ban + prediction interval), interval-censoring / PH-violation / CIF-horizon survival rules, reporting-framework base+extension naming, a classical-style body lint, a PROSPERO ID format gate, and a pagination-placeholder citation gate.

Earlier in this series: analysis-integrity guards (confounding completeness, claim-vs-artifact, structural-zero handling), a multi-agent /self-review --panel mode, and shared domain-probe modules vendored byte-identical into /peer-review and /self-review with a CI drift gate.

Why This Repo?

| | MedSci Skills | Broad skill aggregators | |---|---|---| | Citation quality | Every reference passes reference-verification gates (PubMed / Semantic Scholar / CrossRef) and citation-audit workflows before inclusion. | No verification -- citations generated from model memory | | Pipeline integration | Skills call each other in defined chains. design-study -> calc-sample-size -> write-protocol. | Standalone stubs with no cross-skill interaction | | End-to-end coverage | From IRB protocol to journal submission: sample size, data cleaning, analysis, writing, compliance, journal selection, cover letter. | Gaps at every transition -- no protocol, no journal matching, no cover letter | | Battle-tested | Used on real manuscript submissions by a practicing physician-researcher | Unknown provenance and validation | | Depth per skill | 150-600 lines of documentation + bundled reference files (curated journal profile library, checklists, formula sheets, code templates) | Typically thin SKILL.md templates |

MedSci-Audit — the verification edge in the first rows above is a named suite of 36 deterministic detectors (citation & reference integrity, cohort & pool arithmetic, scope/estimand contracts, reporting compliance, and more) that catch fabricated or drifted content before a manuscript reaches a reviewer. See MEDSCI_AUDIT.md for the suite, its six families, and its evaluation evidence.

What This Is NOT

This is not a broad scientific-tooling library — for cheminformatics, structural biology, or genomics pipelines, see K-Dense scientific-agent-skills. It is not a biomedical-skill aggregator — for a broad curated collection, see OpenClaw Medical Skills. For how MedSci Skills compares to these catalogs, see docs/competitive_positioning.md. For verified cross-agent install paths (Claude Code, Codex, Cursor, GitHub Copilot), see docs/host_compatibility.md.

MedSci Skills is opinionated and narrow on purpose: a single physician-researcher's medical-manuscript pipeline, biased toward radiology, diagnostic accuracy, observational EMR studies, and systematic review / meta-analysis. If you write IMRAD manuscripts for clinical journals, audit reporting compliance against EQUATOR guidelines, or run SR/MA workflows end-to-end, this is built for you. For wet-lab protocols, drug discovery, or single-cell genomics, the repos above are better fits.

Skills

📖 Per-skill reference: docs/skills/ — one page per skill (what it does, when it activates, its Quality Card — purpose, safety boundaries, known limitations, validation, evidence — and bundled resources), generated from each SKILL.md + skill.yml. See docs/skills/AUDIT.md for how the skills are validated.

🧠 ML / DL method coverage: docs/method_coverage_map.md — which machine-learning and deep-learning method families (CNN / transformer / segmentation / detection / foundation / diffusion / SSL for imaging; the full classical family — penalised regression, SVM, k-NN, trees, boosting [XGBoost / LightGBM / CatBoost], MLP, ensembles, clustering — for radiomics/tabular; LLM/MLLM) map to which skills for selection, production, validation, interpretation, and reporting.

                              ┌─────────────────────────────────┐
                              │  orchestrate: single entry point │
                              │  classifies intent, routes to    │
                              │  the right skill or chains them  │
                              └───────────────┬─────────────────┘
                                              │
                  ┌───────────────────────────┼───────────────────────────┐
                  │                           │                           │
            intake-project              (main pipeline)             grant-builder
            (new/messy projects)              │                    (proposals)
                  │                           │
                  ▼                           ▼
                                    ┌── calc-sample-size ──┐
                                    │                      ▼
ma-scout -> search-lit -> fulltext-retrieval -> design-study ──> write-protocol -> manage-project
   │            │
   │            └── find-cohort-gap (DB variables → literature gap → ranked topic proposals)
   │                                    │
   │                                    ▼
   │                         deidentify -> clean-data -> analyze-stats -> make-figures -> write-paper
   │                                                        │                                │
   │                                           replicate-study (paper → new DB)         humanize
   │                                           cross-national (parallel survey)              │
   │                                           batch-cohort (N × M matrix)                   ▼
   │                                                                          find-journal <── self-review
   │                                                                               │                    │
   │                                                                               │                    ▼
   │                                                                               │          humanize -> academic-aio (AI-search visibility)
   │                                                                               ▼
   │                                                    [cover-letter] -> check-reporting -> revise -> present-paper
   │                                                                                                       │
   └── meta-analysis                                                                                  peer-review
                         lit-sync (Zotero + Obsidian sync)     author-strategy (PubMed profile analysis)

                              ┌─────────────────────────────────────────────┐
                              │  publish-skill: package any skill above for │
                              │  open-source distribution (PII audit,       │
                              │  license check, generalization)             │
                              └─────────────────────────────────────────────┘
                              ┌─────────────────────────────────────────────┐
                              │  add-journal: add new journal profiles to   │
                              │  the database (write-paper + find-journal   │
                              │  dual profile generation with quality gates)│
                              └─────────────────────────────────────────────┘

By research stage

All 55 skills, grouped by where they fit in the clinical-manuscript and medical-AI lifecycle. Full descriptions are in the table below; one page per skill lives in the per-skill reference.

| Stage | Skills | |-------|--------| | 🔭 Discover & scope | ma-scout · find-cohort-gap · search-lit · fulltext-retrieval · lit-sync · author-strategy | | 📐 Design & plan | design-study · calc-sample-size · define-variables · write-protocol · fill-protocol · design-ai-benchmarking | | 🧹 Data & analysis | deidentify · clean-data · generate-codebook · version-dataset · analyze-stats · batch-cohort · cross-national · replicate-study | | 🤖 Medical-AI model engineering | preprocess-imaging · architecture-zoo · model-scaffold · model-validation · model-evaluation · uncertainty-imaging · explainability · radiomics-ml · model-card · mllm-eval | | ✍️ Write & visualize | write-paper · make-figures · review-paper · present-paper · humanize · polish-language · academic-aio | | ✅ Comply & verify | check-reporting · self-review · verify-refs · manage-refs | | 📤 Submit & respond | find-journal · add-journal · sync-submission · revise · peer-review · fill-icmje-coi | | 🧭 Orchestrate & manage | orchestrate · intake-project · manage-project · meta-analysis · grant-builder · publish-skill · render-pdf-doc · setup-medsci |

Available Now

| Skill | What It Does | |-------|-------------| | orchestrate | Single entry point for the full bundle. Classifies your request and routes to the right skill -- or chains multiple skills for multi-step workflows. Full Pipeline Mode runs analyze-stats → make-figures → write-paper → check-reporting → self-review end-to-end. --e2e flag for fully autonomous execution with post-skill validation and halt-on-failure. | | find-cohort-gap | Research gap finder for longitudinal cohort databases. Profiles cohort strengths, matches PI expertise, scans literature saturation via 6-Pattern scoring, and outputs ranked topic proposals with comparison tables and one-pagers. Works with any cohort: NHIS, UK Biobank, institutional EMR, health checkup registries. | | search-lit | PubMed + Semantic Scholar + bioRxiv search with anti-hallucination citation verification. Token-efficient error handling -- CrossRef failures are silently batched, not repeated. BibTeX output tags each entry with verified/verified_by/verified_on fields so downstream skills can trust the citation provenance. | | verify-refs | Pre-submission reference audit for .md, .docx, .bib, or .tsv inputs. Extracts references, verifies DOI/PMID via CrossRef/PubMed when available, and writes qc/reference_audit.json as the sole output — row-level status (OK / MISMATCH / UNVERIFIED / FABRICATED) lives inside the JSON records[] block. /search-lit produces candidate BibTeX; /lit-sync owns manuscript/_src/refs.bib. | | fulltext-retrieval | Batch open-access PDF downloader. Unpaywall → PMC → OpenAlex → CrossRef pipeline. OA-only -- no paywall bypass. Input: DOI list or TSV. Optional PDF→Markdown conversion via pymupdf4llm for token-efficient LLM analysis of academic papers. | | check-reporting | Manuscript compliance audit against 46 reporting guidelines and risk of bias tools (STROBE, STROBE-MR, RECORD, STARD, STARD-AI, TRIPOD, TRIPOD+AI, TRIPOD-LLM, PGS-RS, CHEERS 2022, CROSS, SRQR, COREQ, PRISMA, PRISMA-DTA, PRISMA-P, PRISMA-ScR, MOOSE, ARRIVE, CONSORT, CONSORT-AI, CARE, SPIRIT, SPIRIT-AI, CLAIM, DECIDE-AI, SQUIRE 2.0, CLEAR, GRRAS, MI-CLEAR-LLM, SWiM, AMSTAR 2, QUADAS-2, QUADAS-C, RoB 2, ROBINS-I, ROBINS-E, ROBIS, ROB-ME, PROBAST, PROBAST+AI, NOS, COSMIN, RoB NMA). Machine-readable JSON summary with compliance_pct and fixable_by_ai flags for automated pipeline integration. | | analyze-stats | Statistical analysis code generation (Python/R) for diagnostic accuracy, DTA meta-analysis (bivariate/HSROC), inter-rater agreement, survival analysis, demographics tables, regression (logistic/linear), propensity score (matching/IPTW/overlap weighting), and repeated measures (RM ANOVA/GEE/mixed models). Calibration mandatory for prediction models. | | meta-analysis | Full systematic review and meta-analysis pipeline (8 phases). DTA (bivariate/HSROC) and intervention meta-analysis. Protocol to submission-ready manuscript with PRISMA-DTA compliance. | | make-figures | Publication-ready figures and visual abstracts: ROC curves, forest plots, PRISMA/CONSORT/STARD flow diagrams, Kaplan-Meier curves, Bland-Altman plots, confusion matrices, and journal-specific visual/graphical abstracts (python-pptx template-based). Communication-first design principles (Nat Hum Behav 2026 — key message, audience, cognitive load, figure-vs-table decision) and five flow-diagram production lessons (official-template fidelity, VML fallback PDF export, docx XML escape, sequential placeholder mapping, version freeze); critic rubric Section G adds 5 communication-first checks. --study-type auto-generates the full required figure set; structured _figure_manifest.md output for downstream pipeline consumption; D2 enforced as default for flow diagrams. | | design-study | Study design review: identifies analysis unit, cohort logic, data leakage risks, comparator design, validation strategy, and reporting guideline fit. | | design-ai-benchmarking | Design and validity review for benchmarking AI system(s) against a human-expert panel: evaluation-question and arm definition, decoupled multi-dimensional rubrics with anchors, planted calibration probes (positive-control / known-bad / instability / mechanism-contradiction), reviewer-panel construction with per-reviewer randomization, inter-rater reliability targets with separate control-item reliability, LLM-as-judge vs human-as-judge adjudication, construct-independence guards, and a structured JSON rating-export schema. Locks the rubric before data collection. | | model-validation | Design or audit the clinical-validation study for an engineer-built medical-imaging model (segmentation / classification / detection): patient-level split disjointness and the data-leakage taxonomy, tuning-on-test, internal vs genuine external validation, comparator design, single-run vs multi-seed variance, task-correct metric selection (Metrics Reloaded), test-set sizing, and CLAIM 2024 / TRIPOD+AI / STARD-AI reporting fit. Ships a deterministic split-leakage gate that proves patient disjointness by set arithmetic on the emitted split table. Integrates with MONAI / nnU-Net — does not replace them. | | preprocess-imaging | Design or audit the data-preparation stage of a medical-imaging model — DICOM/NIfTI intake, resampling, intensity normalization, and the augmentation plan — so the pipeline is leakage-safe before model-scaffold builds the training repo. Emits a declarative preprocessing manifest and a deterministic data-stage leakage gate (check_preprocessing_leakage) that catches what the split table cannot see: a dataset-level normaliser fit on non-train data (NORMALIZATION_LEAKAGE), a data-fitted transform run before the split (PREPROCESS_BEFORE_SPLIT), and a patient's slices crossing splits (PATIENT_CROSS_SPLIT). Integrates MONAI / TorchIO transforms; never reimplements them or touches real patient data. | | model-scaffold | Generate a reproducible, runnable PyTorch training repo for a medical-imaging task — segmentation (U-Net), classification, detection, image-to-image synthesis, self-supervised pretraining, or fine-tuning a pretrained backbone (transfer learning) — the missing middle link between choosing an architecture and validating a trained model. Emits a patient-level seed-locked split as an auditable artifact, a task-appropriate model, train/evaluate scripts that seed every RNG and infer under eval mode, a config, requirements, a reproducibility record, and a Methods stub with VERIFY placeholders (no fabricated numbers). Fine-tuning mode (--task finetune) adds a frozen→unfrozen schedule, discriminative learning rates, and a pretrained-weight provenance record (PRETRAINED.md), with a MedSAM-adaptation + train-only diffusion-augmentation guide. Reproducibility holds by construction; ships a check_training_hygiene AST gate (RNG seeding, eval-mode inference, train-split-only loaders, pretrained-provenance) + a network-free build→validate challenge. Integrates with MONAI / nnU-Net / TorchIO / timm / torchvision for production-grade models. | | architecture-zoo | "Which architecture for which research question" decision tool: maps task (classification / segmentation / detection / transfer), modality, data scale, and class imbalance to a paper-grounded architecture shortlist. Curates the foundational curriculum (ResNet / DenseNet / EfficientNet / ViT / Swin; U-Net / 3-D U-Net / Attention & Residual U-Net / nnU-Net / Mask R-CNN; SAM/MedSAM / TotalSegmentator / BiomedCLIP / DINO / MAE / SimCLR) — each with core idea, when-to-use, medical-imaging use, reference implementation, validation setup, and the matching model-scaffold template. Advisory; teaches archetypes, not a live SOTA leaderboard. | | model-card | Generate the documentation an engineer-built medical-imaging model must carry — a Model Card (Mitchell et al. 2019), a Datasheet for its dataset (Gebru et al. 2021), and a METRIC-informed data-quality pass — filled from user-supplied facts (never fabricated), then verify every required section is present and non-empty with a deterministic completeness gate (check_model_card_complete). Model Card / Datasheet are documentation standards vendored as templates, not counted reporting checklists. | | model-evaluation | Compute and report task-correct held-out metrics for a trained medical-imaging model — segmentation (Dice + a boundary metric HD95/NSD, per structure), classification (AUROC + AUPRC + sensitivity/specificity with bootstrap CIs at the deployment prevalence), or detection (FROC/mAP with a stated IoU criterion) — plus calibration and subgroup slices. Emits a per-case table for analyze-stats and gates the metric choice against Metrics Reloaded / CLAIM 2024 (check_metric_reporting). Numbers come only from executed code. | | explainability | Produce or audit the interpretability/explainability analysis of a medical-imaging model — Grad-CAM / Grad-CAM++ / attention-rollout / saliency / integrated-gradients — so it clears the rigor bar: mandatory Adebayo sanity checks (model- and data-randomisation), a quantitative localisation metric against ground truth (IoU / pointing game / Dice) instead of eyeballed examples, a cohort-level result, and attribution framing rather than "proof the model is correct". Emits an explainability-report manifest + a deterministic gate (check_explainability_report): SALIENCY_AS_VALIDATION, NO_SANITY_CHECK, NO_LOCALIZATION_METRIC (Major); INSUFFICIENT_SANITY, CHERRY_PICKED_EXAMPLES, MISSING_METHOD (Minor). Integrates captum / pytorch-grad-cam; never reimplements them or touches real patient data. | | radiomics-ml | Produce or audit a radiomics / tabular clinical-ML study — features → random forest / XGBoost / regularised logistic → clinical outcome — the most common solo-doable clinical-ML workflow (no GPU, no engineer). Emits a pipeline manifest + a deterministic rigor gate (check_radiomics_ml): NO_NESTED_CV, HIGH_DIM_LOW_EVENTS, SELECTION_OUTSIDE_CV (Major); NO_FEATURE_STABILITY, NO_CALIBRATION, NO_EXTERNAL_VALIDATION (Minor). pyradiomics/IBSI settings, nested CV with in-fold selection, ICC stability, SHAP, calibration + decision curve, CLEAR/TRIPOD+AI/PROBAST-AI reporting. Integrates scikit-learn / xgboost / pyradiomics; never reimplements them. | | uncertainty-imaging | Design or audit the uncertainty / out-of-distribution / selective-prediction layer of a deployment-framed medical-imaging model — so a clinical-use claim carries calibrated per-case uncertainty (MC-dropout / deep ensemble / conformal / Bayesian), an OOD guard validated on a held-out OOD set, an abstention rule at a pre-specified operating point, and calibration checked under distribution shift. Emits an uncertainty manifest + a deterministic gate (check_uncertainty_reporting): POINT_PREDICTION_NO_UNCERTAINTY, CONFORMAL_NO_COVERAGE_VALIDATION, OOD_NO_HELDOUT_SET (Major); ENSEMBLE_NOT_INDEPENDENT, MCDROPOUT_DISABLED_AT_INFERENCE, SELECTIVE_NO_TARGET, NO_CALIBRATION_UNDER_SHIFT (Minor). Integrates MAPIE / captum / OOD scorers; never reimplements them or touches real patient data. | | mllm-eval | Model-agnostic evaluation harness (closed API or open weights) for an LLM/MLLM on a clinical task — radiology report generation, VQA, clinical text extraction — covering the adjudicated reference standard, clinical-efficacy metrics (RadGraph-F1 / CheXbert-F1 beyond BLEU/ROUGE), faithfulness/hallucination, pretraining-contamination, prompt sensitivity, and a reader study; gates the plan with check_mllm_eval_completeness and routes the reviewer audit to the MLLM probe. | | intake-project | Classifies new research projects, summarizes current state, identifies missing inputs, and recommends next steps. | | grant-builder | Structures grant proposals: significance, innovation, approach, milestones, and consortium roles. | | present-paper | Academic presentation preparation: paper analysis, supporting research, speaker scripts, slide note injection, and Q&A prep. | | publish-skill | Convert personal Claude Code skills into distributable, open-source-ready packages. PII audit, license compatibility check, generalization, and packaging workflow. | | write-paper | Full IMRAD manuscript pipeline (8 phases). Outline to submission-ready manuscript with critic-fixer loops, AI pattern avoidance, and journal compliance. Anti-interpretation guardrails in Results; interactive Discussion planning with anchor paper input. Case report mode (CARE 2016, 1000-word short-form). Optional cover letter generation (Phase 8+). LLM Disclosure: auto-generates disclosure statements in Methods, Acknowledgments, and Cover Letter (opt-out via --no-llm-disclosure). --autonomous flag skips all user gates for fully automated manuscript generation; Phase 2 auto-calls /make-figures --study-type with manifest verification; Phase 7 enforces strict sequential QC chain (check-reporting → search-lit → self-review fix loop → DOCX build). | | review-paper | Scaffold and draft a literature review — narrative (SANRA), scoping (PRISMA-ScR + JBI), or systematic (PRISMA 2020). Asks for the spine axis (modality / task / lifecycle), builds a 7-part skeleton with a required Intro scope/n