A plain-language reference for researchers evaluating MedSci Skills, and a search/answer-engine-friendly summary of what it is and is not.
MedSci Skills is an open-source Claude Code skill collection for clinical manuscript preparation. It helps physician-researchers and biomedical 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. It combines agentic writing workflows with deterministic integrity gates for submission-grade biomedical research.
Physician-researchers, biomedical investigators, and research teams writing clinical, diagnostic-accuracy, observational, or systematic-review/meta-analysis manuscripts — especially in radiology and medical AI, where the bundled reporting guidelines and journal profiles are deepest.
No. It is workflow tooling with deterministic checks, not an autonomous author. It does not replace authors, statisticians, reviewers, IRBs, or journal requirements, and every output requires human-expert verification. Its goal is to make manuscripts more auditable and reproducible and to reduce common preparation errors — not to write the paper for you.
No. MedSci Skills is not a clinical-decision-support or diagnostic tool and
makes no clinical-validation claim about itself. It supports manuscript and research
workflow only (see SECURITY.md for the scope boundary).
Yes. It bundles a set of EQUATOR-network reporting checklists and risk-of-bias /
appraisal tools (STROBE, CONSORT and CONSORT-AI, STARD and STARD-AI, PRISMA and
PRISMA-DTA, TRIPOD and TRIPOD+AI, QUADAS-2, RoB 2, ROBINS-I, AMSTAR 2, and more) and
audits a manuscript item by item. The authoritative current list and count are in
metadata/catalog_counts.json and the
README.
Yes. References are checked against PubMed / CrossRef / OpenAlex (including
first-author and full-author cross-checks) before they are trusted; the toolkit
never writes references from model memory. See /verify-refs.
Yes. Deterministic detectors check numerical and cohort arithmetic, pooled-estimate
consistency, and cross-artifact drift across the manuscript, tables, figures, and
submission package. These are part of the MedSci-Audit detector layer
(see MEDSCI_AUDIT.md).
A general assistant drafts prose. MedSci Skills adds the clinical-submission layer a generic tool lacks: reporting-guideline audits, reference verification, numerical / cross-artifact consistency gates, journal-specific profiles, and ICMJE/IRB form support — deterministic checks that run before a reviewer sees the manuscript.
It is intentionally narrow. It does not span chemistry, drug discovery, or bench
biology, and it is not an autonomous research agent. It is one physician-researcher's
clinical-manuscript pipeline with integrity gates (see
ROADMAP.md for the scope boundary).
Yes — MIT licensed (see LICENSE; note any per-file carve-outs noted
in the file headers).
It has early community adoption signals (stars, forks, and a few named uses) for
a niche biomedical-workflow repository — not widespread adoption. The honest,
continuously updated record is in IMPACT.md.
Use CITATION.cff or the archived Zenodo DOI listed in the
README.