Aperivue/medsci-skills

skills/peer-review/references/exemplar_reviews/ models how a reviewer turns a finding into a well-phrased comment (currently: AI overclaiming, reference-standard validity, data leakage, missing calibration). Add one more finding type — e.g. missing blinding, small-sample overclaim, inadequate baseline comparator, or multiplicity not addressed. What to do Author a synthetic, from-scratch example (no copied review text, no real citations, English) following exemplar_reviews/README.md. Show the four moves: anchor the location → state the gap → partner-voice (Aczel-compliant) phrasing → severity calibration (design-level → Major #1). Acceptance: new .md matching the pattern, synthetic only, CI green. Pointers: skills/peer-review/references/exemplar_reviews/README.md + the existing examples. Open it in your browser: skills/peer-review/references/exemplar_reviews/ — you can create or edit the file right there, no git and no clone.

Much of the medical-AI agent-skill ecosystem is Chinese-led; a README.zh-CN.md would widen reach. What to do Translate README.md to Simplified Chinese as README.zh-CN.md. Add a one-line language switcher near the top of both READMEs (e.g. English | 简体中文). Keep code blocks, commands, and links unchanged. Acceptance: README.zh-CN.md present + cross-linked, no broken links. Pointers: README.md. Open it in your browser: README.md — you can create or edit the file right there, no git and no clone.

make-figures carries a curated exemplar gallery the Critic Loop scores drafts against. It has two kinds: Flow diagrams (PRISMA / STARD / CONSORT / STROBE / pipeline) live in exemplar_diagrams/ as an image + metadata triple. Non-flow figure "anatomy models" live in skills/make-figures/references/exemplar_plots/ as a single hand-authored .md file that describes, element by element, what a publication-grade plot of that type must contain — authored from scratch as a teaching model, not extracted from any published figure (so no image and no attribution are needed). This issue is about the second kind (the exemplar_plots/ .md models). Already covered (don't re-add these): bland_altman, calibration_plot, clinical_timeline, confusion_matrix, decision_curve, forest_plot, imaging_panel, km_curve, manhattan_plot, mrmc_roc, roc_pr. Add a plot type we don't cover yet — for example: a funnel plot (meta-analysis publication bias), violin / box comparison, swimmer plot, waterfall plot (RECIST

/manage-refs renders references via journal CSL styles (14 included). Add one for a medical / medical-AI journal that lacks one. What to do Download the official CSL from the Zotero Style Repository (https://www.zotero.org/styles) for a journal not yet covered. Add it to skills/manage-refs/citation_styles/ and note it in that folder's README.md. Sanity-check it renders with pandoc citeproc. Acceptance: valid .csl, README updated. Pointers: skills/manage-refs/citation_styles/ (+ its README). Open it in your browser: skills/manage-refs/citation_styles/ — you can create or edit the file right there, no git and no clone.

/deidentify ships 11 country PHI locale packs (kr, us, jp, cn, de, uk, fr, ca, au, in, it — Italy arrived in #544). Add one more — e.g. Brazil, Spain, Netherlands, Mexico, Taiwan… What to do Copy skills/deidentify/locales/_template.json to <cc>.json. Fill in that country's PHI patterns (national-ID / phone / postal-code formats), following an existing locale like us.json or de.json. Cite where the format comes from in a comment/PR note. Optionally add a small test row. Acceptance: new <cc>.json valid + patterns documented, CI green. Pointers: skills/deidentify/locales/, skills/deidentify/SKILL.md. Open it in your browser: skills/deidentify/locales/ — you can create or edit the file right there, no git and no clone.