Lean and assumes competence: no explanation of what a forest plot, OR, or heterogeneity is — only the error-prone conversion formulas, decision tables, and the command to run the tool. Every line delivers non-obvious operational detail rather than padding Claude already knows.

Fully executable: a copy-paste `tu run MetaAnalysis_run ...` JSON invocation with realistic study rows, a runnable `python .../meta_analysis.py --input studies.csv` with exact CSV column sets, and explicit formulas (e.g. `(ln(U) − ln(L)) / (2 × 1.96)`) for the conversions the script performs.

Clear 5-step sequence with an explicit validation feedback loop — the 'Scale foot-gun' note is a checkpoint that tells Claude to ignore the tool's interpretation field and re-`exp()` ratio results, and 'Honest limitations' / retraction checks act as quality gates before trusting inputs.

Well-organized single-file skill with one-level-deep pointers: the body keeps the overview and decisions inline while offloading the tedious conversions to `scripts/meta_analysis.py` and pointing to sibling-skill references (e.g. `references/cox_regression.md`) for adjacent detail. References are real, clearly signaled, and not nested.

Lists multiple concrete actions: 'pool effect sizes across studies (odds ratios, risk ratios, hazard ratios, mean differences, correlations, GWAS betas)', 'fixed- or random-effects models', 'quantify heterogeneity (Q, I², τ²)', 'build a forest plot', and 'converting OR/HR/CI, two-group means±SD, proportions, and correlations into the (effect, SE)'. Each is a specific, concrete capability.

Explicitly answers both 'what' (pool effect sizes, quantify heterogeneity, forest plot, effect-size+SE preparation) and 'when' via 'Use when you have results from MULTIPLE studies and need a single pooled estimate, or to synthesize evidence from a systematic review / multiple GWAS / replicated experiments.'

Natural terms a user would say appear throughout: 'multiple studies', 'single pooled estimate', 'systematic review', 'multiple GWAS', 'replicated experiments', plus measure names (OR, RR, HR) users cite. Good coverage of phrasings users actually produce.

Clear niche — quantitative pooling of effect sizes across multiple studies — with distinct triggers ('single pooled estimate', 'heterogeneity', 'forest plot') that separate it from the sibling literature-search skill, which is named explicitly as the complementary tool.