Content
60%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-structured, highly actionable statistical-analysis body with genuine reference and script bundles that match their in-SKILL.md documentation. Its weakness is token efficiency: generic statistics pedagogy, plotting boilerplate, textbook lists, and repeated reference pointers inflate it well beyond what the overview needs, and two code snippets have API-signature bugs.
Suggestions
Cut the 'Key Advantages', 'Best Practices', 'Common Pitfalls', and 'Support and Further Reading' sections — that content is either generic statistics knowledge Claude already has or belongs in references/assumptions_and_diagnostics.md; keep the Getting Started Checklist as the workflow spine.
Move the four full worked code examples (t-test, ANOVA, regression diagnostics with plotting, Bayesian t-test) into a references/analysis_recipes.md and keep only one compact quick-start snippet inline.
Fix the two API mismatches: pingouin's `compute_effsize_from_t` returns a single float (compute the CI separately), and `FTestAnovaPower().solve_power` takes `nobs` (and `k_groups`/proportion kwargs) rather than `ngroups=3`.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is noticeably verbose with several padded sections that teach Claude things it already knows: ~30 lines of matplotlib residual-plot boilerplate, 'Key Advantages' explaining what Bayesian inference is, 'Best Practices'/'Common Pitfalls' lists of generic statistics-ethics definitions (p-hacking, HARKing), a textbook citation list, and online-resource links. It exceeds the anchor-3 'some unnecessary explanation' profile — there are multiple whole sections that earn no token — but is not severely padded throughout (the test-selection table, effect-size benchmarks, and APA templates are genuinely valuable), so it lands at 2 rather than 1. | 2 / 5 |
Actionability | Mostly executable, copy-paste-ready code for the common cases: pingouin t-test/ANOVA with result extraction, statsmodels power analysis, pymc Bayesian t-test, and fill-in-the-blank APA report templates. It is not a 5 due to minor correctness gaps: `d, ci = compute_effsize_from_t(...)` treats pingouin's function as returning a (d, CI) tuple when it returns a single float, and `FTestAnovaPower().solve_power(..., ngroups=3, ...)` does not match statsmodels' actual signature (which takes nobs and derives groups). Far above the anchor-3 pseudocode profile. | 4 / 5 |
Workflow Clarity | A clear sequence is present — 'Getting Started Checklist' runs question → test selection → power analysis → assumption verification via assumption_checks.py → analysis → effect sizes → reporting → sensitivity analysis — with explicit checkpoints ('ALWAYS check assumptions before interpreting test results') and violation-remediation branches (Welch's t-test, non-parametric alternatives). It falls short of 5 because the 'Workflow Decision Tree' is shallow (every NO branch just says 'Continue') and there are no error-recovery loops for runtime failures, e.g. convergence failures in pymc sampling. | 4 / 5 |
Progressive Disclosure | Structure is good and the references are real and one level deep: all five `references/*.md` files exist as listed in the Resources section, are clearly signaled with per-topic descriptions, and `scripts/assumption_checks.py` exists with exactly the six functions the body documents. It is not a 5 because the overview itself carries long inline content that belongs in the reference files — four full code examples plus a 30-line plotting block push SKILL.md well past overview length, and some sections repeat the same reference pointers three times. | 4 / 5 |
Total | 14 / 20 Passed |