Content
75%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-organized, lean instruction-only skill with a genuinely concrete seven-step methodology, specific statistical method mappings, a validation checklist, and honest error handling. The main gaps are minor: no worked example or executable snippet anywhere, validation as a post-hoc checklist rather than inline checkpoints, and a redundant Input Validation section duplicating Error Handling.
Suggestions
Add one short worked example (e.g., a minimal R0-from-growth-rate or epidemic-curve snippet) so the most technical steps are executable rather than directional.
Fold the "Input Validation" section into "Error Handling" to remove the duplicated stop-on-missing-input guidance.
Fix the "Hills criteria" typo (should be "Hill's criteria for causation") and consider promoting the Quality Checklist items into inline checkpoints at the relevant methodology steps.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The 65-line body is efficient and imperative ("Compute incidence rate (person-time denominator)", "draw a DAG to identify confounders and colliders") with no padding or explanation of concepts Claude already knows. It stays at 4 rather than 5 because the "Input Validation" section largely restates "Error Handling" (both tell the model to stop on missing inputs/out-of-scope requests) and the canned refusal quote is template boilerplate that could be trimmed. | 4 / 5 |
Actionability | Concrete method-to-data mappings appear throughout: "logistic for OR, Poisson/negative binomial for rates, Cox for time-to-event", "incidence rate (person-time denominator)", "Estimate R0 from early epidemic growth rate or next-generation matrix", plus named tools (EpiEstim, DAGitty). Per the code_vs_instruction note the absence of code is not itself penalized, but it stays at 4 rather than 5 because several steps remain directional ("Apply propensity score methods if needed", "Translate findings into public health actions") with no worked example, formula, or command, and 3 would undervalue how specific the method selection guidance is. | 4 / 5 |
Workflow Clarity | The seven numbered steps give a clear, correctly sequenced methodology (question → data → descriptive → measures → analytic → modeling → communication), and the Quality Checklist supplies validation checkpoints ("Confidence intervals provided for all estimates", "Confounders identified via DAG and adjusted for"). This is analysis work, not destructive/batch operations, so the workflow-clarity cap does not apply; it falls short of 5 only because validation lives in a separate end-of-work checklist rather than inline checkpoints with explicit fix-and-retry loops between steps. | 4 / 5 |
Progressive Disclosure | The body is cleanly sectioned (When to Trigger, Methodology, Databases, Output Format, Checklist, Error Handling) with no buried references and no nested files — the bundle directories (references/, scripts/, assets/) don't exist, so nothing dangles. It scores 4 rather than 5 because at 65 lines it exceeds the under-50-line simple-skill threshold and is entirely monolithic — all guidance is inline (e.g., the databases list) with nothing split out — yet it is well above 3 since no content is misplaced or hard to navigate. | 4 / 5 |
Total | 16 / 20 Passed |