Content
57%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 knowledgeable, well-structured skill body with concrete tools and a clear workflow, weakened by over-explanation of basic genomics concepts, missing validation checkpoints in a batch workflow, and a monolithic structure with no progressive disclosure to reference files.
Suggestions
Cut or externalize the 'Key Concepts' section — Claude already knows GWAS, beta/OR, LD, and population stratification; keep only the skill-specific guidance such as the ancestry transferability numbers and 'LOOK UP DON'T GUESS' directive.
Add an explicit validation/verification checkpoint to the workflow (e.g., sanity-check SNP count and effect-size distribution against a PGS Catalog model, and re-query if the association set is unexpectedly small) to satisfy the batch-operation feedback-loop requirement.
Split detailed reference material (Data Sources tool catalog, Applications, Limitations) into one-level-deep files like DATA_SOURCES.md and LIMITATIONS.md, leaving SKILL.md a concise overview that links out with clearly signaled references.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient with valuable skill-specific material (tool names, QC thresholds, ancestry caveats), but the 'Key Concepts' section extensively explains genomics concepts Claude already knows (GWAS study design, beta/OR basics, LD, population stratification) and could be trimmed. | 3 / 5 |
Actionability | Provides mostly concrete, executable guidance — named ToolUniverse tools with parameters, the PRS/z-score formulas, QC filters, and percentile risk categories — with the minor gap that the workflow's headline functions (build_polygenic_risk_score, calculate_personal_prs) are not listed among the documented tools. | 4 / 5 |
Workflow Clarity | A clear six-step sequence is present, but validation/verification checkpoints are missing; because PRS construction is a batch operation over many variants, the rubric caps workflow clarity at 3 absent explicit validate-fix-retry feedback loops. | 3 / 5 |
Progressive Disclosure | Well-organized with clear section headers, but it is a 282-line monolithic file with no external references; conceptual exposition (Key Concepts) and detailed Applications/Limitations sections are inlined that could be split into one-level-deep reference files. | 3 / 5 |
Total | 13 / 20 Passed |