Content
56%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.
The tool-facing half of this skill is excellent — a precise parameter table, hard-won gotchas, executable script modes, and clear workflow chains. The theoretical half is a genetics textbook cram sheet that re-teaches standard concepts Claude already knows and should be split out into reference files, which is what holds the skill back on both conciseness and structure.
Suggestions
Move the Theoretical Reasoning, Mendelian Framework, Hfr Mapping, and trap sections into references/ files (e.g., references/formulas.md, references/traps.md), keeping SKILL.md to the tool table, gotchas, workflows, and script modes with one-line pointers to the references.
Trim or delete explanations of textbook concepts Claude already knows (Punnett-square mechanics, HWE genotype expectations, Fis/Fst/Fit definitions, penetrance vs expressivity), keeping only the non-obvious curated items like the PGS-vs-h² trap and the duplex-sequencing detection threshold.
Add one worked example with concrete inputs and expected outputs (e.g., a real rsID traced through gnomad_search_variants -> gnomad_get_variant, and an HWE computation with numbers), and list the parameter signatures for the PopGen_hwe_test / PopGen_fst / PopGen_inbreeding / PopGen_haplotype_count tools.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Roughly two-thirds of the body re-explains textbook genetics Claude already knows — Punnett-square step-by-steps ('Draw Punnett square(s) for each gene', 'Multiply independent gene ratios (e.g., 3:1 x 3:1 = 9:3:3:1)'), HWE genotype expectations ('expected genotypes AA=p^2, Aa=2pq, aa=q^2'), Fis/Fst/Fit definitions, penetrance-vs-expressivity, and Bridges' non-disjunction basics. These are padded sections competing with the context window, even though the tool-specific material (gotchas, workflows, script modes) is genuinely additive. | 2 / 5 |
Actionability | Highly concrete: a tool table with required parameters ('variant_id (REQUIRED)... Param is "variant_id" NOT "rsid"'), copy-paste CLI invocations verified against the bundled script ('popgen_calculator.py --type hwe --AA N1 --Aa N2 --aa N3'), and four named tool-chain workflows. Minor gaps: the PopGen_* tools are named without parameter specs, and there is no worked example showing inputs and expected outputs. | 4 / 5 |
Workflow Clarity | Clear sequences throughout — 'gnomad_search_variants -> gnomad_get_variant(dataset="gnomad_r4") -> MyVariant_query_variants' chains, an explicit tools-first-then-script-then-formulas priority, and a validation checkpoint in 'Verify ratios sum to 1.0' with error-recovery guidance. Not a 5 because validation is present only in the Mendelian framework, not in the tool-query or computation workflows (e.g., no cross-checking of tool results against the script). | 4 / 5 |
Progressive Disclosure | The bundle is real and correctly wired — scripts/popgen_calculator.py exists and its CLI flags match the documented modes, referenced one level deep. But the body is a ~360-line monolith where the theoretical formulas, Mendelian framework, Hfr mapping, and trap catalogs clearly belong in references/ files, leaving SKILL.md as an overview plus the tool-critical material. | 3 / 5 |
Total | 13 / 20 Passed |