Content
63%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 body delivers strong domain-specific reasoning frameworks and decision thresholds that are hard to reconstruct from general knowledge, but suffers from redundancy (two overlapping reasoning chains, a Key Concepts section teaching basics Claude already knows), a long time-sensitive dead-API digression, and a fully inlined structure with no progressive disclosure into reference files. The one broken internal cross-reference ("Limitations section above") adds minor confusion.
Suggestions
Merge the "REASONING STRATEGY — Start Here" chain into the "Fine-Mapping Reasoning Framework" (they cover the same LD-first → functional-tiebreak → eQTL-bridge logic) and delete the "Key Concepts" section, which defines credible sets and posterior probability Claude already knows.
Move the dead GWASSumStats_* tool narrative into a short 'Deprecated tools' callout (one sentence plus the deprecation link), and fix the incorrect cross-reference that calls the Limitations section "above" when it appears below the DIY tools section.
Split the tool catalog, DIY calculator details, and PP/L2G interpretation tables into reference files (e.g. references/tools.md, references/interpretation.md) referenced one level deep from SKILL.md, keeping the main file a lean overview and reasoning framework.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with genuinely non-obvious domain guidance (r² thresholds, PP and L2G interpretation bands, tiebreaker ordering), but it is padded: the "Key Concepts" section defines credible sets and posterior probability Claude already knows, the "REASONING STRATEGY — Start Here" chain duplicates the "Fine-Mapping Reasoning Framework" Steps 1–5 (LD first → functional tiebreak → eQTL bridge), and the dead-API paragraph is long and time-sensitive ("as of this writing") without being framed as a deprecated section. Mostly efficient with some clearly trimmable material — fits anchor 3 better than anchor 2, since the padding is a minority of the file. | 3 / 5 |
Actionability | Concrete, executable guidance throughout: named tool calls with explicit actions ("Always call OpenTargets_get_variant_credible_sets... If the posterior probability is < 0.5..."), calculator tools with full parameter signatures, and verified live outputs. This is an instruction/tool-orchestration skill, so the absence of runnable code is acceptable per the rubric's scoring notes; the minor gap is that the calculator signatures (beta=[...], se=[...]) are placeholders rather than copy-paste ready, and no worked Python example is shown despite the "COMPUTE, DON'T DESCRIBE" directive. | 4 / 5 |
Workflow Clarity | A clear 5-step sequence (lead SNP vs causal → LD structure → credible set analysis → colocalization → tiebreakers) with explicit decision checkpoints at each step ("If the posterior probability is < 0.5, the lead SNP is likely NOT causal", L2G thresholds of 0.7/0.5/0.3). This is not a destructive or batch operation, so no validation cap applies; what keeps it at 4 rather than 5 is the absence of any validate/fallback loop for the anticipated failure path where a locus has no precomputed credible set — the branch is described but not sequenced. | 4 / 5 |
Progressive Disclosure | No bundle files exist and the ~185-line body is monolithic: well-organized section headers aid navigation, but everything — the tool catalog, DIY calculator details, interpretation bands, FAQ, and references — is inlined in SKILL.md with no one-level-deep reference files. Content that clearly belongs in separate files (tool reference, interpretation guide) is inline, matching anchor 3; there is also an incorrect cross-reference ("This skill's 'Limitations' section above" at the DIY section — Limitations appears below it). | 3 / 5 |
Total | 14 / 20 Passed |