Content
92%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-engineered skill body: fully executable commands with a complete worked example, count-before-select validation checkpoints with explicit incomplete/truncated recovery guidance, and a clean one-level-deep split between the overview and two reference files. The main improvable point is redundancy — Quick Start and Typical Workflows duplicate the same command sequence, and the question-to-command mapping is stated twice.
Suggestions
Collapse 'Typical Workflows' into the Quick Start (or reduce it to the two cases the Quick Start does not already cover, e.g. hom-ref and population-cohort composition) to remove the duplicated command sequence.
Merge the 'When to Use' bullet list and the 'Command Selection Guide' into a single question-to-command mapping, keeping the counting-first ordering in one place.
State the coordinate-verification rule once (in Coordinate Provenance, with the CAUTION box) and reference it from Core Rules and Common Mistakes instead of restating it three times.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly every token is service-specific knowledge Claude could not know (sentinel '-1' semantics, 'A zero numeric filter is unset on the server', pagination/RAM behavior, mutual-exclusivity rules, annotation-release caveats), with no explanation of concepts Claude already knows. Minor instances could be trimmed: the 'Typical Workflows' section repeats the Quick Start's identical command sequence with placeholders, the 'When to Use' bullet list and 'Command Selection Guide' both map questions to the same commands, and the coordinate-verification warning is stated three times (Core Rules, Coordinate Provenance, Common Mistakes). Efficient with minor trimmable redundancy — anchor 4 rather than 5. | 4 / 5 |
Actionability | All guidance is executable: fully specified 'uv run scripts/onekgpd_api.py count-samples --chrom chr17 --start 43044295 --end 43170327 --consequence MISSENSE_VARIANT ...' commands with a complete worked Quick Start (real resolved BRCA1 coordinates, source URL, and retrieval tag), concrete meta-script invocations, and placeholder templates explicitly labeled 'replace all angle-bracket placeholders'. Nothing within the skill's scope is pseudocode — the one non-executable step (external coordinate resolution) is outside the skill's commands and is still concretely exemplified — so anchor 5 fits better than 4's 'minor gaps'. | 5 / 5 |
Workflow Clarity | Multi-step processes are clearly sequenced with explicit validation checkpoints and feedback loops: mandatory coordinate resolution first, 'Call the count command FIRST to size the result set, then select only if the count is manageable', 'result_incomplete=true means results cannot support a definitive zero/absence claim. Re-run after service recovery', truncated-cap handling, and a Common Mistakes section with mistake→fix pairs. These are read-only queries, so the destructive/batch cap does not apply, and the count-first rule plus completeness checks constitute explicit validation — matching the anchor-5 feedback-loop pattern. | 5 / 5 |
Progressive Disclosure | The body is a clear overview that keeps only summaries inline (command list, the 22 returned keys, filter-flag digest) and pushes full detail one level deep into two clearly signaled, purpose-described reference files — references/onekgpd_commands.md ('full per-command argument tables and the returned-variant output schema') and references/annotation_vocabularies.md ('the controlled-vocabulary terms accepted by the CSV filter flags') — both verified to exist with matching content. Scripts and the bundled offline data file (assets/kgpe.json, used by onekgpd_meta.py) are referenced by exact path. No nested references, no orphaned or buried content — anchor 5. | 5 / 5 |
Total | 19 / 20 Passed |