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.
The body is action-rich with executable CLI, config schema, and realistic output examples, and the workflow is clearly sequenced for a read-only fetch. Its main weakness is progressive disclosure: multiple referenced bundle files (dataset index, demo input, the script itself) are absent from the skill, breaking the navigation the body promises.
Suggestions
Ship the referenced bundle files the body depends on, or remove/guard the references — the script at skills/eqtl-catalogue-region-fetch/eqtl_catalogue_region_fetch.py, data/dataset_index_r7.tsv, data/dataset_index_r7.provenance.json, and examples/input.json are all missing.
Tighten Gotcha #6: split the single long sentence with stacked parentheticals (33-of-758 probed, QTD000584 row counts) into a short rule plus a compact table or a one-line example, so the core guidance is not buried.
Make workflow validation explicit: add a checkpoint after the tabix fetch (e.g., 'verify n_variants > 0 and the expected columns are present before writing') rather than leaving verification implicit.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly lean and assumes Claude's intelligence with no generic padding, but Gotchas #6 is a very dense run-on with stacked parenthetical statistics, and the Trigger 'Do NOT fire when' list overlaps the Scope section, leaving minor tightening opportunities. | 4 / 5 |
Actionability | Fully executable guidance throughout: copy-paste CLI commands, a concrete JSON config schema, and realistic example outputs (manifest.yaml, variants.tsv rows, report.md) covering the common cases. | 5 / 5 |
Workflow Clarity | A clearly sequenced 5-step workflow (Resolve dataset_id → Pick region → Tabix fetch → Filter by molecular_trait_id → Write outputs) with named error states; validation is mostly implicit rather than called out as explicit checkpoints, and it is a read-only fetch so the destructive cap does not apply. | 4 / 5 |
Progressive Disclosure | Sections are well-organized, but the body repeatedly references bundled files that do not exist in the skill — data/dataset_index_r7.tsv, data/dataset_index_r7.provenance.json, examples/input.json, and skills/eqtl-catalogue-region-fetch/eqtl_catalogue_region_fetch.py — so navigation to those materials is broken. | 2 / 5 |
Total | 15 / 20 Passed |