Content
71%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 an efficient, highly actionable operator's manual built around real API gotchas, executable code patterns, and a gated nine-step workflow with self-critique. Its main weakness is the broken reference layer: all docs/ paths cited (API reference, interpretation guide, report templates, golden examples) are missing from the bundle, and two bundle files are orphaned, so the skill cannot be followed end-to-end as shipped.
Suggestions
Include the docs/ tree (or fix the paths): docs/alphagenome-api.md, docs/interpretation-guide.md, docs/report-templates.md, and docs/examples/* are referenced as MANDATORY workflow inputs and report templates but do not exist in the bundle, so Steps 0, 5, and 6 cannot be executed as written.
Mention the orphaned bundle files in the body — reference references/citation.bib where AlphaGenome citation is appropriate and add scripts/generate_ontology_mapping.py to the Script Reference table so users can discover them.
Make the Extended Search code block self-contained (define or clearly annotate its dependence on the prior snippet's 'df') and consolidate the duplicated 'uv run' explanation into a single section to tighten token usage.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious, hard-won operational knowledge (e.g., 'score_variant does NOT accept ontology_terms', 'GTF Feather file uses Capitalized column names', 'Sashimi does NOT accept a strand argument') and largely assumes Claude's competence. It falls at score 4 rather than 5 because of minor redundancy — the 'uv run' rule is explained twice (Core Rules and Environment Setup) — and the nested Splicing Zoom Strategy sub-list under References is over-detailed for its placement. | 4 / 5 |
Actionability | The Broad Discovery Scan code block is complete and executable (dotenv loading, client creation, scorer filtering, ranked output) and commands like 'uv run --project $SKILL_DIR /tmp/my_analysis.py --arg1 val1' are copy-paste ready. It stays at 4 rather than 5 because the Extended Search snippet is a fragment depending on an undefined 'df', and the workflow steps point to template/doc files that are not present in the bundle, leaving minor execution gaps. | 4 / 5 |
Workflow Clarity | The nine-step checklist is clearly sequenced with explicit mandatory gates ('Step 0: Review Golden Examples (MANDATORY)', 'Step 5: MANDATORY: Read interpretation-guide.md before interpreting results') and a self-critique validation step ('Step 7: Self-Critique... verify links & claims'), and Common Issues provides error-recovery guidance. It does not reach 5 because each step is a terse one-liner whose operational detail is deferred to doc files that are absent from the bundle, so the checkpoints cannot actually be followed as written. | 4 / 5 |
Progressive Disclosure | The structure and signaling are good — one-level-deep references with descriptions and a script reference table — but checked against the actual bundle, every 'docs/' reference (alphagenome-api.md, interpretation-guide.md, report-templates.md, and all docs/examples/* directories) points to files that do not exist, while bundle files references/citation.bib and scripts/generate_ontology_mapping.py are never mentioned anywhere. Navigation is therefore broken at the most critical points, which is worse than the minor-gaps level of score 4 but not the inlined-wall-of-text pattern of score 2. | 3 / 5 |
Total | 15 / 20 Passed |