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.
A highly actionable skill body with excellent copy-paste tooling, explicit decision routing, and real error-recovery loops. Its weaknesses are redundancy (the simplify and tool-choice guidance is repeated in multiple sections), a competing mid-document workflow, inlined bulk that belongs in references, and six broken reference-file links.
Suggestions
Consolidate the three separate explanations of simplify/p.adjust-denominator behavior and the two gseapy-vs-clusterProfiler tool-choice sections into a single authoritative section, cutting the duplicated guidance in PRIMARY SCRIPTS and Analysis conventions.
Create or remove the six missing references files (cross_validation.md, id_conversion.md, organism_support.md, common_patterns.md, multiple_testing.md, report_template.md) — every 'See:' pointer in the body should resolve to a real file.
Move the six-tool Sample-Level scoring table's 'verified live' output transcripts and the bulk of the Tool Reference tables into references/tool_parameters.md, keeping only the input/output/answers distinction inline.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~500-line body is dense with genuinely non-obvious operational knowledge (tie-breaks, library snapshots, DEG filter conventions), but has clear tightening targets: the simplify/p.adjust-denominator explanation appears in three places (PRIMARY SCRIPTS, Analysis conventions, and the enrichgo runner notes), the 'verified live' run-transcript dumps under the six scoring tools are padding, and the Input Parameters table duplicates what references/tool_parameters.md covers. Mostly efficient content with several unnecessary or duplicated sections — matching the middle anchor rather than the noticeably-verbose level below, since almost none of it explains concepts Claude already knows. | 3 / 5 |
Actionability | Fully executable, copy-paste-ready CLI invocations with flags for all three scripts (e.g., the gseapy_enrichment_runner.py and enrichgo_runner.py bash blocks), parseable output markers ('# TIES_AT_TOP: n=K', '# CANDIDATE_RANK', '# TOPN_BY_ADJ_PVALUE:'), a question-phrasing-to-filter mapping table, and concrete tie-break conventions. The specific commands cover the common cases directly. | 5 / 5 |
Workflow Clarity | RULE ZERO gives an unambiguous first step (scan for pre-computed results), the PRIMARY SCRIPTS section provides explicit 'When to use' selectors, and the document includes genuine feedback loops ('Re-run after fixing and the ties at top should drop sharply'; 'Cross-check before reporting: count your filtered gene list'). It falls short of the top anchor because the mid-document Quick Start Workflow and Core Principles sections present a second, competing entry point (direct gseapy.enrichr + ToolUniverse cross-validation) that tensions with the scripts-first guidance at the top, leaving the canonical sequence ambiguous in the middle of the file. | 4 / 5 |
Progressive Disclosure | Structure and signaling are good (section headers, one-level-deep references, four working reference files plus four real scripts), but 6 of the 11 cited reference files do not exist (cross_validation.md, id_conversion.md, organism_support.md, common_patterns.md, multiple_testing.md, report_template.md) — navigation to them dead-ends. Additionally, bulk that belongs in references is inlined (the six-tool scoring table with live-verification output dumps, the full Tool Reference tables). This lands on 'some structure but could be better organized' rather than 'good structure' above it, which requires references to be mostly clear AND present. | 3 / 5 |
Total | 15 / 20 Passed |