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 is highly actionable with a clear phased workflow, verified tool examples, and unusually honest live-tested gotchas. It is held back by inlined textbook biology and reference tables that inflate tokens, monolithic structure with no external reference files, and small consistency defects in the Core Tools table.
Suggestions
Move the TargetScan and miRTarBase 'Computational Procedure' download-and-process blocks (and the well-studied target/disease reference tables) into a references/ file, keeping SKILL.md as a lean overview with one-level-deep links.
Trim textbook content Claude already knows — the ncRNA class size/function list and mechanism recaps — keeping only the operational guidance (which database to query, which tool to call, how to grade evidence).
Fix the Core Tools table: remove the duplicate 'miRBase_get_mirna' row and correct the shifted LNCipedia descriptions so each tool's 'Use For' matches its actual capability.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The bulk is genuinely operational (tool tables, verified call/response examples, gotchas like the broken MODOMICS_get_modification tool), but several sections restate textbook biology Claude already knows — e.g. the ncRNA class list ('rRNA (120-5000 nt): Ribosome components', 'tRNA (~76 nt): Amino acid delivery') and recaps of miRNA/lncRNA mechanisms — and the well-studied target and disease-association tables pad the token budget. A duplicated '## Limitations' heading also signals unedited accumulation. | 3 / 5 |
Actionability | Mostly executable guidance: copy-paste-ready TargetScan download/rank code, miRTarBase processing code, and verified `tu.run_tool(...)` calls with real response examples. Minor gaps: the Core Tools table has duplicate and mismatched rows ('miRBase_get_mirna' listed twice; LNCipedia rows where descriptions appear shifted, e.g. 'LNCipedia_get_lncrna_publications | lncRNA sequence (FASTA format)'), which could misroute a tool choice; call syntax also varies between `tu.run_tool` and bare `tool(param=...)` styles. | 4 / 5 |
Workflow Clarity | A clear five-phase sequence (Identity → Targets → Expression → Disease → Interpretation) with an evidence-grading framework (T1–T4) and report-structure checklist that function as validation checkpoints ('look for CLIP-seq or reporter assay evidence before asserting', 'always verify via DisGeNET or PubMed'). Not a 5 because there are no explicit error-recovery loops (e.g. what to do when a tool call fails or returns no results), leaving minor validation gaps. | 4 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, assets/), so everything lives in one ~290-line SKILL.md. Section headers are clear, but content that clearly belongs in separate reference files is inlined — the two long 'Computational Procedure' download-and-process blocks and the well-studied target/disease tables — matching the 'some structure but content that should be separate is inline' anchor. | 3 / 5 |
Total | 14 / 20 Passed |