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.
Highly actionable content with live-verified tool examples, a well-sequenced five-phase workflow, and honest evidence grading, but token efficiency and organization lag: the tool table contains duplicated/mislabeled rows, a section heading is duplicated, and lengthy lookup tables and download procedures are inlined instead of split into reference files.
Suggestions
Deduplicate and correct the Core Tools table: remove the second miRBase_get_mirna and PubMed_search_articles rows, and fix the 'LNCipedia_get_lncrna_publications' and 'LNCipedia_search_ncrna_by_type'/'get_lncrna_xrefs' descriptions that currently describe other tools' functions.
Rename the first '## Limitations' section (line 226) to '## Computational Procedures (Download-and-Process)' so it no longer duplicates the true Limitations section at line 291.
Move the curated lookup tables (well-studied miRNA targets, key ncRNA-disease associations) and the TargetScan/miRTarBase download pipelines into a references/ file, keeping SKILL.md as a lean overview with one-level-deep pointers.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient domain-specific guidance, but the tool table has duplicated rows with mismatched descriptions (miRBase_get_mirna listed twice, PubMed_search_articles twice, 'LNCipedia_get_lncrna_publications' labeled 'lncRNA sequence (FASTA format)'), and the '## Limitations' heading appears twice (once wrapping computational procedures). Several blocks could be tightened but the content is largely non-redundant domain knowledge. | 3 / 5 |
Actionability | Fully executable throughout: verified live tool calls with real responses (RNAcentral URS examples, MODOMICS queries with output shown), concrete parameters (clip_min=1, taxid=9606), and complete copy-paste pandas pipelines for TargetScan download and miRTarBase processing. It even documents a verified-broken tool (MODOMICS_get_modification) with the exact failure mode and a safe alternative. | 5 / 5 |
Workflow Clarity | A clear Phase 0–4 sequence with class-identification decision rules, an evidence-grading framework (T1–T4) acting as a validation checkpoint, and explicit fallback loops (broken tool -> search/list; download unavailable -> reference table or PubMed). Falls short of 5 because there is no explicit cross-source verification step for results (e.g., confirm a target in both ENCORI and literature before reporting). | 4 / 5 |
Progressive Disclosure | No bundle files exist and the skill is a single ~290-line file; structure is well-sectioned, but ~60 lines of download-and-process procedures (TargetScan, miRTarBase) and long curated association tables are inlined where reference files would fit, and the duplicated 'Limitations' header hurts navigation. This sits between 'content that should be separate is inline' (3) and 'minor organization gaps' (4) but nearer the former. | 3 / 5 |
Total | 15 / 20 Passed |