Content
60%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.
Operationally rich content — specific tools, SOAP quirks, fallback strategies, and a genuinely useful evidence-grading reasoning framework — undermined by padding in the Summary/Key Features/Best for sections and by a broken bundle layout: it references files that don't exist and ignores the script it actually ships. Trimming the redundant sections and fixing the reference structure would lift this substantially.
Suggestions
Fix the reference structure: either add the referenced QUICK_START.md and .env.template to the bundle, or remove those pointers; and reference scripts/metabolism_ref.py from SKILL.md so the shipped script is discoverable.
Delete the redundant "Summary", "Key Features", and "Best for" sections (~25 lines) — they restate the Use Case and Research Workflow sections verbatim in concept and add no operational value.
Move the Input Parameters tables and Usage Patterns to a one-level-deep reference file (e.g., USAGE.md), keeping SKILL.md as a concise overview of the 4-phase pipeline, SOAP quirks, and reasoning framework.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient — the SOAP parameter handling, response-format variations, fallback hierarchy, and adduct-mass tables are non-obvious and earn their tokens — but there is clear padding: the "Summary" section restates the 4-phase pipeline, "Key Features" ("✅ 100% test coverage with working pipeline") is self-congratulatory metadata, and "Best for" duplicates the "Use Case" section. Not a 2 because the majority of the body is dense operational guidance rather than concept explanation, but it could be tightened by roughly a fifth. | 3 / 5 |
Actionability | Concrete, executable guidance throughout: exact tool names ("HMDB_search", "metabolights_get_study"), exact parameters ("operation=\"search\"", "do not use endpoint or method"), numeric thresholds ("±5 ppm", "cosine > 0.7", "fold-change > 1.5"), and response-format handling ("Always check response type with isinstance()"). Per the instruction-skill scoring note, absent code is not penalized when guidance is actionable, but no inline runnable example exists and the promised Python examples live in a QUICK_START.md that is not in the bundle — minor gaps that keep this below anchor 5. | 4 / 5 |
Workflow Clarity | The 4-phase pipeline is clearly sequenced with explicit fallback checkpoints (primary → fallback → "continue with remaining phases"), per-phase numbered steps, and a Common Issues section with cause/solution diagnosis. This is a batch operation, and validation is largely addressed (response-type checks before field access, fallback on failure), so the missing-validation cap at 3 does not apply; it is not 5 because there is no explicit verify-the-output step before finalizing the report. | 4 / 5 |
Progressive Disclosure | The file structure is misaligned with the actual bundle: the body's only outbound references — "See QUICK_START.md" and "check .env.template" — point to files that do not exist, while the one real bundle file (scripts/metabolism_ref.py) is never referenced from SKILL.md at all. Additionally, ~290 lines of input-parameter tables, usage patterns, and reasoning framework are inlined in SKILL.md that could live in reference files. Well-labeled section headers prevent a score of 1, but broken references plus orphaned bundle content match anchor 2 (content that belongs in separate files inlined / references not usable). | 2 / 5 |
Total | 13 / 20 Passed |