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 well-structured planning skill with excellent progressive disclosure — a full one-level-deep reference bundle, every file real and mapped to a specific output section — and a clear, ordered workflow with a mandatory pre-output consistency gate. Weaknesses are moderate redundancy (hard rules restated in three places, a trigger section duplicating input-validation examples) and the absence of any worked output example or explicit failure-handling for the Step 5 gate.
Suggestions
Consolidate the never-fabricate/never-overstate rules into the Hard Rules section only, and delete the redundant 'Core Function should not' bullets and 'What This Skill Should Not Do' items that restate them — keeping one canonical rule list would remove ~40 lines of repetition.
Drop the 'Sample Triggers' section (or fold it into the frontmatter description); it duplicates the Input Validation examples and trigger guidance belongs in the description, not the body.
Add explicit failure handling for the Step 5 dependency check (e.g., 'If any check fails, revise the affected sections and re-run this checklist before generating output') and a brief worked example of one output section (e.g., a filled-in Section C workload table) to make the output format fully concrete.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient instruction (no biology pedagogy Claude already knows), but it is noticeably repetitive: the never-fabricate rules appear in "Core Function," "Hard Rules 1–3," and "What This Skill Should Not Do"; the Dataset Disclaimer placement is stated in Step 6, Section H, and Hard Rule 2; and "Sample Triggers" largely reiterates the Input Validation examples and content that belongs in the description. This fits the level-3 anchor (mostly efficient but could be tightened); it does not reach level 4's 'minor instances of over-explanation,' since whole duplicated sections could be consolidated into Hard Rules alone. | 3 / 5 |
Actionability | For an instruction-only skill the guidance is largely executable: a fixed 7-step execution order, a mandated output structure (Sections A–L) with per-section content specs, a copy-paste redirect template for out-of-scope input, and concrete method constraints ("count matrices should map to DESeq2 by default; non-count normalized expression matrices should map to limma by default"). It falls short of level 5 because there is no worked example of any output section (e.g., a filled-in Section C comparison table), leaving the exact output format partly to inference; it is clearly above level 3, which expects pseudocode-level vagueness. | 4 / 5 |
Workflow Clarity | Steps are clearly sequenced ("7 Steps (always run in order)") with an explicit mandatory gate — "Step 5 — Dependency Consistency Check (mandatory before output)" with a six-item checklist — plus out-of-scope redirect-and-stop handling and a final self-critical risk review with a fallback plan. It sits at level 4 rather than 5 because the recovery loop is implicit: Step 5 says the check is mandatory before output but never instructs what to do when a check fails (revise which sections, re-run the check), and no checkpoint exists between generating the workflow (Step 6) and the risk review (Step 7). | 4 / 5 |
Progressive Disclosure | The body is a genuine orchestrating overview: the "Reference Module Integration" section maps each of the nine reference files to the specific output section where it must be used (e.g., "references/study-patterns.md → use when selecting the dominant single-cell study pattern in Section B"), references are one level deep, clearly signaled, and all nine referenced files exist in references/. This matches the level-5 anchor (clear overview with well-signaled one-level-deep references, easy navigation); there is no nesting or buried reference content. | 5 / 5 |
Total | 16 / 20 Passed |