Content
68%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 instructional core is excellent: a concrete, domain-specific algorithm and a complete worked example make the skill highly actionable, and reference files are real and well-integrated. Weaknesses are a generic boilerplate tail (~36% of the body) that wastes tokens, and abstract validation steps where a concrete completeness checklist would serve better.
Suggestions
Delete the generic template sections (Required Inputs, Recommended Workflow, Output Contract, Failure Handling, Input Validation, Quick Validation, User Checkpoints); they restate what Claude already does by default and consume roughly a third of the token budget without skill-specific value.
Replace the abstract final validation pass with a concrete Methods-completeness checklist (e.g., verify every applicable CONSORT/STROBE/PRISMA item is addressed, statistical reporting includes effect sizes with 95% CIs and missing-data strategy, and Methods content matches what Results reports).
Move the full worked-example Methods draft into a reference file (e.g., references/example_rct_methods.md) and keep a short excerpt plus pointer in SKILL.md to reduce inline weight.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The core sections (When to Use, Implementation Details, worked example) are dense and domain-specific, but roughly a third of the body is generic template boilerplate ('Required Inputs', 'Output Contract', 'Failure Handling', 'Quick Validation' containing only 'No local script validation step is required', 'User Checkpoints') that adds no skill-specific value. Not 2: the majority of tokens earn their place; not 4: the padded tail is more than minor over-explanation. | 3 / 5 |
Actionability | The 10-step algorithm specifies executable, concrete details (block randomization size 4, Shapiro-Wilk, Holm correction, Cohen's d with 95% CI, IUPAC nomenclature, gene-symbol italics), and the worked example provides a complete input-to-output manuscript draft that is directly adaptable. Per the scoring notes, an instruction-only skill with guidance this concrete and a full example covering the flagship case (RCT) merits the top score. Not 4: no material execution gaps exist for the primary use case. | 5 / 5 |
Workflow Clarity | The Methods-section algorithm is clearly sequenced (identify study type, map checklist, write prose, specify details, statistics, bias control, ethics, data, terminology, journal adaptation), but validation checkpoints are abstract ('Run a final validation pass for completeness, consistency, and safety') and the Recommended Workflow section is generic filler. Not 4: no concrete verification step or checklist of what to check is written; not 2: a real, coherent 10-step sequence is present. | 3 / 5 |
Progressive Disclosure | Three real, one-level-deep reference files (imrad_structure.md, reporting_guidelines.md, writing_principles.md) are clearly signaled both in a reference list and inline in the algorithm steps. Not 5: the full ~40-line sample Methods draft is inlined in SKILL.md rather than split into an examples reference, keeping the main file heavier than an overview needs to be. | 4 / 5 |
Total | 15 / 20 Passed |