Content
72%Weight 40%Scale 1-3Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-structured, highly actionable API skill body with strong progressive disclosure (real, clearly-signaled reference files) and copy-paste-ready examples. The main gaps are redundant '--fields' warnings that hurt conciseness and the absence of an explicit validation/feedback loop for batch fetches, which caps workflow clarity.
Suggestions
Consolidate the repeated '--fields' guidance: state the rule once in Core Rules and reference it from the Context Efficiency Warning instead of restating it, to reduce redundancy.
Add an explicit validation checkpoint for batch workflows, e.g. after '--count-total', instruct to inspect totalCount and adjust '--limit'/filters before fetching, and note how to handle an empty or error response.
Consider moving the full per-flag enum value lists (status, phase, age-group, study-type) into references/clinical_trials_api.md and keeping only the most common values inline, since they already exist in the reference.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient — flag references, enum values, and worked examples all earn their tokens — but the '--fields always' guidance is repeated across Core Rules, the Context Efficiency Warning, the Response Layout Summary, and the flag reference, and the Context Efficiency Warning largely duplicates the Core Rules; could be tightened. Not a 3 due to this redundancy; not a 1 because nothing explains concepts Claude already knows. | 2 / 3 |
Actionability | Fully executable, copy-paste-ready commands with multiple worked examples (cystic fibrosis, atezolizumab, NCI, pancreatic, NYC), a complete flag reference with concrete enum values, and explicit output paths — matching the score-3 anchor. | 3 / 3 |
Workflow Clarity | Core Rules give a clear recommended sequence ('--count-total first', then fetch with '--fields', paginate with '--page-token') and pagination is well documented, and the prerequisite license-check has a conditional checkpoint; however there is no explicit validation/error-recovery feedback loop for the batch-fetch workflow, which the rubric flags for batch operations, so it caps at 2 rather than 3. | 2 / 3 |
Progressive Disclosure | SKILL.md is an overview with common commands inline and pushes detail to two real, correctly-named, one-level-deep reference files (references/studies_schema.md and references/clinical_trials_api.md, both present in the bundle), clearly signaled via TIP callouts, inline 'See references/...' notes, and a References section — matching the score-3 anchor. | 3 / 3 |
Total | 10 / 12 Passed |