Content
53%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 body is well-organized domain guidance with concrete tool mappings and a sensible reasoning-first workflow, but the progressive-disclosure layer is broken — all six referenced files are missing — and the workflow lacks validation checkpoints. Code examples are near-executable but contain tool-name inconsistencies between sections.
Suggestions
Bundle the six referenced files (REPORT_TEMPLATE.md, STUDY_DESIGN_PROCEDURES.md, WORKFLOW_DETAILS.md, BEST_PRACTICES.md, EXAMPLES.md, QUICK_START.md) into the skill's references directory; currently every 'See X for detailed steps' path in the body is a dead end, leaving the 14-section report template and per-path procedures unreachable.
Add validation checkpoints and fallback loops to the workflow: what to do when a tool returns no hits (e.g., alternate disease names, looser filters), and cross-checking enrollment projections and effect-size assumptions against the precedent trials found in PATH 4 before writing them into the report.
Tighten the body: merge the 'LOOK UP DON'T GUESS' block into the preceding 'look them up from precedent trials' paragraph, move the version/date block into a reference file, and fix the Quick Start's tool-name inconsistencies (ClinicalTrials_search_studies vs search_clinical_trials) so the example matches the tool reference tables.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly dense domain-specific guidance, but it duplicates itself ('Look them up from precedent trials and FDA guidance — do not derive them from first principles' is restated as the 'LOOK UP DON'T GUESS' block), carries a dated version block ('Version 1.0.0', 'Last Updated: February 2026'), and inlines a generic ClinicalTrials.gov pagination script whose details it simultaneously defers to the data-wrangling skill. This fits anchor 3 ('mostly efficient but could be tightened') rather than 4, where only minor trimming would be needed. | 3 / 5 |
Actionability | Concrete tool names are mapped to each of the 6 research paths and the Quick Start is near-executable Python, but the example calls tu.tools.ClinicalTrials_search_studies and OpenTargets_get_associated_phenotypes_by_disease_efoId while the tool reference tables list search_clinical_trials and different OpenTargets methods, so the naming is inconsistent and not reliably copy-paste ready; the example also trails off at 'Compile into feasibility report...'. Mostly executable with minor gaps fits anchor 4; anchor 5 would require consistent, complete, common-case-covering code. | 4 / 5 |
Workflow Clarity | The macro sequence is present and sensible (answer 4 reasoning questions, run 6 parallel paths, build the report first and update progressively), but there are no validation checkpoints or feedback loops anywhere — no 'if a tool returns no hits, try X', no cross-check of enrollment projections against precedent trial enrollment — and the per-path step sequences are deferred to a file that is not in the bundle. Anchor 3 ('sequence present but checkpoints missing or implicit') fits; anchor 4 requires most checkpoints to be present. | 3 / 5 |
Progressive Disclosure | The inline structure would be excellent — clearly signaled, one-level-deep references plus a reference table for REPORT_TEMPLATE.md, STUDY_DESIGN_PROCEDURES.md, WORKFLOW_DETAILS.md, BEST_PRACTICES.md, EXAMPLES.md, and QUICK_START.md — but the bundle contains no files at all: every 'See X for detailed steps / full templates / complete workflow' is a dead end, leaving the skill an overview whose detail layer is entirely missing. Scored against the actual (empty) bundle structure, navigation fails outright, fitting anchor 2; anchor 3 requires the structure to be only imperfectly organized, not broken. | 2 / 5 |
Total | 12 / 20 Passed |