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 domain-rich, actionably written skill body with concrete tool parameters, mandatory verification checkpoints, and a phased workflow. Its two real weaknesses are inconsistent phase numbering between the overview and the detailed sections, and a References section pointing at files that are absent from the bundle (while the bundle's one script is orphaned).
Suggestions
Reconcile phase numbering: the Workflow Overview lists "Phase 5.5: Literature" and "Phase 6: Report Synthesis", but the Key Tools sections label "Phase 5.5: Safety & Pharmacogenomics" and "Phase 6: Literature" — renumber so overview and details match, and add a Phase 6 Report Synthesis section covering the report-first principle and output format.
Fix the reference bundle: TOOLS_REFERENCE.md, API_USAGE_PATTERNS.md, TREATMENT_ALGORITHMS.md, REPORT_TEMPLATE.md, EXAMPLES.md, and CHECKLIST.md are linked in the References section but do not exist in the bundle — either ship them in references/ or trim the list to files that exist; likewise reference or remove the orphaned scripts/gdsc_drug_response.py.
Move the cancer-type biomarker-to-drug mappings and resistance-mutation catalogs into TREATMENT_ALGORITHMS.md, keeping only the evidence-hierarchy logic and a few exemplar mappings inline to slim the overview and reduce stale-approval risk in SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and specialized — biomarker-to-drug mappings, resistance frequencies ("C797S (10-25%), MET amp (15-20%)"), and parameter gotchas — knowledge Claude cannot reliably recall, so it earns its tokens rather than padding. Not 5 because the cancer-type drug-mapping blocks and resistance catalogs largely duplicate what TREATMENT_ALGORITHMS.md is supposed to hold, and could be trimmed from the overview. | 4 / 5 |
Actionability | Highly concrete tool guidance: exact parameters with corrections ("`variant_id` (numeric, e.g., 4170)", "param: `condition`, NOT `disease`", "`ensemblId`, camelCase"), a runnable command ("python3 skills/tooluniverse-drug-drug-interaction/scripts/pharmacology_ref.py --type cyp_substrate --drug drugname"), and example query values ("molecular_profile='EGFR C797S'"). Not 5 because there is no complete end-to-end example tool call in the body — worked examples are delegated to EXAMPLES.md. | 4 / 5 |
Workflow Clarity | A clear phase sequence (0→6) with per-phase tool assignments and an explicit hard checkpoint ("Phase 5.5: Safety & Pharmacogenomics (MANDATORY — do NOT skip)" / "You MUST call FAERS... before finalizing the report") plus a Phase 0 tool-verification table. Held below 5 by a numbering inconsistency: the Workflow Overview defines "Phase 5.5: Literature → Phase 6: Report Synthesis" while the detail sections label "Phase 5.5: Safety & Pharmacogenomics" and "Phase 6: Literature", and no detailed Report Synthesis section exists. | 4 / 5 |
Progressive Disclosure | The References section is well-signaled and one level deep (TOOLS_REFERENCE.md, TREATMENT_ALGORITHMS.md, EXAMPLES.md, etc.), but none of those six files exist in the provided bundle — there is no references/ directory — so the navigation paths dangle. Additionally, the one bundle file present (scripts/gdsc_drug_response.py) is never referenced by the body. Structure matches the 4 anchor, but unverifiable/broken references push it to 3. | 3 / 5 |
Total | 15 / 20 Passed |