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.
An information-dense, highly actionable instruction skill with a clearly sequenced 10-phase workflow, concrete tool parameter tables, and real error-handling guidance. Its main defects are structural: the KEY PRINCIPLES list is split/orphaned mid-document, and all seven referenced bundle files are missing, leaving key details (scoring logic, matching algorithms, report template) unreachable.
Suggestions
Ship the referenced bundle files (MATCHING_ALGORITHMS.md, SCORING_CRITERIA.md, TOOLS_REFERENCE.md, REPORT_TEMPLATE.md, TRIAL_SEARCH_PATTERNS.md, EXAMPLES.md, QUICK_START.md) or remove/inline the dead links — progressive_disclosure currently points at files that do not exist, so delegated details like scoring logic and the report template are unreachable.
Fix the broken 'KEY PRINCIPLES' list: items 4–10 (Evidence-graded through English-first queries) are stranded under the '## Molecular Matching Priority' heading; rejoin them with items 1–3 under a single list or renumber them as their own section.
Add one or two fully-specified example tool invocations (e.g., a `search_clinical_trials` and `CTIS_search_trials` call with concrete argument values) so the top-scoring use cases are copy-paste ready without consulting the missing reference files.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dominated by dense, non-obvious reference material (MCP tool parameter tables, required-parameter warnings, CIViC gene IDs, point-valued scoring tables) that Claude cannot know. Minor trimmable redundancy: the opening paragraph restates the frontmatter description, and 'LOOK UP, DON'T GUESS' spends lines telling Claude to prefer verified data. Not 5: these minor padded spots exist; not 3: no explanations of concepts Claude already knows. | 4 / 5 |
Actionability | Concrete, near-executable guidance throughout: exact tool names with required parameters ("`drugbank_get_targets_by_drug_name_or_drugbank_id` | `query`, `case_sensitive`, `exact_match`, `limit` (ALL REQ)"), batch sizes ("process in batches of 10"), explicit scoring arithmetic, and parallelization groups. Not 5: no example tool invocations with actual argument values, and critical logic (biomarker parsing, detailed scoring, report template) is delegated to reference files; not 3: far more executable than pseudocode-level guidance. | 4 / 5 |
Workflow Clarity | The 10-phase workflow is clearly sequenced with a parallelization strategy and error handling that includes try/except wrapping, empty-result checks, fallback tools, and failure documentation in a completeness checklist — so batch operations do have validation/verification steps. Not 5: the 'KEY PRINCIPLES' numbered list is structurally broken — items 4–10 are orphaned under the 'Molecular Matching Priority' heading after its paragraph, muddling the principle list — and phase details are thin (one line each), leaving some checkpoints implicit. | 4 / 5 |
Progressive Disclosure | The body is well-sectioned and the trailing 'Reference Files' table clearly signals one-level-deep references, but every referenced file (TOOLS_REFERENCE.md, MATCHING_ALGORITHMS.md, SCORING_CRITERIA.md, REPORT_TEMPLATE.md, TRIAL_SEARCH_PATTERNS.md, EXAMPLES.md, QUICK_START.md) is absent from the bundle — the references/ directory does not exist, so all seven links are dead and navigation breaks. Not 4: scoring against the actual bundle structure, clearly-signaled references that do not resolve cannot count as good structure; not 2: the in-body organization and reference signaling are present and reasonable. | 3 / 5 |
Total | 15 / 20 Passed |