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.
The body is an efficient, highly actionable phase-based workflow with exact tool calls, parameter corrections, fallback chains, and scoring thresholds. Its main defects are three dead references in the Reference Files section (the per-phase code examples and report template do not exist in the bundle) and a few phases that name tools without call syntax.
Suggestions
Create the missing DIAGNOSTIC_WORKFLOW.md, REPORT_TEMPLATE.md, and CHECKLIST.md files (or remove/correct the Reference Files links) so the referenced per-phase code examples, report template, and completeness checklist are actually reachable.
Give Phases 5–6 the same treatment as Phases 2–4: one concrete example call each (e.g., AlphaFold2/InterPro and PubMed/OpenAlex syntax) instead of bare tool names.
Add one explicit validation checkpoint between phases (e.g., "confirm ClinGen classification before adding a gene to the panel; if Disputed/Refuted, exclude and record") to make the existing implicit gates into stated steps.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and table-driven with no explanations of concepts Claude already knows, and every workflow line carries operational content. However, the PanelApp section includes justification prose ("a genuinely different signal from MARRVEL/ClinGen's disease-gene curation, since it reflects what a real diagnostic lab panel would include today") and rationale sentences that could be tightened to bullets — minor instances of over-explanation, matching the level-4 anchor rather than level 5's every-token-earns-its-place. | 4 / 5 |
Actionability | Concrete, executable guidance dominates: a parameter-corrections table with exact WRONG/CORRECT values, specific calls like `Orphanet_search_diseases(operation="search_diseases", query=keyword)` and `FAVOR_annotate_variant("chr-pos-ref-alt")` with expected outputs (panel id 285), plus exact thresholds (gnomAD frequency classes, ACMG codes, overlap-score bands). Minor gaps remain: Phases 5–6 list only tool names ("AlphaFold2, InterPro domains", "PubMed, BioRxiv/MedRxiv, OpenAlex") without call syntax, keeping it at level 4 rather than fully copy-paste-ready level 5. | 4 / 5 |
Workflow Clarity | The Phase 0–7 sequence is explicit and ordered, with fallback chains for tool failure, evidence-grading tiers, and ClinGen confidence gates that function as decision checkpoints. This is an analysis workflow (not destructive/batch), so the validation cap doesn't apply; it sits at level 4 because explicit validate-then-proceed checkpoints between phases are implicit rather than stated, and the completeness checklist (CHECKLIST.md) that would serve as the final checkpoint is a dead reference. | 4 / 5 |
Progressive Disclosure | The body is well structured with clear section headers and a clearly signaled Reference Files list, but scored against the actual bundle, 3 of the 4 referenced files (DIAGNOSTIC_WORKFLOW.md, REPORT_TEMPLATE.md, CHECKLIST.md) do not exist — only scripts/clinical_patterns.py is present. The promised per-phase code examples and report template are unreachable, so navigation breaks where it matters. This lands at level 3 (structure present, but the reference organization fails to deliver) rather than level 2, since nothing is buried or inlined — the content is simply missing. | 3 / 5 |
Total | 15 / 20 Passed |