Content
78%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 highly actionable — exact tool signatures, thresholds, an explicit phase workflow, a combination algorithm, an output template, and worked patterns — and reasonably concise for the domain. Its main weakness is structure: everything lives inline in one long file with no one-level-deep references, and tool-failure recovery loops are largely unspecified.
Suggestions
Move the 17-row Tool Parameter Reference table to references/tool-reference.md and keep a 3-4 line quick-reference (validation, frequency, ClinVar lookup) inline so SKILL.md stays an overview.
Move the four worked Common Patterns to references/patterns.md, linked one level deep from the body, reserving SKILL.md for the phase workflow and classification algorithm.
Add explicit error-recovery feedback loops for the tool chain: what to do when VariantValidator or ClinVar returns an error or conflicting transcripts, and a re-validation checkpoint after Phase 0 before any criterion is assessed.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and domain-specific with concrete thresholds and little concept-teaching, but a few passages could be trimmed — e.g. 'Population AF is among the strongest evidence in either direction' and the tangential ESM mechanism-complement paragraph in Phase 2. This fits 'efficient; minor instances of over-explanation' rather than the fully lean top anchor. | 4 / 5 |
Actionability | Fully executable guidance throughout: tool calls with exact parameter names and real example values (e.g. VariantValidator_validate_variant with NM_000059.4:c.5946delT, GRCh38), concrete numeric thresholds (REVEL >= 0.7, mis_z > 3.09, pLI >= 0.9, SpliceAI >= 0.5), a complete output template, and four worked patterns using real variants. Not 4: no gaps — the common cases are covered copy-paste ready. | 5 / 5 |
Workflow Clarity | A clear Phase 0-6 sequence with an explicit validation phase up front ('Wrong HGVS or wrong transcript cascades errors... Validate first') and a complete combination-rule algorithm, but error-recovery feedback loops are thin — only one fallback is written ('If gnomAD data is unavailable, note the gap and continue') and there is no guidance for VariantValidator/ClinVar failures or transcript conflicts. Fits 'clear sequence with most checkpoints present; minor validation gaps' rather than the feedback-loop-rich top anchor. | 4 / 5 |
Progressive Disclosure | The body is a ~240-line monolith with clear section headers but no bundle files at all: the 17-row Tool Parameter Reference table and the four worked Common Patterns are reference-style material inlined in SKILL.md rather than split one level deep. Fits 'some structure but could be better organized; content that should be separate is inline' rather than the 'minor organization gaps' anchor above. | 3 / 5 |
Total | 16 / 20 Passed |