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 well-structured, information-dense workflow document: clearly sequenced phases, disease-type routing, exact tool parameters, and a quantitative scoring framework. Its two real weaknesses are missing validation/feedback checkpoints in the workflow and — most seriously — five reference files cited as living in the same directory that are absent from the bundle, so the progressive-disclosure structure is promised but not delivered.
Suggestions
Ship the five referenced files (TOOLS_REFERENCE.md, SCORING_REFERENCE.md, REPORT_TEMPLATE.md, EXAMPLES.md, QUICK_START.md) in the bundle, or remove/inline the pointers — currently every reference is a dead link.
Add explicit validation checkpoints to the workflow, e.g., after each tool call verify the expected response shape (OpenTargets nested data wrapper, PubMed plain list) and a fallback action on failure, mirroring the gnomAD 'skip gracefully' pattern.
Show one minimal executable example (a sample tool invocation or the Python analysis pattern demanded by 'COMPUTE, DON'T DESCRIBE') so the guidance moves from tool names to copy-paste-ready usage.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious, high-value specifics — tool parameter gotchas ('MyGene_query_genes: param is query (NOT q)', 'DrugBank tools: ALL require 4 params'), phase-specific tool lists, and exact enrichment library names — with almost no explanation of concepts Claude already knows. It falls short of anchor 5 only through minor redundancy (e.g., the repeated 'Scoring: See SCORING_REFERENCE.md' lines and the ~30-line 'Common Use Patterns' section that partially duplicates the reference-file pointers). | 4 / 5 |
Actionability | Concrete, executable guidance dominates: exact tool names per phase, exact parameter corrections ('variant_id (NOT rsid)', 'species=homo_sapiens'), specific libraries ('KEGG_2021_Human', 'Reactome_2022'), numeric scoring bands, and a tier table mapping scores to management. It stops short of anchor 5 because no actual invocation syntax or code snippet is shown for any tool — the 'COMPUTE, DON'T DESCRIBE' directive says to write Python but gives no example — leaving a gap between tool name and copy-paste-ready call. | 4 / 5 |
Workflow Clarity | The 9-phase sequence is clearly ordered with explicit routing logic ('Route to the correct Phase 3 path BEFORE running Phase 2 tools'), a disease-type classification table driving Phase 3 branches, and graceful failure handling ('gnomAD: May return Service overloaded - skip gracefully'). It matches anchor 4 rather than 5 because there are no explicit validate-then-proceed checkpoints or feedback loops — e.g., no step verifying that tool responses parsed correctly or what to do when an expected field is missing (the OpenTargets/gnomAD notes hint at this but are not woven into the workflow). | 4 / 5 |
Progressive Disclosure | The structure is well-signaled and one level deep — a dedicated 'Reference files (same directory)' section listing TOOLS_REFERENCE.md, SCORING_REFERENCE.md, REPORT_TEMPLATE.md, EXAMPLES.md, and QUICK_START.md — but none of these files exist in the bundle (no references/, scripts/, or assets/ directories; no such files anywhere in the skill directory). Per the guideline to score against the actual bundle structure, the references are dangling: navigation promised by the body is unresolvable, which drops it below anchor 4's 'most content is appropriately placed; references mostly clear'. | 3 / 5 |
Total | 15 / 20 Passed |