Conciseness

The tool parameter blocks and Common Mistakes are genuinely non-obvious and earn their tokens, but the body also explains things Claude already knows: "H3K27ac -- Active enhancers and promoters", "H3K27me3 -- Polycomb-repressed regions", and an opening paragraph reiterating that direct binding evidence outweighs co-expression. Fits the 'mostly efficient but includes some unnecessary explanation or could be tightened' anchor rather than 4, where over-explanation would be only minor.

Actionability

Concrete tool names, exact parameter lists, copy-paste JSON call examples, and documented return shapes for the core tools (JASPAR, Enrichr, GTEx, STRING), plus a 'COMPUTE, DON'T DESCRIBE' mandate. Not 5 because several tools (RegulomeDB, IntAct, BioGRID, STRING_functional_enrichment, OLS, PubMed) have parameter tables but no call example or return-shape example.

Workflow Clarity

Phases 0-7 give a clear interrogable sequence and five named use patterns map question shapes to tool chains; the 'LOOK UP DON'T GUESS' rule and the Common Mistakes list act as checkpoints. It falls short of 5 because there are no explicit result-validation steps (e.g., verifying a JASPAR matrix ID before use is stated, but no check that returned enrichment terms are meaningful), though for read-only retrieval this is a minor gap, not the score-3 cap scenario.

Progressive Disclosure

Section headers are clean and navigable, but there are no bundle files at all — ~200 lines of per-tool API reference (parameters, returns, and the mistakes list) live inline in SKILL.md where the rubric's ideal is a lean overview pointing one level deep into references/. This is the 'content that should be separate is inline' anchor; it avoids a 2 only because the inline content is well-structured with clear headers rather than a wall of text.

Specificity

Quotes several concrete actions — "TF-target inference (JASPAR motifs, ChIP-seq)", "motif scanning", "eQTL integration", "perturbation evidence (knockout/overexpression)" — each naming the underlying data source. It stops short of 5 because actions the skill also supports (protein interaction networks, literature context) are omitted, a minor coverage gap.

Completeness

The what is explicit and concrete ("TF-target inference... motif scanning, eQTL integration, perturbation evidence") and the when is an explicit "Use for" clause with concrete trigger phrases, matching the score-5 anchor exactly. It would need to lose the 'Use for' clause to drop to 4.

Trigger Term Quality

Natural user phrasings are present verbatim: "'which TF regulates gene X'", "'which genes does TF Y target'", "regulatory pathway reconstruction", alongside "gene regulatory network analysis", "ChIP-seq", "eQTL". Not 5 because common variants like "transcription factor binding sites", "gene regulation", or spelled-out "transcription factor" phrasing are missing, and "TF-target inference" leans technical.

Distinctiveness Conflict Risk

A clear niche (gene regulatory networks) with domain-specific triggers — JASPAR, ChIP-seq, eQTL, TF-target — that a user would only say when they want this exact skill; minimal overlap risk with general bioinformatics or protein-interaction skills.