Names the domain (prose/writing) and some actions ('drafting, editing, or reviewing text'), but the core capabilities are described in abstract, jargon-heavy language ('voice basins', 'retrieval-shaped positive constraints', 'diagnostic taxonomy') rather than concrete, understandable actions.

Explicitly answers both what ('Produce clean, human-register prose by routing into specific voice basins') and when ('Use when drafting, editing, or reviewing text for AI patterns, or when matching a writer's voice from a sample'). The 'Use when' clause is present and provides clear trigger guidance.

Includes some useful trigger terms like 'drafting', 'editing', 'reviewing text', 'AI patterns', and 'writer's voice', but relies heavily on specialized jargon ('voice basins', 'anti-slop', 'retrieval-shaped positive constraints') that users would rarely use naturally. Missing common terms like 'writing style', 'tone', 'rewrite', 'sound more human'.

The focus on AI-pattern detection and voice matching gives it some distinctiveness, but 'drafting, editing, or reviewing text' is broad enough to overlap with general writing or editing skills. The specialized jargon helps differentiate but may not be recognized by the selection mechanism in practice.

The skill is well-written but verbose for its purpose. The 'Architecture Principle' section explains negative prompting theory Claude already understands, the 'Source Basis' section is attribution that doesn't help Claude execute the skill, and the 'Voice Within the Register' section could be tightened. However, the core constraints and routing checks are efficiently stated.

The six core routing constraints each include concrete ✓/✗ examples showing exactly what passes and fails. The derivation fields are specific and enumerated, the quick routing checks are concrete presence tests, and the Final Audit Pass gives a precise two-step procedure. The scoring rubric provides a clear numeric threshold for re-derivation.

The workflow is clearly sequenced: (1) complete derivation fields before writing, (2) apply six simultaneous constraints during generation, (3) run quick routing checks before delivery, (4) optionally run the Final Audit Pass on finished drafts. The audit pass includes an explicit feedback loop (step 1 diagnoses, step 2 fixes). The separation of generation context from diagnostic context is a critical validation checkpoint that is clearly enforced.

The skill provides a clear overview with well-signaled one-level-deep references to four separate files (derivation.md, routing-patterns.md, examples.md, diagnostic-taxonomy.md). Each reference includes a brief description of its contents and when to use it. The diagnostic taxonomy is explicitly scoped to editorial review only, showing thoughtful content separation.