The body is lean and assumes Claude's competence, covering only what a technical editor needs ('What you're hunting for', 'Confidence calibration', 'What you don't flag') with no padding or basic-concept explanations.

Concrete, executable guidance throughout: named finding categories with worked examples, numeric confidence thresholds (0.80+, 0.60-0.79, below 0.50 suppress), and explicit output directives like 'Flag with autofix_class: auto and group by logical theme, keeping original R# IDs'.

It is a single unambiguous review task, which the simple-skill note allows to score 3, and the confidence-calibration block acts as an explicit gating checkpoint (suppress below 0.50) before emitting findings.

Under 50 lines with no bundle files and no need for external references, yet well-organized into clear sections, satisfying the short-skill progressive-disclosure criterion.

Lists multiple concrete review actions -- 'contradictions between sections, terminology drift, structural issues, and ambiguity where readers would diverge' -- matching the multiple-specific-actions anchor rather than the single-domain anchor at 2.

It clearly states what the skill does, but the only 'when' guidance is 'Spawned by the document-review skill', which is an implied spawn context rather than explicit trigger guidance, so the missing-'Use when' cap applies.

Relevant terms like 'contradictions', 'terminology drift', and 'internal consistency' appear, but there is no 'Use when...' clause and common user phrasings are missing, placing it at 'some relevant keywords but missing common variations' rather than full coverage.

The internal-consistency niche is sharply delineated from feasibility or security reviewers, and as a spawned sub-agent it is unlikely to be triggered for the wrong skill, matching the clear-niche anchor.