The body is lean (~45 lines) with no padding or explanations of concepts Claude already knows; every line (format spec, tags, examples, boundaries) earns its place. Not below 3 because there is no verbosity or unnecessary context.

Provides a concrete output format ("L<line>: <tag> <what>. <replacement>."), five defined tags with replacement guidance, and five copy-pattern examples. Per the code-vs-instruction note, absence of executable code is not penalized for an instruction-only skill with this concrete, specific guidance.

A simple single-task skill (review → one line per finding → end with "net: -<N> lines possible" or "Lean already. Ship.") with an unambiguous action and explicit terminal condition. The simple-skills carve-out allows 3 without a multi-step workflow; no destructive/batch validation is needed since it only lists findings.

Under 50 lines with no need for external references and no bundle files present (references/, scripts/, assets/ absent), yet organized into clear sections (Format, Examples, Scoring, Boundaries). The under-50-lines carve-out permits 3 for well-organized sections with no external references.

Lists multiple concrete actions/targets: "Finds what to delete: reinvented stdlib, needless deps, speculative abstractions" plus "One line per finding" — specific, not vague. Not below 3 since it names several concrete capabilities rather than a single domain.

Clearly states what it does ("Review a diff for over-engineering. Finds what to delete...") but has no explicit "Use when..." trigger, so the "when" is only implied. Per the rubric, a missing explicit trigger clause caps completeness at 2.

Contains natural terms like "diff" and "over-engineering" a user might say, but coverage is narrow with no common variations. Not 3 because it lacks broad natural-term coverage; not 1 because the keywords present are genuinely natural rather than jargon.

The niche is distinct — over-engineering review focused on finding what to cut — with triggers unlikely to fire for unrelated skills. Not 2 because the framing ("over-engineering", "reinvented stdlib", "speculative abstractions") carves a clear, narrow scope.