Conciseness

The body is dense, directive prose with no padding or explanation of concepts Claude already knows — every section issues rules rather than background. It falls short of the 5 anchor because the editing-pass rules are stated redundantly across Workflow step 9, the Repair path, Stop conditions, and Output, which could be consolidated. It is well above the 3 anchor since there is no genuinely unnecessary explanation.

Actionability

Process guidance is concrete ("increment the editing-pass count from 0 to 1", "Change only the spans implicated by the blocking preservation errors", "Return the full text exactly once under **Final rewrite**"), but the core how-to-rewrite rules are delegated: "Rewrite text using the complete rules in `../avoid-ai-writing/SKILL.md`. Its editing contract is the authority for … mechanics", and that file is not part of this bundle. This matches the 3 anchor (some concrete guidance but incomplete, missing key details) rather than the 4 anchor, whose gaps are minor rather than the skill's central mechanics.

Workflow Clarity

A clear 10-step workflow with an explicit verification checkpoint ("Send before/after content to `preservation-verifier`") and a genuine feedback loop (FAIL → repair implicated spans → re-verify once → stop), satisfying the anchors above 3. It misses the 5 anchor because step 9 packs the entire pass-counting protocol into one convoluted compound sentence and pass-state rules are scattered across four sections, making the sequence harder to follow than the anchor's crisp checklist style.

Progressive Disclosure

Sections are well organized (Connection contract, Conditional guard, Workflow, Repair path, Voice handling, Stop conditions, Output) and references are clearly signaled, but every substantive detail file — the editing contract, handoff contract, skill graph, and role lenses — lives outside this bundle ("../avoid-ai-writing/SKILL.md", "../avoid-ai-writing-router/references/…") and no references/, scripts/, or assets/ directories exist here, so the overview-to-detail navigation cannot be exercised within the bundle. This sits between the 3 anchor (organization present but the detail layer fails) and the 4 anchor; the missing authority content is more than a minor organization gap.

Specificity

The description lists several concrete actions ("rewrite, humanize, clean up, or remove AI-isms from text") plus a specific preservation constraint list ("voice, facts, intent, structure, register, and protected material"), but the verbs are near-synonyms within a single domain and coverage has minor gaps (no mention of the verification loop or text types handled). It exceeds the 3 anchor (only 1-2 concrete actions) but does not fully match the 5 anchor's comprehensive multi-capability coverage.

Completeness

It explicitly answers both questions: what ("rewrite, humanize, clean up, or remove AI-isms from text while preserving the writer's voice, facts, intent, structure, register, and protected material") and when ("Use when the user asks to…") with concrete trigger phrases, matching the 5 anchor pattern exactly. Not the 4 anchor, since the 'when' clause is explicit and specific rather than improvable.

Trigger Term Quality

"rewrite", "humanize", "clean up", and "remove AI-isms" are terms users naturally say when needing this skill. A few common variations are missing (e.g., "make it sound human/natural", "less robotic", "make it sound like me"), which keeps it below the 5 anchor's comprehensive synonym coverage but clearly above the 3 anchor's partial keyword set.

Distinctiveness Conflict Risk

"remove AI-isms" and "humanize" give the skill a clear niche with minimal conflict risk, but the broad verbs "rewrite" and "clean up" overlap with generic editing/proofreading skills, matching the 4 anchor (mostly distinct, minor overlap risk). It is more distinct than the 3 anchor's 'somewhat specific' and not the 5 anchor because the generic rewrite trigger could fire for ordinary editing requests.