Conciseness

The ~45-line body is lean throughout: no concept Claude already knows is explained, and every rule is operational ("Rewrite only the sentences a test fails on", "One device proves nothing", "Keep exact identifiers, paths, commands, and thresholds"). One sentence — "Note that the pattern catalog did not apply inside detect findings... and nowhere else" — is dense enough to need rereading, but that is a clarity cost, not token waste, so it does not drop below anchor 5's "every token earns its place".

Actionability

The guidance is concrete and directly executable for an instruction-only skill: seven named tests each pair a check with a remedy ("Does the sentence say what the thing does, or how it feels? Replace the feeling with the fact"), mode defaults are deterministic ("no draft means author; an imperative on a draft means edit; a question about a draft means detect"), and boundaries are explicit ("never touch code blocks, quoted text, frontmatter, link targets, identifiers"). It misses anchor 5 because no worked before/after example of a test applied to a sentence appears in the body — examples live entirely in references/patterns.md.

Workflow Clarity

The sequence is clear (determine mode → apply tests → per-mode action → return) and validation exists in three places: the "Done" line is an explicit completion check ("every fact, number, name, quote, and citation in the input survives, and nothing was added"), edit mode carries an idempotency check ("a second pass on the returned text changes nothing"), and detect output format is specified ("Name each pattern found, quote the line, give the fix in a few words"). It falls just short of anchor 5 because fact-survival is stated as an outcome rather than an explicit re-verify step inside the edit workflow, so there is no stated feedback loop of re-checking the output against the input.

Progressive Disclosure

The split is exemplary: the core method (tests, modes, register, boundaries) is inline while the bulk material — a 7.6KB pattern catalog and a terminology guide — lives in references/patterns.md and references/terminology.md, both verified to exist and to deliver what is promised. Each reference is one level deep and cited with a conditional read trigger rather than a bare pointer: "For edit and detect, and for an author passage the tests alone do not settle, read references/patterns.md" and "For wording choices, read references/terminology.md".

Specificity

Names three concrete actions ("Rewrite, check, or draft prose") with three explicit outcome criteria ("carries no AI writing tells, reads plainly on the first read, and keeps every source fact"), covering all three of the skill's modes. It falls short of anchor 5 only because the tells themselves are never made concrete — no example patterns like "seamless" or "it's not just X, it's Y" appear, leaving a minor coverage gap for a detection capability.

Completeness

Explicitly answers both questions: the "what" is "Rewrite, check, or draft prose so it carries no AI writing tells, reads plainly on the first read, and keeps every source fact", and the "when" is a fully explicit trigger clause — "Use when asked to make writing plainer or free of those tells, to check writing for them, or when drafting from supplied content" — with three concrete triggers. This matches the anchor-5 example's structure exactly; it is not a 4 because the when-clause leaves nothing implicit.

Trigger Term Quality

Includes several natural phrasings a user would actually say — "make writing plainer", "free of those tells", "check writing for them", "drafting from supplied content" — matching anchor 4's "good keyword coverage; a few natural terms missing". It misses common synonyms such as "AI-sounding", "robotic", or "sound human", so it is not a 5.

Distinctiveness Conflict Risk

The niche (prose free of AI tells) is distinct, and the description actively disambiguates from the nearest sibling skill with "Use ce-promote for channel-specific marketing copy" — an explicit carve-out that reduces conflict risk below even the anchor-5 example. Not a 4: the overlap risk with other prose skills is addressed, not merely minor.