Conciseness

The body is dense and skill-specific — no generic concept explanations, and the heavy pattern data is correctly left in scripts/patterns.js. It stops short of anchor 5 because the "AI-engineering evidence lens" bullets partially restate cautions already covered in the Authority section, and three near-identical bash examples (stdin, file, rendered-markdown) could be compressed to two. It is well above anchor 3 since nothing explains material Claude already knows.

Actionability

The Preferred path gives copy-paste-ready commands ("printf '%s' \"$TEXT\" | node scripts/detect.js --context general", "--file path/to/draft.md --source-mode rendered-markdown") whose flags and values match the actual bundled scripts/detect.js interface, covers the common input cases, specifies decision criteria for --context and --source-mode, and defines the fallback when Node is unavailable. This matches anchor 5: fully executable guidance covering the common cases.

Workflow Clarity

The Preferred path is a clear numbered sequence with condition points, an explicit fallback branch, an execution-integrity checkpoint ("Never claim execution unless the command actually ran", enforced by execution_evidence executed vs model_only), and explicit Stop conditions. It misses anchor 5 because there is no guidance for handling command failure (detect.js exits 2 on usage or I/O errors), leaving a minor validation gap — though the operation is read-only, so the destructive/batch cap does not apply.

Progressive Disclosure

Structure is good and matches the actual bundle: the rulebook detail is delegated to ../avoid-ai-writing/SKILL.md, graph policy to the router references, and pattern mass to scripts/, with one-level-deep, clearly named paths. It falls short of anchor 5 because references are bare code paths rather than navigable links, and the lens bullets plus several contract details are inlined in SKILL.md where anchor 5's example would signal them as short pointer references.

Specificity

The description lists several concrete actions ("detect, scan, audit, score, or flag AI-writing patterns") plus a scope boundary ("without rewriting the text") and a mechanism condition ("deterministic local detector result when the host can execute Node"). It sits at anchor 4 rather than 5 because the verbs are near-synonyms around one detection capability and it never states what output the skill produces (score, label, findings by severity).

Completeness

The "when" is explicit and detailed ("Use when the user asks to detect, scan, audit, score, or flag...") and a "what" is implied by those same verbs, but there is no standalone third-person capability statement (e.g., "Runs a detect-only analysis and reports score, label, and findings"). That fits anchor 4 — both present, one could be more explicit — rather than 5, and it is clearly above anchor 3 where the "when" would be missing or weakly implied.

Trigger Term Quality

Good keyword coverage with five natural verbs users would say ("detect, scan, audit, score, or flag") plus the domain phrase "AI-writing patterns". It misses anchor 5 because common noun-phrase variations such as "AI-generated text", "machine-written", or "ChatGPT-style writing" are absent.

Distinctiveness Conflict Risk

"without rewriting the text" carves out a clear detect-only niche distinct from rewrite-oriented skills, but the trigger verbs would equally fire a broader sibling skill (e.g., a general avoid-ai-writing skill that includes rewriting). Mostly distinct with minor overlap risk against closely related skills is exactly anchor 4; it does not reach anchor 5's "minimal conflict risk" given that sibling overlap.