Conciseness

The body is dense and free of padding or explanations of concepts Claude already knows, but the detect-only/no-rewrite constraint is repeated across the intro ('Never rewrite unless the user changes the request'), the produce/outgoing rules, 'Stop conditions', and 'Output' ('no rewritten version unless control has explicitly passed to a rewrite owner'), which could be consolidated.

Actionability

The body gives copy-paste-ready commands for the common cases (stdin via "printf '%s' \"$TEXT\" | node scripts/detect.js --context general", file input, and --source-mode rendered-markdown for frontmatter-bearing Markdown), with concrete flag values that match the actual scripts/detect.js contract, plus an explicit fallback when Node is unavailable.

Workflow Clarity

A clear 6-step numbered 'Preferred path' with explicit stop conditions and a no-Node fallback sequences the work well, but there is no error-recovery guidance for command failure (detect.js exits 2 on usage or I/O errors, yet the body never says what to do on a non-zero exit). No destructive/batch cap applies since the workflow is read-only.

Progressive Disclosure

The ~94-line body is a well-sectioned overview that correctly pushes the 180KB pattern engine into scripts/ and references the real, one-level-deep scripts/detect.js. However, four referenced paths (../avoid-ai-writing/SKILL.md and the three ../avoid-ai-writing-router/references/ files) point outside the shipped bundle and cannot be navigated from what is present here.

Specificity

'detect, scan, audit, score, or flag AI-writing patterns without rewriting the text' plus 'a deterministic local detector result when the host can execute Node' names several specific actions and a concrete execution capability, but stops short of stating what the output actually contains. Not 3 (more than 1-2 actions with a concrete boundary); not 5 (coverage of the produced result is a gap).

Completeness

An explicit 'Use when the user asks to detect, scan, audit, score, or flag AI-writing patterns' trigger clause answers 'when' concretely, and the verb list plus the Node-execution clause conveys 'what'. Not 5 because the 'what' is embedded in the when-clause rather than stated as a standalone capability sentence.

Trigger Term Quality

'detect, scan, audit, score, or flag AI-writing patterns' gives good natural keyword coverage users would actually say. A few natural synonyms are missing, such as 'AI-generated text', 'machine-generated', or tool names like 'ChatGPT writing'.

Distinctiveness Conflict Risk

'AI-writing patterns without rewriting the text' carves a clear detect-only niche and explicitly differentiates the skill from rewriting/mutation skills, so the risk of triggering the wrong sibling skill is minimal.