Conciseness

At ~35 lines, every section earns its place: a one-line mission statement ("专治空洞拔高、排比三连、官方黑话、模板化结尾" — useful domain knowledge, not padding), trigger conditions, two lean command blocks, and a one-line zero-dependency note. It matches 'Lean and efficient; assumes Claude's competence; every token earns its place'; nothing explains concepts Claude already knows.

Actionability

Both usage paths are copy-paste-ready and verified against the bundled script: `export LLM_API_KEY=…`, `python3 scripts/deai.py 稿子.txt -o 改好.txt`, and `--prompt-only` all match deai.py's argparse/config. It matches 'Mostly executable guidance; concrete code or commands with minor gaps' — the trigger section promises pasted-text input ("粘贴或文件") but the steps only show file input (the script's `-t`/stdin support and the temp-file workaround are never mentioned). Not 5 because the common pasted-text case lacks an executable example.

Workflow Clarity

Two clearly sequenced alternative workflows (API-key mode marked 推荐, then the zero-cost `--prompt-only` fallback) each have an unambiguous order: set env → run → get output. It matches 'Clear sequence with most checkpoints present; minor validation gaps' — there is a gap between the "粘贴或文件" trigger and the file-only steps, and no note on what happens on failure (e.g. missing key), even though the script itself handles it. This is not a destructive or batch operation, so no cap applies; not 5 because of the paste-path gap and absent error/fallback checkpoint.

Progressive Disclosure

Under 50 lines with no need for external references, and the single referenced path (`scripts/deai.py`) exists in the bundle. Sections (触发条件 / 使用步骤 / 零依赖) are well organized and easy to navigate, matching the rubric's simple-skill exception: 'under 50 lines, no need for external references, progressive disclosure can score 5 with just well-organized sections.'

Specificity

The description names the domain ("AI 生成的文本（公众号/小红书/电商文案等）" / "AI-generated copy") and one core action ("rewrite … into natural, human-sounding Chinese"), but the capability coverage is essentially a single verb with no distinct sub-actions. It matches the anchor 'Names domain and 1-2 concrete actions, but not comprehensive'; not 4 because it does not list several specific actions, and not 2 because the domain and action are concrete rather than generic.

Completeness

Both questions are answered explicitly: what — "输出自然有人味的中文。De-AI writer: rewrite AI-generated copy … into natural, human-sounding Chinese text"; when — the explicit use-when clause "觉得'太 AI 味'要改得更自然、更像人写的时使用" with concrete trigger phrasing and input types. This matches anchor 5; it is above anchor 4 because the 'when' is not merely present but specific and concrete.

Trigger Term Quality

Strong natural trigger terms a Chinese user would actually say — "觉得'太 AI 味'"、"要改得更自然、更像人写的"、"公众号/小红书/电商文案" — plus an English mirror ("AI-generated copy", "human-sounding Chinese"). It matches 'Good keyword coverage; a few natural terms missing': common variants like "润色" (polish), "AI 感/机器味", or English triggers ("make this sound human", "too robotic") are absent, so it falls short of the comprehensive anchor 5.

Distinctiveness Conflict Risk

The "去 AI 味/太 AI 味" niche trigger is distinctive and bilingual, minimizing conflict with unrelated skills. It matches 'Mostly distinct; minor overlap risk' — the broader "改得更自然/rewrite copy" phrasing could also fire for generic Chinese copy-polishing/editing skills. Not 5 because generic polishing requests could plausibly match both this and a general editing skill.