The body is mostly operational and avoids explaining basic concepts Claude already knows, but it repeats the four agent roles twice (in the 操作步骤 workflow and again in the 四大智能体角色职责 section) and the 使用示例 section restates the workflow, so it could be tightened. It is not 3 due to this redundancy; it is not 1 because it is not padded with generic concept explanations.

Concrete parameter-level guidance is present (1080×1920, 9:16, 5s/幕精确到秒, 色温4000K-5500K, 16kHz单声道wav) and each act's required components are itemized, but the actual prompt templates, character fixed-feature text, and scene templates live in the referenced bundle files rather than inline, leaving key generative detail external. It is not 3 because copy-paste-ready prompt scaffolding is not in the body; it is not 2-below because structural and parameter guidance is specific and actionable.

The process is explicitly sequenced (提示词生成 → 质量核查 → 知识库归档 → 跨工具集成) with an explicit validation checkpoint: '通过（pass）…不通过（fail）：生成问题清单，反馈给提示词生成师调整（最多2次）' is a validate→fix→retry feedback loop. It is not 2 because validation and the retry gate are explicit, not implicit.

The body is an overview with a dedicated 资源索引 section giving '何时读取' navigation cues for each bundle file, all six referenced paths (5 in references/, 1 in assets/examples/) exist, and references are one level deep. It is not 2 because navigation is clearly signaled and files are real and well-split; not below because there is no deep nesting.

The description enumerates multiple concrete components and parameters — '角色固定特征+动作时序+场景环境+音频匹配+光影氛围+技术约束', '9:16竖屏、5s/幕', 'Suno+chinese-wav2vec2-base', '一致性管控（角色/视觉/情绪）' — matching the anchor for listing several specific concrete actions. It is not score 2 because it goes well beyond naming only a domain and some actions; the capabilities are itemized with explicit technical detail.

It clearly answers 'what' (generates structured prompts for the shopping-avatar livestream video and integrates with model inference), but provides no explicit 'Use when...' clause or equivalent trigger guidance, so 'when' is only implied. Per the judging guideline, a missing explicit trigger clause caps completeness at 2; it is not 3 for this reason and not 1 because the 'what' is thoroughly stated.

Some natural user terms appear ('导购员数字人带货提示词生成', '带货视频'), but the description is dominated by technical jargon users would not naturally say ('Image-to-Video模式', 'chinese-wav2vec2-base', 'TeaCache', 'int8量化'), missing common conversational variations. It is not 3 because coverage of plain natural phrasing is thin and buried under project-internal jargon; it is not 1 because relevant domain keywords are present.

The niche is sharply defined to one project and one persona ('InfiniteTalk项目小省导购员数字人带货提示词生成'), with triggers unlikely to overlap with other skills. It is not 2 because the specificity is far beyond 'somewhat specific but could overlap'; the project- and character-bound scope makes accidental triggering very unlikely.