Content
70%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body delivers dense, genuinely non-obvious domain knowledge with strong workflow sequencing and validation gates, and concrete examples throughout. Its main weaknesses are localized verbosity (generic camera-language list, repeated priority/caveat statements) and a fully monolithic structure with no progressive disclosure into reference files.
Suggestions
Move the 18-item generic camera-angle list in §3 to a brief pointer or trim it to the project-specific choices, since Claude already knows standard cinematography terms; keep only the composition-selection rules that encode this skill's defaults.
Split bulky reference material into one-level-deep bundle files (e.g., references/entity-schema.md for §5's VisualEntity field rules and references/moderation-pitfalls.md for the 审核误判 case study), leaving SKILL.md as a shorter overview with clearly signaled links.
State the '用户明确要求 > 默认推荐' priority and the '不指定渲染媒介' caveat once each in a shared 原则 section instead of repeating them in §1, §2, and §3, and add one fully assembled example prompt to make the 7-dimension structure copy-paste concrete.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Most content is non-obvious pipeline-specific knowledge (field contracts, square-grid aspect-ratio math, the 2026-08-26 moderation case), but §3's 18-item camera-angle list is standard cinematography knowledge Claude already has, and the '用户明确要求 > 默认推荐' priority and the '不指定渲染媒介' caveat each repeat three or more times across sections. | 3 / 5 |
Actionability | Guidance is concrete and executable: a 7-dimension prompt structure, exact field names (canonical_variant_id, reference_artifact_version_ids), precise grid rules ('2–4 格→2×2，5–9 格→3×3'), specific moderation word families to avoid, and worked examples (湿噜噜, 庭院水池→夕阳). It falls short of fully copy-paste-ready because no complete assembled example prompt shows the target output. | 4 / 5 |
Workflow Clarity | The 主图与衍生图制作顺序 is a numbered 1–5 workflow with explicit gating ('主图未生成…衍生 prompt 保持为空', '收到主图就绪…重新读取实际选中的 ArtifactVersion') and error-recovery loops (审核拒绝→改写措辞后重试; 空间矛盾→'先修正再供下游使用'; '实际检查过图像才可声明一致性已确认'). | 5 / 5 |
Progressive Disclosure | The body has real structure (numbered sections, subsections, a selection-rules table) but is a single ~250-line file with zero reference files; content such as the §5 entity-definition spec or the moderation case study clearly could be split into separate one-level-deep references. | 3 / 5 |
Total | 15 / 20 Passed |