Content
75%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.
A well-structured, highly actionable skill body with executable examples and a clear async workflow backed by real bundle scripts. The main weakness is redundancy between the example and scenario sections and repeated parameter detail, which hurts token efficiency.
Suggestions
Consolidate '使用示例' and '常见使用场景' into a single section to remove overlapping bash command examples and reduce token cost.
De-duplicate ratio/duration guidance that appears in both '参数说明' and '高级选项', keeping the table as the canonical source and linking from the advanced section.
Consider moving the full model-by-model '模型选择' reference table into a references/ file, keeping only the recommendation summary inline in SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient and free of basic-concept padding, but the 5 '使用示例' and 4 '常见使用场景' sections overlap heavily, and '高级选项' repeats ratio/duration info already in '参数说明', so several sections could be tightened. | 3 / 5 |
Actionability | Fully executable copy-paste bash commands with real script paths ($SKILLS_ROOT/seedance/scripts/generate-video.sh), concrete model IDs, and working flags cover all common cases (T2V, I2V, first/last frame, multi-ref, audio). | 5 / 5 |
Workflow Clarity | The async workflow is clearly sequenced (submit → poll until 'succeeded' → download) with a status table and a 24-hour retention warning, though there is no explicit post-download verification checkpoint. | 4 / 5 |
Progressive Disclosure | Good structure with clearly signaled one-level-deep references to real bundle scripts (generate-video.sh/.ps1/generate_video.js) that hold the implementation; the body itself is long, so some reference material could be split out. | 4 / 5 |
Total | 16 / 20 Passed |