Content
71%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 is highly actionable and well-sequenced, encoding proprietary format and Seedance-specific constraints Claude would not otherwise know. Its main weaknesses are token efficiency (duplicated full examples) and a monolithic structure that would benefit from offloading examples and the dialogue-tag spec to reference files.
Suggestions
Collapse Example A and Example B into one full worked example plus a short contrast note showing only the English/Chinese language-handling differences, cutting roughly half the example token volume.
Move the detailed rule-11 dialogue-tag taxonomy (forms a/b/c, emotion labels, multi-speaker chaining) and/or the full worked examples into a references/ file (e.g. DIALOGUE_TAGS.md, EXAMPLES.md) and link from SKILL.md, improving progressive disclosure.
Add an explicit pre-emit validation checklist ('verify sum(SHOT.DURATION_S)==OVERVIEW.DURATION_S ±2s; verify each IMAGE_PROMPT/VIDEO_PROMPT starts with byte-identical IDENTITY_ANCHOR; verify --ar and aspect_ratio tokens present') to turn implicit rules into a feedback checkpoint and raise workflow clarity.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The rules encode genuinely novel downstream-constraint knowledge (strict format, Seedance dialogue-tag mechanics) and are mostly tight, but two near-identical full 5-shot examples (Example A Chinese + Example B English) plus model-behavior commentary in rule 8 ('Seedance (ByteDance) is in fact a Chinese-first model...') are padded and could be trimmed — matching 'mostly efficient but includes some unnecessary explanation'. Not a 4 because the duplicated full examples are a clear, sizable waste. | 3 / 5 |
Actionability | Fully executable: exact field names, literal required tokens ('--ar 9:16', 'aspect_ratio: 9:16'), explicit dialogue-tag formats ('Dialogue (CharacterName, emotion): "..."'), char budgets (≤500/≤220), sentinel values for missing inputs, and complete worked examples covering both Chinese and English cases — matching the score-5 anchor. | 5 / 5 |
Workflow Clarity | Numbered rules 1-12 give a clear sequenced contract with a strict OVERVIEW→SHOT_K output structure and explicit constraints (duration math, byte-identical identity anchor), but there is no explicit 'validate your output before emitting' feedback loop/checklist step framing those constraints as checkpoints — matching 'clear sequence with most checkpoints present; minor validation gaps'. Not capped at 3 because the skill emits text rather than performing destructive/batch operations. | 4 / 5 |
Progressive Disclosure | The body is well-sectioned (Inputs, Output format, Rules, Style presets, Examples, Non-goals) but monolithic: the two bulky full worked examples and the detailed rule-11 dialogue-tag spec are inlined rather than offloaded to reference files, and no bundle files exist. This matches 'some structure but could be better organized; content that should be separate is inline'. | 3 / 5 |
Total | 15 / 20 Passed |