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.
The body delivers dense, domain-specific, executable guidance — a fixed five-segment structure, timing formulas, narration reading conventions, exact overlay parameters, and a delegation workflow capped by an explicit self-check. The weaknesses are minor: some redundancy between §4 style rules and the §5 sceneStyle template, no worked example of a segment prompt, and an implicit rather than explicit error-recovery loop.
Suggestions
Add one compact worked example of a single segment — the motion_clip creation.prompt (with LaTeX, Chinese operation label, and attached narration text), the matching audio creation.script sentence, and its subtitle overlay span — to make the central artifact unambiguous and lift actionability.
After the §5 self-check step, state the recovery loop explicitly: which Element to fix and which checks to re-run when a check fails (e.g., 字幕与旁白不一致时改 overlay 的 text 后复核), completing the workflow's feedback cycle.
Consolidate §4 with §5's sceneStyle="edu_steps" template description: since the template enforces the layout skeleton and subtitle spacing, §4 can shrink to content-only prompt requirements, removing the current overlap and tension about whether prompts should describe visual style.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Dense with non-obvious domain rules (语速约 4.5 字/秒, 单句不超过 25 字, x=0.50 y=0.88 width=0.80 height=0.14) with zero padding about concepts Claude already knows. Not 5 because §4's hand-written visual style requirements overlap with §5's sceneStyle="edu_steps" template description ("不需要描述视觉风格"), which could be consolidated; not 3 since nearly every token carries skill-specific rules. | 4 / 5 |
Actionability | Highly concrete for an instruction-only skill: element types (motion_clip/audio/overlay), field names (creation.prompt, creation.script, text), fixed五段式 durations, exact subtitle coordinates, and exact parameters captionStyle="uniform" and sceneStyle="edu_steps" with a named delegation target (ai_editing_director 到 timeline:<timelineId>). Not 5 because there is no worked example of a segment prompt or task book, and TTS/compose steps defer to unnamed "既有 compose 流程"; far above 3 since the guidance is specific and directly executable rather than pseudocode. | 4 / 5 |
Workflow Clarity | §5 lays out a clear four-step sequence (拆解规划 → 委派创建 → 确定性合成 → 自查) ending in an explicit self-check checklist (公式推导正确、标注全中文且与旁白一致、每句旁白有字幕且不重叠、样式一致、时长覆盖). Not 5 because the error-recovery loop after the check (fix which element and re-verify) is left implicit; not 3 since the sequence is complete and the validation checklist is explicit. | 4 / 5 |
Progressive Disclosure | Well-organized into five numbered sections with clean internal cross-references (第 1–2 节, 第 5 节), a timing table, and no external file references — consistent with the actual bundle, which contains no references/scripts/assets directories. Not 5 because at ~90 lines the detailed §4 style specification is inlined where a one-level-deep reference file could carry it, exceeding the under-50-line simple-skill case; not 3 since structure and navigation are good and nothing is buried. | 4 / 5 |
Total | 16 / 20 Passed |