Content
81%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.
An exceptionally actionable and well-sequenced operational manual with explicit checkpoints, quantified quality gates, and strong reference signaling. Its main weaknesses are cross-section repetition of the same rules, which hurts token efficiency, and inlined engine/spec detail that could live in the reference files it already points to.
Suggestions
State each hard rule once — in 硬性原则 or the relevant reference file — and cross-reference it from the workflow stages and 质量标尺 instead of restating it; e.g. "单镜头 ≥120 帧 / 短句并入相邻镜头" and "最长静止 ≤3 s / 末拍稳定期 ≥30 帧" each currently appear in three or more places.
Move the quantified TTS engine details (kokoro ≈2.3 词/秒, edge-tts rates, the ×0.78 word-count conversion) into reference/narration-storyboard.md and keep only the operational default and pointer in the body.
The 13 narration writing principles are enumerated twice (阶段 2 parenthetical and the narration-guidance.md row of the 关键文件 table) — keep one enumeration and reduce the other to a pointer.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and almost entirely skill-specific (no explanation of concepts Claude already knows), but the same rules are restated across sections: "单镜头 ≥120 帧、短句并入相邻镜头" appears in 硬性原则 8, 阶段 3, and 质量标尺; "最长静止 ≤3 s、末拍稳定期 ≥30 帧" appears in at least four places; the narration principles are enumerated both in 阶段 2 and in the key-files table. Mostly efficient, but the cross-section repetition means it could be meaningfully tightened — fitting the level-3 anchor rather than level 4. | 3 / 5 |
Actionability | Fully executable guidance: copy-paste commands with arguments ("template/scripts/new_project.sh <工作目录> <slug>", "python3 scripts/tts_build.py", "scripts/preview.sh 30", "VER=v1 scripts/render.sh", "python3 scripts/motion_check.py <Gn>"), quantified thresholds (主角 ≥170px、字幕 ≤16 字、单镜头 ≥120 帧), and a duration/word-count table mapping user choices to concrete budgets. This matches the copy-paste-ready level-5 anchor. | 5 / 5 |
Workflow Clarity | Eight sequenced stages with time estimates, four explicit stop-and-wait confirmation checkpoints ("必须停下来等用户回话,不要自己往下走"), and validation with feedback loops throughout ("差 >15% 就加/删句子重跑", add "## gap 15–30" and re-run, QC → 修复 → v2 → 复验 → 小修 → v3, final 成片复测). This matches the level-5 anchor: explicit validation steps, feedback loops, and checklists (质量标尺). | 5 / 5 |
Progressive Disclosure | Good structure: a 关键文件 table mapping every referenced path to its role, per-section pointers ("细则 reference/composition-and-light.md §7", "见 reference/narration-storyboard.md §2.5"), and references kept one level deep. However, the referenced bundle directories (reference/, template/, examples/) are not present alongside SKILL.md so the file structure cannot be verified, and some granular detail (TTS engine word rates, ×0.78 conversion, navigation-label budgets) is inlined in the body rather than deferred — minor organization gaps fitting level 4 rather than the clean overview of level 5. | 4 / 5 |
Total | 17 / 20 Passed |