From short-video-director
Plans all keyframes and asset inventory for a story episode. Reads outline, novel, config, and existing assets, then produces keyframes.json and updates outline.md with asset list. Useful for narrative production workflow.
How this skill is triggered — by the user, by Claude, or both
Slash command
/short-video-director:director-keyframessonnetdirectorThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
- `story/episodes/$ARGUMENTS[0]/outline.md` — 必须读取(剧情骨架/信息传达/集尾钩子)
story/episodes/$ARGUMENTS[0]/outline.md — 必须读取(剧情骨架/信息传达/集尾钩子)story/episodes/$ARGUMENTS[0]/novel.md — 必须读取(视觉细节、动作、情绪、场景的来源)config.md — 必须读取(语言/风格/每集时长目标/每集分镜数)assets/**/*.md — 必须 Glob 列出所有现有资产文件名(决定复用 vs 新增)story/arc.md — 若存在则读取(跨集视觉/情绪连贯)story/episodes/ep{N-1}/keyframes.json — 续集模式必须读取最后一张关键帧(衔接校验)story/episodes/$ARGUMENTS[0]/keyframes.json — 仅在增量模式存在时读取story/episodes/$ARGUMENTS[0]/.review-keyframes-narrative.md — 仅增量模式必须读取(含本轮 review 意见)skills/director-keyframes/rules.md — 必须读取并严格遵循(schema 字段、composition 规则、数量规则、增量模式工作流)bash scripts/read-config.sh "每集分镜数" — 读取本集最大分镜数 N,用于关键帧数量校验$ARGUMENTS[0] — 当前集数(如 ep01)$ARGUMENTS[1] — 模式(可选,full 或 incremental,缺省为 full)把本集小说原文转化为剧情视觉节点序列:枚举本集所有不可省略的画面节点,为每个节点写完整描述,登记本集所有资产引用。不切分 shot、不规划镜头语言——shot 切分与运镜是 storyboarder 的职责。下游消费者是 director-review-keyframes-narrative(叙事评审)、creator-create-assets(出资产卡)、creator-keyframe-images(出关键帧图)、storyboarder-storyboard(消费关键帧并切分 shot)。关键帧不是"挑几张代表画面",而是剧情节奏的视觉锚点——评审者凭关键帧序列就能判断本集剧情完整性、因果性、节奏张弛;下游图像/视频生成依赖关键帧作为参考图保证一致性。
<资产名> 包裹所有出场角色/场景/道具,便于下游扫描资产引用.review-keyframes-narrative.md(定位最后一个 ## 第 N 轮 heading,用 grep ^## 第 [0-9]+ 轮 找最大 N 段,用该段内的意见列表作为变更依据),判断哪些 keyframes 需要修改/新增/删除,其他保持原样;变更后受影响的关键帧 id 不能被回收复用——新增帧必须用新 id(避免下游重抽误判)<xxx> 都要校验现有资产或加入新增清单skills/director-keyframes/rules.md — 必须读取并严格遵循story/episodes/$ARGUMENTS[0]/keyframes.json(全量模式)story/episodes/$ARGUMENTS[0]/keyframes.json(增量模式)story/episodes/$ARGUMENTS[0]/outline.md 末尾追加(或更新)「本集资产清单」章节npx claudepluginhub wddxh/shortvideodirectorGenerates video storyboard prompts from novel chapters, referencing keyframes and assets, with a self-check loop for dialogue density (up to 3 rounds).
Orchestrates story-to-video pipeline: breaks text into scenes, generates consistent Z-Image hero/refs + Qwen Edit frames, WAN FLF clips, ffmpeg concatenation.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.