skill存档

sceneAssets Data Structure

Use sceneAssets to separate production assets from scene structure.

Core Rule

sceneSpecs = what should appear
sceneAssets = how the thing is produced, found, generated, replaced, or verified

This keeps sceneSpecs stable and lets visual assets, prompts, audio, subtitles, and generated files evolve without rewriting the scene plan.

Suggested Types

export type SceneAssetKind =
  | 'placeholder'
  | 'svg'
  | 'image'
  | 'video'
  | 'audio'
  | 'subtitle'
  | 'prompt'
  | 'style-reference';

export type SceneAssetStatus =
  | 'requested'
  | 'planned'
  | 'generated'
  | 'linked'
  | 'verified'
  | 'rejected'
  | 'error';

export type SceneAsset = {
  id: string;
  sceneId: string;
  targetId?: string;
  kind: SceneAssetKind;
  status: SceneAssetStatus;
  role:
    | 'subject'
    | 'background'
    | 'text'
    | 'transition'
    | 'narration'
    | 'music'
    | 'sfx'
    | 'caption'
    | 'reference';
  description: string;
  sourceSkill?: string;
  prompt?: string;
  filePath?: string;
  durationInFrames?: number;
  timeRange?: {
    startFrame: number;
    endFrame: number;
  };
  styleTags?: string[];
  constraints?: string[];
  metadata?: Record<string, string | number | boolean>;
  error?: string;
};

Example

const sceneAssets: SceneAsset[] = [
  {
    id: 'asset_scene_06_index_card_prompt',
    sceneId: 'scene_06_skill_index',
    targetId: 'index',
    kind: 'prompt',
    status: 'planned',
    role: 'subject',
    description: 'SKILL_INDEX card visual prompt for flat-vector library metaphor.',
    sourceSkill: 'aigc-prompt-optimizer',
    prompt: 'flat-vector index card labeled SKILL_INDEX, clean educational explainer style',
    styleTags: ['flat-vector', 'library', 'vivid-controlled'],
  },
  {
    id: 'asset_global_bgm_brief',
    sceneId: 'global',
    kind: 'audio',
    status: 'requested',
    role: 'music',
    description: '90-second friendly intelligent explainer BGM brief.',
    sourceSkill: 'suno-music-brief',
  },
];

Binding To sceneSpecs

• sceneId must match a scene id in sceneSpecs, except global for whole-video assets.
• targetId must match subjects[].id when role is subject.
• Audio and subtitle assets may omit targetId.
• filePath appears only after an asset is linked/generated.
• prompt can exist before any file exists.

Current RemotionSkill DryRun Binding

E:\knowledge-base\07_skill存档\remotion\src\skills\RemotionSkill.ts now creates planned placeholder assets for:

• each scene background;
• each subjects[].id as a subject-bound asset;
• simulated outputs from requested existing Skills, such as prompt assets from aigc-prompt-optimizer or a global editing-plan asset from blind-editing-workflow.

DryRun binding confirms structure only. It does not render video, execute ffmpeg, generate audio, or link production files.

Production States

Use the smallest honest status:

• requested: the need is known, but no plan exists.
• planned: prompt or production method exists.
• generated: a file or concrete output exists.
• linked: Remotion project references the file.
• verified: preview/render check passed.
• rejected: human review rejected it.
• error: generation/linking/verification failed.