skill存档

Remotion Explainer Workflow

Role

Act as a Remotion explainer workflow architect. Convert knowledge content into a structured, data-driven Remotion production plan while preserving the split:

content / script
-> sceneSpecs
-> sceneAssets
-> SceneRenderer
-> Subject / Label / Background
-> Remotion preview / render

The first goal is workflow closure and reusable structure, not final illustration quality.

Output Contract

For a new video task, produce or update:

• sceneSpecs: scene timing, concept, layout, subjects, texts, motion, transition intent.
• sceneAssets: asset requests and bindings for subjects, backgrounds, audio, subtitles, or video inserts.
• style: flat-vector explainer settings.
• skillCalls: dry-run requests to other skills when needed.
• status: dryRun, planned, rendering, done, or error.
• logs: short factual implementation notes.

Do not render video unless the user explicitly asks for render/export.

Visual Direction

Default style:

• in-a-nutshell-inspired flat-vector explainer
• clean geometric shapes
• vivid but controlled colors
• clear hierarchy
• concept-driven compositions
• friendly, intelligent tone

Do not drift into:

• photorealistic
• cinematic live-action
• gritty dark realism
• 3D realism
• anime / oil painting / fashion editorial
• external asset dependency as a first-pass requirement

Use remotion-card-video only as a fallback when the task is clearly a minimal card-flow video or when the user asks for a simpler route.

Workflow

1. Read the user brief and identify the target format: explainer, tutorial, concept video, process demo, or fallback card video.
2. Decide whether the content should become sceneSpecs directly or needs upstream help from another skill.
3. Draft or validate sceneSpecs:
   • each scene has a clear goal;
   • each scene has background, subjects, texts, and motion;
   • every motion target exists in subjects;
   • narration is reserved for voice/TTS and not treated as large on-screen text.
4. Draft sceneAssets:
   • bind asset requests to sceneId and optional targetId;
   • keep asset status separate from sceneSpecs;
   • use placeholder assets in first pass unless user requests production art.
5. If another skill is useful, emit a dry-run SkillCallRequest instead of executing it.
6. Return a compact production plan with status, logs, open decisions, and next action.

Existing Skill Coordination

Use the skill-call protocol in references/skill-call-protocol.md.

Common upstream/downstream roles:

• ai-short-film-breakdown: decide video type and narrative strategy.
• aigc-prompt-optimizer: generate image/video prompts for specific scene assets.
• prompt-master: generate cross-tool prompts or agent instructions.
• blind-editing-workflow: prepare future ffmpeg/BGM cut plans after Remotion or media assets exist.
• suno-music-brief: prepare BGM brief after timing and tone are stable.
• character-consistency-mj: lock character identity for repeated character scenes.
• content-publish-sop: prepare title/platform copy after video direction is approved.
• aigc-postmortem: review results after preview/export/testing.

For the current first version, only design requests and return shapes. Do not actually Read or execute external SKILL.md files.

sceneAssets

Use references/scene-assets.md when asset binding is needed. Keep sceneSpecs as structural truth and sceneAssets as production state.

Rule:

sceneSpecs describes what appears.
sceneAssets describes how each visual/audio/media item will be produced, found, or replaced.

Status Semantics

• dryRun: only planning; no files changed unless explicitly requested.
• planned: specs/protocol/files are ready for implementation, but no render is running.
• rendering: Remotion render/export has started.
• done: requested deliverable exists and was verified.
• error: a blocking issue occurred; include error and next recovery step.

Relationship To Current Remotion Project

Current implementation module:

E:\knowledge-base\07_skill存档\remotion\src\skills\RemotionSkill.ts

Treat that file as the prototype orchestration module. This SKILL.md is the productized agent instruction layer. See references/remotion-skill-ts-relationship.md for how to keep them aligned.

Current dryRun interface:

createVideo(params: {
  sceneSpecs: SceneSpec[];
  style?: RemotionVideoStyle;
  callExistingSkills?: ExistingSkillName[];
  outputPath?: string;
  dryRun?: boolean;
}) => Promise<RenderResult>

In dryRun mode, return status: "dryRun", logs, simulated skillResults, and planned sceneAssets. Do not start Remotion render/export.

Hard Boundaries

• Do not overwrite existing sceneSpecs or visual components without reading them first.
• Do not collapse everything into one giant component.
• Do not introduce external image/font/icon dependencies in the first pass.
• Do not treat narration as full-screen text.
• Do not make final aesthetic decisions for 跳蛛先生.
• Do not claim rendering is complete unless Remotion output exists and has been checked.