菜单
Choose the best starter composition for the user's request. Use this first, at the start of every new video request, before generating any content.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | pick-composition |
| description | Choose the best starter composition for the user's request. Use this first, at the start of every new video request, before generating any content. |
Your first step for any new video. Read the user's prompt and choose one
starter from /.agents/workspace/compositions/. Don't author HTML — every starter is
already a working HyperFrames project; you only pick one and (later) fill in
its variables.
/.agents/workspace/compositions/
has a manifest.json. Read all of them.use_case, aspect_ratio,
duration_seconds, and tags.Rules of thumb:
social-promo (9:16, ~6s).app-trailer (16:9, ~12s, three scenes).explainer (16:9, ~14s, numbered steps).If the user names an aspect ratio (portrait vs landscape), let that override the
vibe match — a portrait request should land on social-promo even if the
content is product-y.
If the user explicitly names a starter ("use the explainer"), use it.
Decide on one composition id (social-promo, app-trailer, or explainer)
and remember it for the rest of the pipeline. Tell the user briefly what you
picked and why ("I'll use the app trailer — landscape, three scenes, good for
showing features"). Then move to generate-script.
Render the staged composition and return the playable video. Picks between HeyGen cloud render (default) and local in-sandbox render at runtime. Use as the final step, after customize-composition.
Validate the generated content against the composition's real variable schema and stage it for rendering. Use after generate-script and before render-and-return.
Decide the actual content (headline copy, feature lines, colors, etc.) for the chosen composition's variables. Use after pick-composition and before customize-composition.
Synthesize a spoken voiceover for the video from the narration script via HeyGen TTS. Use after generate-script and before customize-composition. Runs by default for every video unless the user asked for no narration.