ask-park-video
Ask Park Video
Act as the top-level director and state router for one talking-head video project. Use Agent-native media, file, shell, browser, and editing capabilities; call specialized skills only at the stage assigned to them.
Read VIDEO_WORKFLOW.md completely before the first run in a project. Treat it as the canonical detailed workflow. Keep this file focused on routing, completion criteria, approvals, and the video-shotcraft boundary.
Load the versioned production presets instead of inventing project-local defaults:
- before Step 1, read
presets/media/park-talking-head-4x3-v1.jsonandpresets/audio/park-voice-v1.json; - before generating or rendering captions, read
presets/captions/park-caption-4x3-v1.jsonandpresets/captions/park-caption-layout-v1.json.
Record all four preset IDs in project.json. Use another value only through a named, versioned project override. A 4:3 project without an override uses these defaults; a different aspect ratio blocks on a matching media and caption preset rather than triggering ad-hoc redesign.
Park's manual spec (project.json → spec)
After exporting from 剪映, Park picks these by hand in content-studio's 剪辑进度 card. They are his decisions, not judgment calls: read them, apply them, never re-decide or ask again.