Video Downloader 2.0

Boundary

This skill owns the complete media-preprocessing workflow:

video link or local file
→ acquire video and platform caption
→ usable timed platform subtitles, otherwise cloud ASR
→ choose keyframes / full-video / Agent clips
→ preview and confirm each exact cloud payload
→ Doubao image or native-video understanding
→ timestamped plain Markdown

Its final handoff is multimodal_transcript.md, a normal text file. Cangjie Skill stays unchanged and only consumes text to produce Skills. Do not move download, ASR, frame, OCR, or visual-model logic into Cangjie Skill.

Requirements