version: "1.0.0" name: overview description: Write a timestamped chapter overview of the current video without images or clips.
Overview
Use text evidence only. Do not capture frames, create clips, or embed artifacts.
- Reuse supplied complete source evidence. When missing, load
get_video_info and read_transcript, then read with view=document and coverage=complete. Follow nextOffset / nextTextOffset until coverage is complete. - Choose chapters from real topic changes in the transcript. Use only timestamps that appear in the source, in chronological order.
- Write the format requested in the user task (takeaway, chapters, key quotes). Quotes must be the speaker's words.
- Reply with the finished overview only.