transcript-video-section-splitter
Installation
SKILL.md
Transcript Video Section Splitter
Use this skill when the user wants a long video divided into meaningful content sections based on what is said. The output should include a timestamped transcript, section metadata, and split video clips.
Core Rules
- Transcribe first; do not guess section boundaries from duration alone.
- Keep the original media unchanged. Split from a derived edited video when the user asks to section that edited version.
- If the user asks to split a regenerated or edited variant "as before", reuse the existing section JSON only when the variant preserves the same timeline and final duration.
- Use topic changes, long pauses, repeated framing phrases, and concrete action shifts to set boundaries.
- Store section boundaries in durable JSON before cutting clips.
- Preserve audio and avoid unnecessary re-encoding when exact keyframe cuts are acceptable; re-encode when frame-accurate starts matter.
- Write variant clips to a variant-specific folder such as
artifacts/.../v2/sections/so older split outputs are not overwritten. - Keep private or raw transcripts separate when the clips or section summaries will be published.