Clip Local

Use this skill for local video clipping, not for publishing or remote editing.

Core workflow

1. Verify tools first:
   • yt-dlp
   • ffmpeg
   • python3
2. Decide the mode:
   • explicit start/end times -> clip directly
   • no times -> identify 3-5 highlight candidates, then ask the user which segment to keep
3. Prefer source subtitles when available:
   • detect original language
   • download original subtitles
   • trim subtitle cues to the selected clip range
4. Only fall back to transcription when needed:
   • if no subtitles exist and GROQ_API_KEY is present, run Whisper fallback
   • otherwise ask whether to continue without subtitles
5. Burn subtitles only when requested:
   • plain clipping -> prefer stream copy when safe
   • karaoke / bilingual output -> use the bundled ASS helper and re-encode intentionally

Execution notes

• The bundled subtitle helper lives at scripts/ass-karaoke.py.
• Keep absolute timestamps in the trimmed VTT; shift relative to clip start with the helper's offset flag.
• Re-resolve media URLs if the clip step fails after a long delay because signed stream URLs expire.

Reference routing

• Read scripts/ass-karaoke.py when you need exact subtitle rendering behavior.
• Read nearby runtime files only if you need to debug clipping, subtitle timing, or fallback transcription.

Working rules

• Do not use auto-translated YouTube subtitles as the primary source.
• Confirm highlights with the user before clipping when they did not specify timestamps.
• Keep clipping local; do not route this skill into external publishing flows.
• Explain when subtitle burn-in forces re-encoding and longer runtime.