cut-segment

name	cut-segment
description	Extract one or more time-range segments from a video. Use when the user says "cut from 1:30 to 3:45", "trim the first 30 seconds", "extract these segments", "split this clip at 5:00". Stream-copies losslessly by default; offers a frame-accurate re-encode mode when the cut must be exact. Multiple segments are written as separate files (or optionally stitched in one pass).
disable-model-invocation	false
allowed-tools	Bash(test *), Bash(ls *), Bash(mkdir *), Bash(jq *), Bash(basename *), Bash(dirname *), Bash(realpath *), Bash(printf *), Bash(awk *), Bash(ffmpeg *), Bash(ffprobe *), Read, Write

Cut Segment

Extract precise time ranges from a video. Replaces the legacy cut-video-segment command.

Procedure

1. Inputs

Field	Default
Source	required
Segments	required: one or more start..end ranges (e.g. 00:01:30..00:03:45) or a JSON array [{start, end, label?}, ...]
Output dir	default: same dir as source (or <project>/exports/ when run inside a project)
Mode	copy (lossless, stream-copy) — precise (re-encode for frame-accurate cuts)
Stitch	no (one file per segment) — yes (concatenate all segments into a single output)

2. Validate ranges

SRC="$(realpath "$INPUT")"
DUR=$(ffprobe -v error -show_entries format=duration -of default=nw=1:nk=1 "$SRC")

to_seconds() {
  awk -F: '{
    n = NF
    if (n == 3)      print $1*3600 + $2*60 + $3
    else if (n == 2) print $1*60 + $2
    else             print $1
  }' <<<"$1"
}

For each segment: convert start/end to seconds; reject if start >= end, if start < 0, or if end > DUR. Report which segments were rejected and abort.

3. Mode: copy (default)

Stream-copy each segment. Fast, lossless, but cuts land on the nearest keyframe (typically within 1–2s of the requested mark).

i=0
for seg in "${SEGMENTS[@]}"; do
  read s e <<<"$seg"
  OUT="$OUTDIR/$(basename "${SRC%.*}")_seg_$(printf '%02d' $i)${EXT}"
  ffmpeg -y -ss "$s" -to "$e" -i "$SRC" -c copy -avoid_negative_ts make_zero "$OUT"
  i=$((i+1))
done

-ss before -i enables fast seek; combined with -c copy this is the lossless path.

4. Mode: precise

Frame-accurate cuts via re-encode. -ss goes after -i to force decode-side seek.

ffmpeg -y -i "$SRC" -ss "$s" -to "$e" \
  -c:v libx264 -crf 18 -preset medium \
  -c:a aac -b:a 192k \
  "$OUT"

If a render profile name was supplied (from render-profiles.json), build the encoder args from that profile instead.

5. Stitch (optional)

If STITCH=yes:

LIST=$(mktemp)
for f in "${SEGMENT_OUTPUTS[@]}"; do printf "file '%s'\n" "$f" >> "$LIST"; done
ffmpeg -y -f concat -safe 0 -i "$LIST" -c copy "$OUTDIR/$(basename "${SRC%.*}")_stitched${EXT}"

If the segments were produced via precise mode they share encoder params and concat-copy works. For copy mode + stitch, fall back to a re-encode through render-from-library if stream-copy concat fails.

6. Verify

For each output:

ffprobe -v error -show_entries format=duration,size -of default=nw=1 "$OUT"

Compare actual duration to requested (end - start). In copy mode, mention the keyframe-snap drift if it's >0.5s.

7. Confirm

Print a table of segment outputs with their durations and sizes. Offer follow-ups (burn-graphics to overlay a label on each, render-from-library to stitch in a richer way, audio-analysis per segment).

Notes

• Default output extension preserves the source's container.
• Multiple segments with the same numeric index won't clobber each other (they get _seg_00, _seg_01, …); supply a label per segment in the JSON form to get human-readable names instead.
• For "trim the first/last N seconds" shorthands, translate to a single segment: 00:00:00..-00:00:N (use ffprobe duration) before validation.
• For multi-file concatenation (the legacy merge-videos command's job), use render-from-library instead.
• For watermarks and overlays (the legacy add-watermark command's job), use burn-graphics instead.