danielrosehill/claude-video-editor-plugin/auto-cut-silences

Auto-Cut Silences

Wrapper around `auto-editor` — detects silent segments based on an audio threshold and rebuilds the file with those segments removed. Re-encodes only the affected ranges; the rest is stream-copied where possible.

Procedure

1. Resolve auto-editor

Auto-editor is installed by deps-setup into the plugin's shared uv venv. Locate it via the venv path recorded in preferences.json:

DATA_DIR="${CLAUDE_USER_DATA:-${XDG_DATA_HOME:-$HOME/.local/share}/claude-plugins}/video-editing"
PREFS="$DATA_DIR/preferences.json"
VENV=$(jq -r '.python_venv // empty' "$PREFS" 2>/dev/null)
 
if [ -n "$VENV" ] && [ -x "$VENV/bin/auto-editor" ]; then
  AUTO_EDITOR="$VENV/bin/auto-editor"
elif command -v auto-editor >/dev/null; then
  AUTO_EDITOR="$(command -v auto-editor)"
else
  echo "auto-editor not found. Run deps-setup and select auto-editor."
  exit 1
fi

2. Inputs

Auto-editor accepts these as --silent-threshold, --margin, and --min-clip-length / --min-cut-length.

3. Run

"$AUTO_EDITOR" "$SRC" \
  --silent-threshold "$THRESHOLD" \
  --margin "$MARGIN" \
  --output "$OUT"

For more aggressive trimming (e.g. talking-head where pauses are abundant), suggest --margin 0.1sec --silent-threshold 3%.

For a preview-only run (don't write a new file, just report what would be cut):

"$AUTO_EDITOR" "$SRC" --export timeline -o /tmp/auto-cut-timeline.json

The exported timeline is a JSON list of kept ranges — useful for showing the user what would be removed before committing.

4. Report

Print before/after duration and the cut ratio:

DUR_IN=$(ffprobe -v error -show_entries format=duration -of default=nw=1:nk=1 "$SRC")
DUR_OUT=$(ffprobe -v error -show_entries format=duration -of default=nw=1:nk=1 "$OUT")
RATIO=$(awk "BEGIN{printf \"%.1f\", ($DUR_IN-$DUR_OUT)/$DUR_IN*100}")
 
echo "In : ${DUR_IN}s"
echo "Out: ${DUR_OUT}s"
echo "Cut: ${RATIO}% removed"
echo "File: $OUT"

If the cut ratio is <2% or >90%, warn — the threshold is probably wrong.

5. Iterate

Auto-editor is fast on clip-length input but slow on long lectures. For tuning, suggest the user:

1. Run on a 60-second extract first to dial in --silent-threshold and --margin.
2. Then run on the full file.

Notes

• Auto-editor handles both video and pure audio. For video, it preserves video stream timing while cutting matching ranges from both streams.
• This skill does not normalize loudness. Pair with normalize-audio afterward if levels are uneven.
• For complex agentic editing ("make a 60-second cut down of this hour-long stream"), use a future agentic-edit skill that wraps video-use / VideoAgent. This skill is for the simple "remove silences" case only.