render-from-library

name	render-from-library
description	Assemble a single video from a folder of clips — concat in a chosen order, optionally with crossfade transitions, normalized to a target resolution/fps. Use when the user says "stitch this folder into one video", "render-from-library", "concat the clips in folder X", "make a montage from these clips". Lossless concat path when all clips share codec/params; transcoding fallback otherwise.
disable-model-invocation	false
allowed-tools	Bash(test *), Bash(ls *), Bash(mkdir *), Bash(find *), Bash(sort *), Bash(jq *), Bash(basename *), Bash(dirname *), Bash(realpath *), Bash(printf *), Bash(awk *), Bash(ffmpeg *), Bash(ffprobe *), Read, Write

Render From Library

Assemble a single deliverable from a folder of clips.

Procedure

1. Inputs

Field	Default
Source folder	required
Output	default: <source>/_assembled_<timestamp>.mp4
Order	name (alphabetical) — mtime (modification time) — manual (user supplies an ordered list)
Recursive	no
Transitions	none (hard cuts) — crossfade:<seconds> (default 0.5s when on)
Normalize	auto — re-encode if clips differ; pick most-common resolution/fps as target
Render profile	optional — apply a named profile from render-profiles.json for the final encode

2. Enumerate

SRC="$(realpath "$INPUT")"
if [ "$RECURSIVE" = "yes" ]; then
  mapfile -t CLIPS < <(find "$SRC" -type f \( -iname '*.mp4' -o -iname '*.mov' -o -iname '*.mkv' -o -iname '*.m4v' \))
else
  mapfile -t CLIPS < <(find "$SRC" -maxdepth 1 -type f \( -iname '*.mp4' -o -iname '*.mov' -o -iname '*.mkv' -o -iname '*.m4v' \))
fi
case "$ORDER" in
  name)  IFS=$'\n' CLIPS=($(printf '%s\n' "${CLIPS[@]}" | sort)); unset IFS ;;
  mtime) IFS=$'\n' CLIPS=($(printf '%s\n' "${CLIPS[@]}" | xargs -d '\n' stat -c '%Y %n' | sort -n | cut -d' ' -f2-)); unset IFS ;;
  manual) ;; # caller supplies the array directly
esac
[ ${#CLIPS[@]} -lt 2 ] && { echo "Need at least 2 clips."; exit 1; }

3. Probe each clip

probe() {
  ffprobe -v error -select_streams v:0 \
    -show_entries stream=width,height,r_frame_rate,codec_name -of json "$1"
}

Build a table of (codec, w, h, fps) per clip. Record the most-common signature.

4. Choose strategy

• Lossless concat if all clips share codec + container + width/height/fps and there are no transitions — use ffmpeg concat demuxer:

  printf "file '%s'\n" "${CLIPS[@]}" > /tmp/concat-list.txt
  ffmpeg -f concat -safe 0 -i /tmp/concat-list.txt -c copy "$OUT"
  

• Filter-graph re-encode otherwise. Build with xfade for crossfades or concat filter for hard cuts. All clips first re-scaled / fps-converted to the target signature with scale + fps.

5. Build the filter graph (re-encode path)

For hard cuts (no transition):

[0:v]scale=W:H,fps=F,setsar=1[v0];[1:v]scale=W:H,fps=F,setsar=1[v1];...
[v0][0:a][v1][1:a]...concat=n=N:v=1:a=1[v][a]

For crossfades:

Chain xfade between consecutive video streams, offsetting at (cumulative_duration - fade_duration). Audio uses acrossfade. The cumulative offset must be computed from each clip's duration via ffprobe — list them in the preview before running.

6. Pick the encoder

If a render profile was supplied, build the output args from it (same shape render-with-profile uses). Otherwise default to -c:v libx264 -crf 20 -preset medium -c:a aac -b:a 192k.

7. Preview

Print:

• ordered clip list with durations + total duration
• chosen strategy (lossless concat / re-encode + transition)
• target resolution/fps and encoder
• estimated output path

Ask for confirm before running.

8. Run + verify

mkdir -p "$(dirname "$OUT")"
ffmpeg ... "$OUT"
ffprobe -v error -show_entries format=duration,size -of default=nw=1 "$OUT"

9. Confirm

Print output path, size, duration. Offer follow-ups (audio-analysis, burn-graphics, generate-deliverables).

Notes

• Crossfades require all clips to be re-encoded — don't promise a "lossless crossfade".
• Audio sample rates that differ across clips are resampled to 48 kHz on the re-encode path.
• For a folder of stills, prefer the audio-to-music-video skill (cover-art + audio composite) — this skill is for video clips.
• Manual ordering: caller passes an explicit ordered list. The skill should accept it as a JSON array argument or read from a library.txt in the source folder, one filename per line.