clip-maker-clip-maker

Converted from Claude Code command /clip-maker. Review and adapt: hooks and MCP tool IDs may need manual mapping for Codex.

Clip Maker — Full Pipeline

Create vertical (9:16) clips from a horizontal video, optimized for Reels/Shorts/TikTok.

Arguments

Parse $ARGUMENTS for:

• <video_path_or_url> — required, path to source video OR YouTube/web URL
• --duration N — target clip duration in seconds (default: 60)
• --auto — skip manual moment selection, use all found moments
• --no-subtitles — don't burn subtitles into clips
• --api — use OpenAI Whisper API instead of local whisper
• --no-copy — skip social media copy generation
• --language LANG — transcription language (default: ru)
• --prompt "terms" — domain-specific terms for whisper (improves recognition of jargon, names, abbreviations)

Pipeline

Execute steps sequentially. Report progress to the user after each step.

Step 1: Check dependencies

bash plugins/clip-maker/scripts/install-deps.sh [--api if user passed --api]

If it fails, stop and report what's missing.

Step 2: Download if URL

If <video_path> looks like a URL (starts with http://, https://, or youtube.com, youtu.be):

VIDEO_PATH=$(bash plugins/clip-maker/scripts/download-video.sh "<url>" ~/Downloads)

The script outputs the downloaded file path. Use it as <video_path> for subsequent steps.

Step 3: Setup output directory

Create output directory next to the video: {video_dir}/{video_name}_clips/

Step 4: Transcribe

bash plugins/clip-maker/scripts/transcribe.sh "<video_path>" "<output_dir>" [--api] [--language LANG] [--prompt "terms"]

This produces <output_dir>/transcript.json.

Step 5: Find interesting moments

Launch the moment-finder agent:

Agent(subagent_type="clip-maker:moment-finder", model="opus", prompt="
Analyze the transcript and find the most engaging moments for vertical clips.

- Transcript path: <output_dir>/transcript.json
- Target clip duration: <duration> seconds
- Output path: <output_dir>/moments.json

Read the transcript, find the best moments, and write moments.json.
")

Step 6: User selects moments (unless --auto)

If --auto is NOT set:

1. Read moments.json
2. Present the moments to the user as a numbered list showing: title, time range, quote, score
3. Ask which moments to keep (user can type numbers like "1,3,5" or "all")
4. Update moments.json to only include selected moments

If --auto IS set: proceed with all moments.

Step 7: Extract frames

bash plugins/clip-maker/scripts/extract-frames.sh "<video_path>" "<output_dir>/moments.json" "<output_dir>"

Step 8: Analyze crop positions

Launch the crop-analyzer agent:

Agent(subagent_type="clip-maker:crop-analyzer", model="sonnet", prompt="
Analyze frames to determine optimal crop positions for vertical clips.

- Frames directory: <output_dir>/frames/
- Source video dimensions: [get from ffprobe]
- Output path: <output_dir>/crop_coords.json

Read each moment's frames, locate the speaker, and write crop_coords.json.
")

Step 9: Cut clips

bash plugins/clip-maker/scripts/cut-clip.sh "<video_path>" "<output_dir>/moments.json" "<output_dir>/crop_coords.json" "<output_dir>"

Step 10: Burn subtitles (unless --no-subtitles)

bash plugins/clip-maker/scripts/burn-subtitles.sh "<output_dir>/moments.json" "<output_dir>/transcript.json" "<output_dir>"

Step 11: Generate social media copy (unless --no-copy)

Launch the copywriter agent:

Agent(subagent_type="clip-maker:copywriter", model="sonnet", prompt="
Generate social media descriptions for the clips.

- Moments: <output_dir>/moments.json
- Output: <output_dir>/copy.md

Read moments.json and write copy.md with platform-specific descriptions.
")

Step 12: Generate manifest and report

Create <output_dir>/manifest.json with metadata for all clips:

{
  "source": "<video_path>",
  "created": "<timestamp>",
  "clips": [
    {
      "file": "clip_01.mp4",
      "title": "...",
      "start": 125.4,
      "end": 183.2,
      "duration": 57.8,
      "has_subtitles": true
    }
  ]
}

Report final summary:

• Number of clips created
• Output directory path
• Total duration of clips
• Whether copy.md was generated

Cleanup

Remove intermediate files:

• <output_dir>/frames/ directory
• <output_dir>/transcript.json — keep (useful reference)
• <output_dir>/moments.json — keep (useful reference)
• <output_dir>/crop_coords.json — remove