Claude Video Editing Agent

A creative assistant specializing in programmatic video and audio editing using the Claude Code SDK.

Features

• Video Processing: Download videos with yt-dlp, edit with ffmpeg
• Audio Editing: Process audio files with ffmpeg
• Transcription: OpenAI Whisper for speech-to-text (supports 99 languages, outputs SRT/VTT/TXT)
• Image Viewing: Claude can directly view images using the Read tool
• Full Filesystem Access: Read, Write, Edit files with pattern matching (Glob, Grep)
• Web Access: WebFetch and WebSearch capabilities
• Task Management: Built-in todo lists and sub-agents

Prerequisites

• Python 3.10+
• Node.js
• Claude Code: npm install -g @anthropic-ai/claude-code

Installation

# Install all Python dependencies
pip install -r requirements.txt

# ffmpeg should be installed via system package manager
# macOS: brew install ffmpeg
# Ubuntu: apt install ffmpeg

Usage

python main.py

Example Requests

• "Download a video from YouTube using yt-dlp"
• "Extract the audio from video.mp4"
• "Transcribe this audio file and generate subtitles"
• "View the image file in the workspace and describe what you see"
• "Create a 10-second video with ffmpeg"
• "Search the web for the latest ffmpeg documentation"

Available Tools

The agent has access to:

• File Operations: Read, Write, Edit, MultiEdit
• Search: Glob (file patterns), Grep (content search)
• Execution: Bash (for ffmpeg, yt-dlp, whisper, etc.)
• Web: WebFetch, WebSearch
• Task Management: Task (sub-agents), TodoWrite, SlashCommand

Workspace

All work happens in the ./workspace/ directory, which is automatically created on startup.

Notes

• Claude's Read tool supports viewing images natively - no custom tools needed!
• yt-dlp can download from YouTube, Vimeo, and 1000+ other sites
• ffmpeg is available for complex video/audio processing tasks
• Whisper supports 99 languages and can output text, SRT, VTT, TSV, and JSON formats
• Whisper model sizes: tiny, base, small, medium, large (larger = more accurate but slower)