Enable coding agents to handle local video editing end-to-end
This tool creates an environment for processing video editing entirely on your computer without cloud uploads or API keys. It provides a workflow where coding agents execute FFmpeg commands directly and verify the results themselves. It is suitable for developers and creators who want to automate video editing but prioritize data security.
The editing flow from measurement to verification
This tool standardizes the series of steps an agent takes to handle video files. First, it uses probe.py to measure the file's duration, frame rate, resolution, and audio configuration. It then performs the necessary editing tasks and verifies that the results match the intent using check.py and look.py. All operations occur locally, ensuring video files are not sent to external servers. The agent decides the next steps based on actual measured data rather than file names.
42 tools and platform-specific templates
It includes 42 Python scripts required for video editing.
- Subtitles and karaoke It generates subtitles using a speech recognition engine or applies existing text files to the video. It also supports karaoke-style animations.
- Audio processing It automatically removes silent sections or normalizes audio to a specific loudness level. It also includes dropout and noise removal features.
- Screen composition It converts videos to vertical format or trims specific sections to fit a desired length. It also supports multicam editing and transition effects.
Installation and requirements
The system must have ffmpeg and python3 installed. Install the tool with the npx ffmpeg-skill command and check the system status with npx ffmpeg-skill doctor. To use speech recognition-based subtitle generation, you must additionally install a local speech engine such as faster-whisper or parakeet-mlx. These engines download model files on first run but operate locally thereafter. In Windows environments, be careful with Python path settings when using Git Bash.
The agent's scope of judgment and limitations
This tool does not make editing judgments. Decisions on where to cut, which subtitles are good, or which scenes are interesting rely on user instructions or the judgment of other agents. The contact sheet generated by look.py is merely material for the agent to visually confirm results; it does not automatically recognize faces or objects. Therefore, important editing decisions require final human review. Additionally, if specific FFmpeg filters or encoders are missing from the system, some features may be limited, so it is advisable to carefully check the results of the doctor command.