The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external media processing binaries to render the final video output.
scripts/render.py uses subprocess.Popen to call ffmpeg for encoding video and audio streams from browser-rendered frames.
The script also calls ffprobe to verify the metadata of the generated MP4 file.
Security measures: Input parameters such as width, height, and FPS are cast to integers and validated before being passed to the command line.
[DYNAMIC_EXECUTION]: The rendering pipeline involves dynamic execution of JavaScript within a browser context.
scripts/render.py utilizes the Playwright library to launch a headless Chromium instance to render the film.html template.
The script inlines local project scripts into the HTML and executes window.renderAt(t) to capture individual frames.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied project configuration files, creating a potential attack surface for indirect instructions.
Ingestion points: The agent is instructed to read brief.js and brief.json from the project directory to define the film's content and timing.
Boundary markers: The skill documentation includes extensive instructions on verifying evidence and clearing assets, though no explicit string delimiters are mandated for data interpolation.
Capability inventory: The skill has the capability to write files, execute subprocesses (ffmpeg), and perform network operations via Playwright (though primarily used for local rendering).
Sanitization: The Python scripts validate numeric inputs (dimensions, duration) before use in shell commands, reducing the risk of command injection via configuration data.
[EXTERNAL_DOWNLOADS]: The skill requires standard Python libraries for data processing and video manipulation.
requirements.txt specifies numpy, Pillow, PyYAML, playwright, and av (PyAV). These are well-known packages from the official PyPI registry.

launch-film