The Agent Skills Directory

External Resource Provisioning: The skill downloads model weights for background removal (e.g., u2net_human_seg) and transcription at runtime. While these are necessary for video processing, the retrieval of binary assets from remote servers is a pattern to monitor in isolated environments.
Browser Security Configuration: The layout measurement engine utilizes headless Chromium with specific flags such as --disable-web-security and --allow-file-access-from-files. These settings allow the skill to measure generated HTML files locally, which is required for pixel-perfect caption positioning, but represents a specialized security configuration for the browser instance.
Extensive Command-Line Orchestration: The rendering pipeline involves significant interaction with the host system through ffmpeg, ffprobe, and node subprocesses. These tools are used to extract audio, extract frames, and composite the final video, which is typical for media manipulation tasks.
Data Ingestion Surface: The skill processes user-provided video files and JSON configurations to generate transcriptions. It lacks explicit boundary markers or instruction-ignoring delimiters for the text generated during transcription, creating a potential surface for indirect prompt injection if the source audio contains malicious instructions.
Dynamic Script Generation: The skill generates local HTML and JavaScript files (e.g., index.html, rail.html) and executes them within a browser environment to calculate layout metrics and render frames.

embedded-captions