The Agent Skills Directory

[COMMAND_EXECUTION]: The provided video_viewer.py tool uses subprocess.run to call external utilities like manim, ffmpeg, and ffprobe. These calls are safely implemented using argument lists, which prevents shell command injection. The tool is designed to facilitate local development workflows.
[INDIRECT_PROMPT_INJECTION]: The video viewer interface renders project files such as plan.md. The markdown parser in ui.html performs basic HTML escaping but lacks sanitization for link protocols, allowing for potential javascript: URI injection. This is considered a low-risk vulnerability surface given the tool is restricted to the local loopback interface and is intended for the user's own project content.
Ingestion points: Project planning documents (plan.md) and Manim source code (script.py) are read by video_viewer.py and displayed in the ui.html frontend.
Boundary markers: None. Content is rendered directly in the preview panels without specific delimiting instructions.
Capability inventory: The local backend can trigger manim re-renders and ffmpeg concatenation tasks through subprocess.run calls.
Sanitization: The UI uses a basic esc() function to replace HTML characters, but does not validate or filter URLs in markdown links.

manim-skill