The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches narration audio from ElevenLabs, a well-known text-to-speech service, and installs rendering dependencies using Homebrew and the Python package registry (PyPI) via uv. These interactions target trusted and well-known services necessary for the skill's primary function.
[COMMAND_EXECUTION]: The scripts utilize system commands including ffmpeg, ffprobe, and manim for video and audio processing. Additionally, it uses the macOS security utility to retrieve the ElevenLabs API key from the system keychain, which is a secure method for handling user credentials.
[DYNAMIC_EXECUTION]: The skill's workflow involves the agent generating a Python script (scenes.py) based on defined templates and then executing it using the Manim animation engine to produce the final video output.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied topics and external sources to generate narration and animation code, which represents a potential surface for indirect prompt injection.
Ingestion points: User prompt and provided reference materials (files, URLs).
Boundary markers: None explicitly defined to isolate source content during script generation.
Capability inventory: Shell command execution (FFmpeg, Manim), network requests (ElevenLabs), and local file operations.
Sanitization: Relies on the underlying AI model's safety filters during code and script generation.

animated-explainer