The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates executable content at runtime to fulfill different output formats. This includes creating self-contained HTML files with inline JavaScript for interactive explainers and Python scripts for the Manim animation library to generate video content.
[INDIRECT_PROMPT_INJECTION]: The skill accepts arbitrary user-provided topics via the $ARGUMENTS parameter to generate complex outputs. This creates a surface for indirect prompt injection where a maliciously crafted topic description could attempt to influence the logic of the generated code (HTML/JS or Manim scripts).
Ingestion points: User-supplied topic requests processed in SKILL.md.
Boundary markers: The instructions do not define explicit delimiters or instructions to the model to ignore embedded commands within the user's input.
Capability inventory: The skill has the capability to write files (SVG, HTML, video), execute shell commands (ffmpeg, manim), and interact with the ElevenLabs network API.
Sanitization: There are no documented procedures for sanitizing or escaping the user's input before it is interpolated into the generated scripts or HTML files.
[COMMAND_EXECUTION]: The skill relies on several external command-line utilities for media processing, including ffmpeg for video assembly and local text-to-speech tools such as Kokoro, Piper, or the macOS say command.
[CREDENTIALS_UNSAFE]: The skill is explicitly instructed to access the ELEVENLABS_API_KEY from the environment. While it follows the best practice of reading from the environment rather than using hardcoded values, it confirms the intent to handle sensitive API credentials.
[EXTERNAL_DOWNLOADS]: The skill instructions allow the agent to install external animation and text-to-speech tools if they are not already present on the system, which involves downloading software from external package registries.

asd