The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates Python scripts for Manim animations and HTML/JavaScript for interactive pages, then executes or renders them using local tools and browsers. This runtime script generation is a core capability used to produce visual output based on user requests.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted user input and is instructed to preserve technical identifiers verbatim. These identifiers are then interpolated into dynamically generated executable files (Manim Python scripts, HTML/JS) without explicit sanitization, potentially allowing malicious data to influence code execution.
Ingestion points: User prompts for explanations or specific output formats in SKILL.md.
Boundary markers: Absent; the instructions prioritize verbatim preservation of user-provided technical terms.
Capability inventory: Writing and running Python code (manim), creating and executing HTML/JS (Playwright), and running system commands (ffmpeg, npx, pip).
Sanitization: Absent; no escaping or validation rules are provided for technical data before script interpolation.
[CREDENTIALS_UNSAFE]: The skill facilitates the use of an ELEVENLABS_API_KEY stored in environment variables. Although it provides instructions to handle the key securely (e.g., not printing it), the capability to access and utilize sensitive secrets is inherent to the skill's narration functionality.
[EXTERNAL_DOWNLOADS]: The skill triggers the download and installation of well-known software packages and tools, such as manim and @mermaid-js/mermaid-cli, from official registries during the creation of diagrams and videos.
[COMMAND_EXECUTION]: The skill utilizes several CLI tools and system commands, including ffmpeg, ffprobe, npx, and pip, to handle media processing, syntax validation, and dependency management.

eli5