The Agent Skills Directory

[SAFE]: The skill consists of detailed instructional guidelines for explaining complex topics and does not contain any malicious scripts or commands. It explicitly includes several security best practices for the AI agent to follow:
Restricted Execution Environment: The instructions for building HTML explainers (in references/html-explainers.md) mandate that files be self-contained, forbidding the use of external libraries, fonts, images, or network requests. This prevents potential data exfiltration or Cross-Site Scripting (XSS) in generated artifacts.
Credential Safety: The video generation section in references/karpathy-output-ladder.md instructs the agent to read API keys for text-to-speech services only from environment variables and explicitly warns to 'keep them out of files and commits.'
User Consent for Dependencies: The skill advises the agent to use already-installed tools (like ffmpeg or Manim) and specifically states to 'install nothing without the user's agreement,' mitigating the risk of unauthorized remote code execution through package installation.
Instructional Focus: The core logic focuses on technical writing standards (STE-inspired) and visual clarity, rather than performing privileged operations or accessing sensitive user data outside of the immediate context required for explanation.

explain