The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches p5.js, p5.sound, and other utility libraries from well-known content delivery networks (CDNs) including Cloudflare (cdnjs), jsdelivr, and unpkg. These are standard resources for web-based visual arts and are considered safe.
[COMMAND_EXECUTION]: The instructions direct the agent to utilize local command-line tools such as node, bash, python3, and ffmpeg for previewing sketches and rendering video files. These operations are restricted to the primary creative and export functions of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user prompts to drive its generative art pipeline. This creates a surface where malicious instructions embedded in a user's creative request could attempt to influence the generated code or the behavior of export scripts. (1) Ingestion points: User creative concepts and parameter requests defined in the Creative Vision stage (SKILL.md). (2) Boundary markers: None explicitly defined to delimit user-provided text from agent instructions. (3) Capability inventory: Local execution of node, bash, and ffmpeg for video rendering, and browser-level file saving (saveCanvas). (4) Sanitization: No specific input sanitization or validation logic is provided in the instructions.

p5js