The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a two-step generation pipeline where user-provided input is first used to create an 'Algorithmic Philosophy' (Markdown), which then serves as the primary instruction set for the p5.js implementation. This architecture creates an indirect prompt injection surface as the secondary code-generation phase consumes data that originated from untrusted user input without explicit sanitization or boundary markers.
Ingestion points: User requests for specific art styles or themes are interpreted to form the 'Conceptual Seed' and 'Philosophy'.
Boundary markers: None present in the instructions to separate user input from agent-generated instructions.
Capability inventory: The agent uses the Read tool to access templates and generates .md, .html, and .js files.
Sanitization: No sanitization or escaping of user input is performed before it is interpolated into the generative process.
[EXTERNAL_DOWNLOADS]: The templates/viewer.html file includes external references to well-known and trusted services for its rendering environment.
The p5.js library is loaded from Cloudflare's CDN (cdnjs.cloudflare.com).
Web fonts (Poppins and Lora) are loaded from Google's font service (fonts.googleapis.com).
These downloads follow safe patterns from established vendors.
[PROMPT_INJECTION]: The instructions contain directives for the agent to reinforce specific behavioral traits through repetition, such as 'Emphasize craftsmanship REPEATEDLY' and using specific phrases like 'painstaking optimization' or 'master-level implementation'. While these are used to control artistic tone and persona rather than bypass safety filters, they utilize reinforcement patterns typical of prompt engineering.

algorithmic-art