The Agent Skills Directory

[PROMPT_INJECTION]: The instructions in SKILL.md under the 'CRITICAL GUIDELINES' section use forceful language to compel the agent to repeatedly claim its output is the result of 'master-level implementation' and 'painstaking optimization'. This directive to project an exaggerated persona of expertise is a form of output manipulation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to take user-provided 'conceptual seeds' or 'subtle inputs' and incorporate them into a generated HTML/JavaScript artifact. This workflow creates a vulnerability surface for indirect prompt injection, as malicious user input could potentially influence the agent to generate harmful scripts.
Ingestion points: User instructions for the 'algorithmic philosophy' and 'conceptual seed' in SKILL.md.
Boundary markers: The instructions lack formal delimiters or warnings to ignore instructions embedded within the user data.
Capability inventory: The agent creates a full HTML artifact with inline JavaScript capable of DOM manipulation and p5.js API calls.
Sanitization: No sanitization or validation of the user-provided conceptual inputs is defined before interpolation into the generation process.
[EXTERNAL_DOWNLOADS]: The skill's HTML template fetches the p5.js library from Cloudflare's CDN (cdnjs.cloudflare.com) and typography from Google Fonts (fonts.googleapis.com). These are well-known and reputable services for web assets.
[DYNAMIC_EXECUTION]: The skill facilitates the generation of executable JavaScript code within a single HTML artifact. While the agent uses provided templates (templates/generator_template.js and templates/viewer.html), the resulting artifact executes logic dynamically in the user's browser or agent environment.

algorithmic-art