The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to produce self-contained HTML/JavaScript artifacts. It implements a 'script generation and execution' pattern where the agent writes code that is subsequently executed in the user's browser environment. Without explicit sanitization of user-provided 'conceptual seeds,' this creates a risk of injecting malicious payloads into the generated scripts.
[PROMPT_INJECTION]: The skill presents a high surface for indirect prompt injection (Category 8). It ingests untrusted data from the user to 'deduce the conceptual seed' and 'articulate the philosophy.' These user-controlled values are then interpolated into the generated code and manifestos without boundary markers or validation, potentially allowing a user to override agent behavior or execute malicious logic in the resulting artifact.
[PROMPT_INJECTION]: The instructions in SKILL.md contain deceptive framing requirements, ordering the agent to 'emphasize craftsmanship REPEATEDLY' and claim the algorithm was 'refined through countless hours' by a 'master of computational aesthetics.' This is an instruction to intentionally mislead the user regarding the AI's generation process.
[COMMAND_EXECUTION]: The skill's implementation logic involves gathering user-influenced 'algorithmic philosophies' and directly applying them to the core parameters of executable p5.js sketches in the 'viewer.html' template, which facilitates the creation of dynamic, executable content from unverified inputs.

algorithmic-art