The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to inject specific, repetitive behavioral cues into its own intermediate output to influence its persona and implementation choices in later stages. The 'CRITICAL GUIDELINES' section in SKILL.md requires the model to stress its craftsmanship by repeating phrases like 'meticulously crafted algorithm' and 'master-level implementation' to steer the perceived expertise of the agent.
[INDIRECT_PROMPT_INJECTION]: The skill uses a multi-step chain where the 'Algorithmic Philosophy' document generated in the first phase acts as a guiding prompt for the implementation phase. The instructions to seed this document with persona-steering language create an indirect injection vector. 1. Ingestion points: The generated philosophy .md file referenced in SKILL.md. 2. Boundary markers: Absent; there are no delimiters separating instructions from data in the generated philosophy. 3. Capability inventory: Generates HTML and JavaScript artifacts that execute in the user's environment. 4. Sanitization: Absent.
[EXTERNAL_DOWNLOADS]: The skill uses a template that loads the p5.js library from a trusted public content delivery network (CDNJS). Evidence: The script tag referencing cdnjs.cloudflare.com in templates/viewer.html.

algorithmic-art