The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied input to generate design philosophies and visual artwork without using explicit boundary markers or sanitization. This creates an attack surface where instructions embedded in a user's design request could potentially influence the agent's behavior.
Ingestion points: User instructions provided as the conceptual foundation for design creation (SKILL.md).
Boundary markers: None present to isolate user-provided text from internal instructions.
Capability inventory: Writing files (.md, .pdf, .png), reading local fonts, and generating/executing code for image rendering.
Sanitization: None present for processing user input.
[DYNAMIC_EXECUTION]: The instructions guide the agent to write, refine, and execute code at runtime to produce visual outputs. This involves generating logic that could be subtly influenced by the design philosophy derived from user input.
Evidence: SKILL.md directs the agent to "call new functions" and "refine the code" to polish the final masterpiece.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to "download and use whatever fonts are needed" if they are not available in the local directory. This may lead the agent to fetch resources from external repositories, which could be exploited to fetch untrusted content if the agent is not restricted to well-known providers.

canvas-design