The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses a "Final Step" section to instruct the agent to behave as though a specific user feedback has already occurred ("The user ALREADY said 'It isn't perfect enough...' "). This technique is designed to override the agent's perception of the current interaction state to force iterative refinement loops.
[EXTERNAL_DOWNLOADS]: The instructions explicitly direct the agent to "Download and use whatever fonts are needed." While the skill includes license files for many common open-source fonts (like those from Google Fonts), the broad directive encourages the agent to fetch external assets from the internet during its execution.
[DYNAMIC_EXECUTION]: To create visual designs, the skill requires the agent to generate and execute code (e.g., to render graphics into .pdf or .png files). The instructions emphasize a process of writing, running, and refining code to achieve high-quality visual output, which involves runtime script generation and execution.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill processes user-supplied conceptual inputs to form the basis of a design philosophy manifesto as described in SKILL.md.
Boundary markers: Absent. There are no specified delimiters or instructions for the agent to distinguish between its own logic and potentially malicious instructions embedded in the user-supplied design concepts.
Capability inventory: The agent has the ability to write multiple file types (.md, .pdf, .png), execute generated code, and potentially perform network operations to download fonts.
Sanitization: Absent. The skill does not provide mechanisms to sanitize or validate the user-provided design inputs before they are incorporated into the agent's logic and manifesto creation.

canvas-design