The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses hardcoded simulation of user feedback in the 'FINAL STEP' section of SKILL.md to override the model's standard output logic. By asserting that the user has already requested higher quality and 'masterpiece' craftsmanship, it forces the agent into a specific refinement mode, which acts as an instruction override technique.
[INDIRECT_PROMPT_INJECTION]: The skill incorporates user-provided conceptual inputs into its design manifesto and visual creation process without sufficient safety controls.
Ingestion points: User-supplied 'subtle input or instructions' described in SKILL.md as the foundation for design philosophy.
Boundary markers: Absent. The skill does not provide delimiters to isolate untrusted user concepts from the agent's core instructions.
Capability inventory: The skill generates visual artifacts (.pdf and .png), which typically involves executing image rendering tools or libraries.
Sanitization: Absent. No instructions are provided to sanitize or escape external content before it is interpolated into prompts or manifesto generation.
[DYNAMIC_EXECUTION]: The skill directs the agent to gather external resources ('Download and use whatever fonts are needed') and iteratively modify rendering logic ('Go back to the code and refine/polish further') during the rendering process, implying runtime resource fetching and code modification.
[EXTERNAL_DOWNLOADS]: The skill includes font license files that reference external repositories and official domains. These references target established organizations and well-known open-source projects, such as Google Fonts, Vercel, and JetBrains.

canvas-design