The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exposes a potential attack surface by ingesting untrusted user prompts to deduce conceptual art themes.
Ingestion points: User-provided requests for art, posters, and specific design topics are processed in SKILL.md and sub-skills/essential-principles.md to establish a conceptual framework.
Boundary markers: The instructions do not define specific delimiters or boundary markers for the interpolation of user-derived concepts into the art generation process.
Capability inventory: The skill is designed to generate and save markdown (.md), PDF (.pdf), and image (.png) files across its operational steps.
Sanitization: There is no explicit sanitization or filtering of the user-provided input before it is used to influence the visual and philosophical output. However, the visual nature of the primary output formats significantly mitigates the risk of these files being used for automated downstream injection attacks.
[EXTERNAL_DOWNLOADS]: In sub-skills/essential-principles.md, the instructions suggest that the agent should "Download and use whatever fonts are needed" to ensure high-quality craftsmanship. While this represents a generic request for external assets, the skill files do not provide specific remote URLs, untrusted sources, or automated scripts to facilitate these downloads. The skill instead references a local ./canvas-fonts directory containing license information for well-known open-source font projects.
[PROMPT_INJECTION]: The skill uses strong imperative language and instructional markers such as "CRITICAL," "IMPORTANT," and "non-negotiable" to steer the agent toward high-quality aesthetic output. For example, sub-skills/essential-principles.md employs a simulated context technique ("The user ALREADY said...") to compel the agent to refine its work to a "masterpiece" level. These patterns are focused on artistic quality control and do not attempt to bypass safety guidelines or override system-level security constraints.

canvas-design