The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of user requests, screenshots, and mockups, which represents a potential surface for indirect prompt injection.
Ingestion points: Data enters the agent context through user instructions and pasted mockups/screenshots as specified in the triggers section of SKILL.md.
Boundary markers: The skill does not implement specific delimiters or instructions to ignore embedded commands within user-provided visual or textual data.
Capability inventory: The skill utilizes Read, Write, and Edit tools to modify the codebase based on interpreted design requirements.
Sanitization: No explicit sanitization or validation of the contents of screenshots or user prompts is performed before they influence the agent's output.
[EXTERNAL_DOWNLOADS]: The skill references and links to reputable external resources and asset providers for design validation and UI elements.
Evidence: REFERENCES.md includes links to well-known services such as WebAIM, Google Fonts, Fontshare, Coolors, Heroicons, and Lucide.

ui-ux-design