The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (HTML source code, URLs) when operating in CRITIQUE or REDESIGN modes. Specifically, agents/design-critic.md reads UI source files and reference/critique.md instructs the agent to render URLs via playwright-cli. The skill has capabilities including command execution (Bash) and file writing (Write). There are no explicit boundary markers or sanitization steps described to handle potentially malicious instructions embedded in this external content, which could lead to indirect prompt injection.
[EXTERNAL_DOWNLOADS]: The skill installs and executes the playwright-cli tool from Microsoft's official repository for surface rendering and verification. It also references numerous professional design libraries and kits from established vendors including Google, Vercel, IBM, and Airbnb. These external resources are from well-known services and trusted organizations and are documented neutrally.

superdesign