The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes message strings that are displayed in the client UI, which presents a surface for indirect injection. 1. Ingestion points: Data enters through the message argument in the setToastCookie function (found in architecture.md). 2. Boundary markers: The skill includes 'Anti-Patterns' and 'Security Considerations' that explicitly warn against including raw user input. 3. Capability inventory: The system renders text content to the user's browser using third-party toast libraries. 4. Sanitization: Documentation advises relying on the library's escaping mechanisms and provides code examples of safe vs. unsafe interpolation.
[DATA_EXPOSURE]: The skill implements a cookie-based notification system where the app-toast cookie is configured with httpOnly: false in architecture.md. This is a documented trade-off necessary for the client-side handler to read the message. The risk is mitigated by explicit instructions to never store sensitive data or credentials within the toast payload.

toast-notifications