The Agent Skills Directory

[SAFE]: The skill documents a secure architecture for UI extensions. It uses iframes as a primary sandbox mechanism, ensuring that widgets (mini-apps) are isolated from the host application's environment.
[COMMAND_EXECUTION]: The framework utilizes specific actions like install-extension and add-extension-slot-target. These are high-level API operations for managing extension metadata and do not allow for arbitrary shell command execution.
[DATA_EXFILTRATION]: While the framework passes context data (such as contact information) to extensions via window.slotContext, this is an intentional design for widget functionality. The skill specifies that these extensions operate within a sandboxed environment, preventing unauthorized exfiltration of host application data.
[REMOTE_CODE_EXECUTION]: The skill explicitly defines extensions as Alpine.js mini-apps and states that the system is not intended for arbitrary code modification. The use of sandboxed iframes provides a security boundary against execution risks.
[PROMPT_INJECTION]: The instructions are strictly technical and architectural. There are no attempts to override agent behavior, bypass safety filters, or extract system prompts.

extension-points