The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill defines a system for runtime UI injection where worker-provided ESM JavaScript and CSS are loaded dynamically via import(). This is an architectural feature designed to extend the console without rebuilds, utilizing hash-deduped hot reloading and content function dispatching.\n- [COMMAND_EXECUTION]: Provides agent-accessible functions to manipulate the console workspace, such as console::workspace::open and console::workspace::propose. These operations facilitate user-mediated navigation and workspace organization.\n- [INDIRECT_PROMPT_INJECTION]: The console's role in rendering external UI and trace data creates a surface for indirect injection. The documentation mitigates this by enforcing error boundaries for renders, scoped CSS attributes for workers, and providing redactRaw hooks to protect sensitive data during display. The mandatory evidence chain includes ingestion at console::ui-content and trace views, boundaries provided by scoped attributes and error boundaries, capabilities restricted to the host.iii bus, and sanitization via redactRaw.\n- [SAFE]: The documentation explicitly instructs against hardcoding secrets, recommending environment-based authentication and host-managed validation boundaries to prevent unauthorized data exposure.

ade