The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to handle untrusted UI payloads from AI agents and implements an extensive security architecture to mitigate this risk.
Ingestion points: Untrusted JSON payloads are received via SSE or REST in A2UIAgentService (documented in reference/a2ui-renderer-services.md).
Boundary markers: The protocol uses a declarative component catalog system. A2UICatalogService (documented in reference/a2ui-renderer-patterns.md) enforces a strict allowlist of component types (Row, Column, Text, etc.).
Capability inventory: The renderer supports interactive actions (forwarded back to the agent) and a controlled openUrl function for navigation.
Sanitization: A2UISanitizerService provides URL protocol validation (allowing only http: and https:) and text length capping. The A2UIRendererComponent utilizes Angular's default interpolation which provides automatic XSS protection. The skill explicitly prohibits eval() and new Function() for agent-provided data.
[EXTERNAL_DOWNLOADS]: The skill references documentation and SDKs from the A2UI project and the daisyUI component library. These are standard resources for frontend development within this specific technical ecosystem and do not present unusual risk.

a2ui-angular