agents-ui
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses the shadcn/ui registry system to fetch component configurations from https://livekit.io/ui/r/{name}.json. This is a standard installation method for shadcn-compatible UI libraries from the official service provider.
- [INDIRECT_PROMPT_INJECTION]: The skill includes components like
AgentChatTranscriptthat process and display data from user speech or chat messages, creating a potential surface for indirect injection if the backend agent acts on this data without validation. - Ingestion points: Data processed and displayed via the
AgentChatTranscriptanduseVoiceAssistantcomponents (SKILL.md, references/components.md). - Boundary markers: None identified in the documented UI component props.
- Capability inventory: The skill provides React UI components; dangerous capabilities like command execution or file writes would reside in the separate backend agent services (Python/Node.js) it connects to.
- Sanitization: Standard React rendering provides protection against XSS, but the skill documentation does not mention specific sanitization for LLM-generated instructions in the transcript stream.
Audit Metadata