agent-ui
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches a component registry definition from
https://ui.inference.sh/r/agent.jsonusing theshadcnCLI pattern. - [COMMAND_EXECUTION]: Instructs the user to run
npmandnpxcommands for package installation (@inferencesh/sdk) and component integration. - [INDIRECT_PROMPT_INJECTION]: The component implements an interface that ingests untrusted data for processing by an LLM.
- Ingestion points: The
Agentcomponent accepts user messages and supports file and image uploads whenallowFilesorallowImagesare enabled. - Boundary markers: The documentation examples do not explicitly demonstrate prompt delimiters or instructions to ignore embedded commands in user-provided files or text.
- Capability inventory: The component supports tool execution, including client-side tools for UI scanning and form filling.
- Sanitization: Input sanitization and validation are handled by the underlying SDK and proxy route; they are not explicitly detailed in the UI-level documentation.
Audit Metadata