agent-ui

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches UI component configuration from the vendor domain ui.inference.sh using the shadcn CLI tool.\n- [EXTERNAL_DOWNLOADS]: Installs the @inferencesh/sdk package to handle proxy routing and SDK features.\n- [PROMPT_INJECTION]: The skill implements a chat interface that processes untrusted user data, creating an indirect prompt injection surface.\n
  • Ingestion points: User inputs and potential file/image uploads processed by the Agent component in SKILL.md.\n
  • Boundary markers: Not explicitly defined in the provided usage examples.\n
  • Capability inventory: Includes client-side tool execution (e.g., scan_ui, fill_field) and human-in-the-loop approval flows.\n
  • Sanitization: Not specified in the component documentation or configuration examples.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 10:49 PM