ai-sdk-ui
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides UI templates for chat and completion interfaces that ingest untrusted data from AI models. While these templates utilize
react-markdownfor safer rendering, the ingestion of arbitrary model output into the DOM represents a standard indirect prompt injection surface. - Ingestion points: Found in
templates/use-chat-basic.tsx,templates/nextjs-chat-app-router.tsx, andtemplates/custom-message-renderer.tsxwhere message content from the hook is rendered. - Boundary markers: The templates do not explicitly implement boundary markers for model output, which is typical for UI-focused skills.
- Capability inventory: The UI templates themselves do not execute commands, but they facilitate communication with backend APIs (
/api/chat,/api/completion) that may have tool-execution capabilities. - Sanitization: The skill recommends
react-markdownfor rendering (e.g., intemplates/custom-message-renderer.tsx), which helps mitigate XSS and other injection risks, though a commented-out alternative usingdangerouslySetInnerHTMLis included as a reference.
Audit Metadata