tools
Pass
Audited by Gen Agent Trust Hub on Jul 11, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill consists of technical reference material for the assistant-ui framework. The provided code snippets are standard implementation examples and do not contain malicious instructions or payloads.
- [PROMPT_INJECTION]: The documentation identifies and addresses indirect prompt injection surfaces within the generative UI system, where an LLM specifies components via JSON.
- Ingestion points: The generative-ui message part receives JSON specifications from the LLM (detailed in references/generative-ui.md).
- Boundary markers: The system implements a component allowlist as a security boundary to restrict available UI components.
- Capability inventory: Documented capabilities include browser-side actions such as clipboard access, local storage, and navigation (detailed in references/make-tool.md).
- Sanitization: The documentation provides explicit developer mitigations, including blocking 'javascript:' URLs and avoiding 'dangerouslySetInnerHTML'.
- [COMMAND_EXECUTION]: Documentation in references/mcp-server.md describes the use of StdioClientTransport for the Model Context Protocol (MCP), which involves spawning local processes for tool servers. This is correctly identified as a development-time feature for local connectivity.
Audit Metadata