mcp-expert
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides blueprints for agents that process external data and render interactive UI components (MCP Apps), creating a potential attack surface if implemented without the recommended guards.
- Ingestion points: MCP Apps render data from remote servers, and servers ingest tool arguments supplied by the agent.
- Boundary markers: The guide explicitly recommends strictly isolated sandboxing for UI components to prevent cross-site scripting (XSS).
- Capability inventory: Blueprints describe tools for data synchronization, notifications, and browser automation.
- Sanitization: The skill recommends using Zod or Valibot for strict runtime schema validation of all inputs.
- [EXTERNAL_DOWNLOADS]: Mentions the use of standard package runners for installing MCP servers and protocol debugging tools.
- Evidence: References the use of
uvx mcp-server-browser-usefor browser automation capabilities. - Evidence: References
@modelcontextprotocol/inspectorvia npx for official protocol testing and verification.
Audit Metadata