vercel-ai-sdk

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). The skill explicitly fetches and uses external third-party content—e.g., creating MCP clients with StreamableHTTPClientTransport/SSE (process.env.MCP_SERVER_URL), calling mcpClient.tools() and mcpClient.readResource(), and example webSearch tools—so the agent ingests and interprets untrusted public/tool-provided content at runtime, which could enable indirect prompt injection.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 1.00). The skill creates an MCP HTTP transport to runtime endpoints like https://mcp-server.example.com/mcp and then calls mcpClient.tools() / mcpClient.getPrompt(), which fetches remote tool definitions and prompts that directly control the agent's instructions and enable remote-executed tool behavior, so this is a high-risk runtime dependency.