ai-sdk-6

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

• Third-party content exposure detected (high risk: 1.00). The skill's documentation and required workflow explicitly show agents fetching and injecting external content—e.g., async prepareCall using vectorSearch (references/agents.md), RAG middleware that adds retrieved sources to model instructions (references/middleware.md), and MCP client/tool discovery reading remote resources (references/mcp.md)—so untrusted third‑party webpages/documents can be ingested and directly influence agent behavior.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill shows runtime calls to MCP servers (e.g., createMCPClient transport.url = "https://your-server.com/mcp") which are used to fetch tools, prompts, and resources that are injected into agent instructions and tool definitions, so external content at that URL can directly control prompts/behavior at runtime.