ai-sdk-6-skill

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly integrates external MCP servers via createMCPClient (e.g., listing tools from mcp), and documents provider-native tools like xAI web_search and x_search plus examples using webSearch/readPage that fetch and ingest public web and social content for agent workflows, exposing the agent to untrusted third-party content.

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

• Potentially malicious external URL detected (high risk: 1.00). The MCP integration example shows a runtime connection to an external Model Context Protocol server (createMCPClient with transport.url='https://mcp.example.com/sse') and then calls mcp.listTools() whose returned tools are merged into the agent, meaning remote content can directly control agent instructions/tool execution.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (high risk: 1.00). The skill prompt includes an explicit tool named "transferFunds" with parameters (amount, to) and an execute implementation placeholder, i.e., a function whose purpose is to move money. That is a specific, finance-facing operation (a "send transaction" style tool), so it meets the criteria for Direct Financial Execution. Other parts of the skill are generic agent/tool patterns and safe controls (needsApproval), but the presence of the transferFunds tool makes this a direct financial execution capability.