openai-agents-setup

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

• Potentially malicious external URL detected (high risk: 0.80). The skill connects at runtime to an MCP server via the MCP_SERVER_URL (e.g., http://localhost:8001 or http://localhost:8001/mcp), and uses MCPServerStreamableHttp to discover and invoke tools on that server—meaning remote code (tools) is executed and can directly affect agent behavior and is a required dependency.