mcp-builder

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). The evaluation harness intentionally forwards MCP tool definitions, tool inputs/outputs and server responses to the external OpenAI API (and by design asks the model to include detailed summaries of tool inputs/outputs), enabling easy exfiltration of sensitive data (including secrets passed via env/subprocess or returned by tools) to a third-party LLM.