mcp-builder

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

• Third-party content exposure detected (medium risk: 0.30). The runtime evaluation harness reads outsider-authored free text from an XML evaluation file (question strings) and injects it directly into the LLM prompt via messages = [{"role":"user","content": question}] in scripts/evaluation.py, enabling indirect prompt injection against the MCP tools it then calls.