langgraph-supervisor

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

• Third-party content exposure detected (high risk: 1.00). The skill ingests arbitrary user/URL content (e.g., the API endpoint analyze_content in backend/app/api/v1/analysis.py accepts url and content) and multiple agent nodes (e.g., security_agent_node and the create_agent_node implementations) directly insert state["raw_content"] into LLM prompts (SECURITY_AGENT_PROMPT.format(content=state["raw_content"][:5000]) and analyze_with_agent), so it clearly processes untrusted third‑party content that could carry indirect prompt-injection.