ai-orchestration-langchain
Fail
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: HIGHDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The 'calculator' tool in 'examples/agents.md' and 'examples/structured-output-tools.md' uses 'eval()' on user-supplied input to perform calculations. This allows for arbitrary JavaScript code execution in the agent environment.\n
- Evidence: 'return String(eval(expression));' in both files.\n- [COMMAND_EXECUTION]: The use of 'eval()' allows for executing JavaScript code that can perform unauthorized actions within the runtime environment.\n- [INDIRECT_PROMPT_INJECTION]: The RAG pipeline in 'examples/rag.md' processes external content from URLs and files, which is then interpolated into prompts.\n
- Ingestion points: 'CheerioWebBaseLoader' (web), 'TextLoader' (files), and 'PDFLoader' (PDFs) in 'examples/rag.md'.\n
- Boundary markers: The base LCEL RAG example in 'examples/rag.md' lacks delimiters or safety instructions, although a 'Security' section elsewhere provides a mitigated example.\n
- Capability inventory: The presence of the 'eval()'-based calculator tool increases the potential impact if malicious instructions are injected into the retrieved data.\n
- Sanitization: No sanitization or validation of the ingested external content is performed.
Recommendations
- AI detected serious security threats
Audit Metadata