langgraph
Fail
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: HIGHDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The code example for the 'Basic Agent Graph' implements a 'calculator' tool that utilizes the Python 'eval()' function to process input strings. This is a high-risk pattern that enables the execution of arbitrary Python code on the host machine.
- Evidence:
return str(eval(expression))inSKILL.md. - [INDIRECT_PROMPT_INJECTION]: The 'calculator' tool implementation creates a vulnerability surface for indirect prompt injection because it processes untrusted inputs through a dangerous function without any validation.
- Ingestion points: The
expressionparameter in thecalculator(expression: str)tool function defined inSKILL.md. - Boundary markers: Absent. There are no instructions or delimiters provided to prevent the LLM from executing malicious logic passed within the expression.
- Capability inventory: Arbitrary Python code execution via
eval(). - Sanitization: Absent. The input is passed directly to the execution function without filtering or escaping.
Recommendations
- AI detected serious security threats
Audit Metadata