prompt-engineering-patterns
Fail
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: HIGHDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The execute_tool function in SKILL.md uses the eval() function to execute a math expression. The input for this expression is generated by an LLM based on user queries, which allows for the execution of arbitrary Python code if the LLM is manipulated.
- [INDIRECT_PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection attacks.
- Ingestion points: user_query in the react_agent function in SKILL.md.
- Boundary markers: The ReAct agent pattern does not utilize boundary markers to isolate user input from the rest of the prompt context.
- Capability inventory: The skill includes an eval() call in SKILL.md capable of executing arbitrary code.
- Sanitization: No input sanitization or validation is applied to the tool input before it is passed to eval() in the ReAct example.
Recommendations
- AI detected serious security threats
Audit Metadata