Reasoning Techniques
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [Indirect Prompt Injection] (LOW): The Python implementation pattern for Chain-of-Thought prompting demonstrates a direct interpolation surface for untrusted data. 1. Ingestion points: 'question' variable in the code block within SKILL.md. 2. Boundary markers: Absent in the example prompt template. 3. Capability inventory: The snippet invokes 'llm.generate' within SKILL.md. 4. Sanitization: No validation or escaping is implemented for the interpolated content.
Audit Metadata