nemo-guardrails
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [External Downloads] (LOW): The skill installs the 'nemoguardrails' Python package via pip. Although NVIDIA is a well-known vendor, it is not on the specific trusted sources list provided. The severity is downgraded to LOW as the dependency is essential for the skill's primary safety purpose.
- [Prompt Injection] (LOW): The skill exposes an indirect prompt injection surface by processing user input to evaluate safety rules. * Ingestion points: User input in 'rails.generate' calls. * Boundary markers: Triple-quoted strings and message schemas are used in the examples. * Capability inventory: Calls LLM generation and internal Python actions for validation. * Sanitization: The skill itself serves as a sanitization and validation framework, reducing the risk of accidental obedience.
Audit Metadata