nemotron-policy-generator
Pass
Audited by Gen Agent Trust Hub on Jun 23, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill contains specific countermeasures against injection. It includes a 'non-negotiable floor' in its Operating Principles that instructs the agent to refuse any user attempts to bypass or weaken safety categories related to the sexualization of minors (S7), even if explicitly requested in the prompt or within an attached policy file.
- [DATA_EXFILTRATION]: No unauthorized data exposure or exfiltration patterns were found. The skill workflow explicitly includes a step to scrub personal identifying information (PII) from existing policies when performing extensions or updates.
- [REMOTE_CODE_EXECUTION]: The skill does not perform any remote code execution. All assets, including the standalone HTML policy generator GUI, are self-contained and do not fetch or execute external scripts.
- [COMMAND_EXECUTION]: There are no shell commands or system-level operations initiated by the skill. It primarily focuses on text processing and file generation for policy artifacts.
- [SAFE]: All external references, such as links to the HuggingFace model cards and datasets, point to official and trusted NVIDIA repositories.
Audit Metadata