prompt-guard
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFE
Full Analysis
- EXTERNAL_DOWNLOADS (SAFE): The skill downloads the 'meta-llama/Prompt-Guard-86M' model from Hugging Face. Since 'meta-llama' is a trusted organization, this download is considered safe for its intended purpose.
- REMOTE_CODE_EXECUTION (SAFE): Model inference is performed locally using the transformers and torch libraries. There is no evidence of piping remote content to a shell or using dangerous execution functions like eval on untrusted data.
- DATA_EXFILTRATION (SAFE): The skill does not access sensitive system files (like SSH keys or credentials) and does not perform network operations to untrusted third-party domains.
- PROMPT_INJECTION (SAFE): Although the skill contains example prompt injection strings, these are provided as test cases for the classification model and are not instructions intended to subvert the agent's behavior.
Audit Metadata