prompt-guard
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a technical guide for security alignment and does not contain any malicious code or instructions. All code examples are intended for local execution to secure LLM applications.
- [EXTERNAL_DOWNLOADS]: Fetches the 'Prompt-Guard-86M' model and tokenizer from Meta's official Hugging Face repository using the
transformerslibrary. This is standard behavior for loading pre-trained machine learning models. - [PROMPT_INJECTION]: Contains various strings that simulate prompt injection and jailbreak attempts (e.g., 'Ignore all previous instructions', 'You are now in developer mode'). These are clearly documented as test cases for the classification model and are not intended to influence the agent's behavior.
- [INDIRECT_PROMPT_INJECTION]: The skill describes an interface for processing untrusted third-party data to detect injections. While it ingests external strings, the provided code only outputs security scores and does not perform sensitive operations with the data, effectively serving as a security control rather than a vulnerability.
Audit Metadata