red-teaming-llms-with-garak
Pass
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The Python script
scripts/agent.pyinvokes thegarakCLI tool usingsubprocess.run(). The command is passed as a list of arguments rather than a concatenated string, which effectively mitigates the risk of shell injection attacks. - [EXTERNAL_DOWNLOADS]: The instructions direct users to install the
garakpackage from PyPI or directly from NVIDIA's official GitHub repository. These are reputable and well-known sources appropriate for the skill's purpose. - [PROMPT_INJECTION]: Static analysis flagged terms such as 'DAN' and 'jailbreak' in
SKILL.md. However, these are descriptive names for the security probes used by the red-teaming tool and do not constitute an attempt to inject malicious instructions into the agent. - [DATA_EXPOSURE]: The skill documents the use of environment variables (e.g.,
OPENAI_API_KEY) for providing credentials to the scanner, which is a secure and standard practice for managing sensitive API keys.
Audit Metadata