agentskillexchange Scan LLM systems for jailbreaks, prompt injections, and unsafe behaviors with garak
Probe a model or agent stack with adversarial test suites so safety failures show up before deployment or review.
Prerequisites
Python 3.10+, target LLM or API credentials, command line access
Installation
Use the upstream install or setup path that matches your environment:
Standard install with pip
- python -m pip install -U garak
- python -m pip install -U git+https://github.com/NVIDIA/garak.git@main
- conda create --name garak "python>=3.10,<=3.12"
Requirements and caveats from upstream:
- For testing. This always generates the empty string, using the test.Blank generator. Will be marked as failing for any tests that require an output, e.g. those that make contentious claims and expect the model to re...
Basic usage or getting-started notes:
garak is a command-line tool. It's developed in Linux and OSX.
Just grab it from PyPI and you should be good to go:
The standard pip version of garak is updated periodically. To get a fresher version from GitHub, try:
Source: https://github.com/NVIDIA/garak
Extracted from upstream docs: https://raw.githubusercontent.com/NVIDIA/garak/HEAD/README.md