nvidia/garak ? reverse-engineered prompt

Build me a Python command line tool called garak that can test an AI model for common security and safety problems like prompt injection, jailbreaks, hallucinations, data leakage, toxicity, and misinformation.

I want to point it at different kinds of models, including Hugging Face, OpenAI, Bedrock, LiteLLM, Replicate, and anything I can reach through a REST endpoint, then run a set of probes and tell me which ones fail. It should show progress while it runs, print a clear summary at the end, and save detailed logs plus a report file I can inspect later.

Make it easy to install and run from the terminal, with sensible defaults so if I just give it a target it can scan it right away. Include a way to choose specific probe groups or a single probe when I want to focus on one test. If you need to check current library docs or examples online while building it, go ahead.