red-teaming-llms-with-garak - Red-Teaming LLMs with garak
Runs NVIDIA garak probe suites against an LLM endpoint and interprets the resulting hit-rate report.
Tags
Updated: 2026-09-24Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Enumerate probes and detectors
- Configure target LLM endpoints
- Run security probe suites
- Interpret hit-rate report logs
- Re-test endpoints after applying mitigations
Inputs
- Target LLM endpoint
- API key
- Written testing authorization
- YAML or JSON configuration file
Outputs
- JSONL report log file
- HTML report file
- Terminal summary output
- Debug log file
- Vulnerability hit log file
Requirements
- Python 3.10+
- garak package