A Claude Code skill for NVIDIA garak - the LLM vulnerability scanner.
Garak is an open-source security testing framework for large language models. Think of it as nmap or Metasploit, but for AI systems. It probes whether LLMs can be manipulated into generating harmful content, leaking data, accepting prompt injections, or failing in other undesirable ways.
Add this skill to your Claude Code environment:
# Clone the skill
git clone https://github.com/evalops/garak-skill.git ~/.claude/skills/garak
# Or add via Claude Code plugin marketplace (if available)Once installed, the skill activates when you ask Claude to:
- "Test this LLM for security vulnerabilities"
- "Red team my model with garak"
- "Run jailbreak tests on GPT-4"
- "Check for prompt injection vulnerabilities"
- "Scan my model for data leakage"
- Complete CLI reference for garak
- 40+ probe categories organized by attack type
- Support for 23+ LLM platforms (OpenAI, Bedrock, HuggingFace, Ollama, etc.)
- Configuration file examples
- Best practices for security assessments
- Troubleshooting guides
- Python 3.10-3.12
- pip
- API keys for target LLM platforms
Apache-2.0
- NVIDIA garak - The underlying vulnerability scanner
- Anthropic Agent Skills - Skill format specification