ai-scanner-garak

What This Skill Does

The ai-scanner-garak skill integrates the NVIDIA Garak LLM vulnerability scanner into the OpenClaw ecosystem. It provides an automated, multi-tenant platform for testing AI models against 179 distinct security probes across 35 vulnerability families, including the OWASP LLM Top 10. By wrapping the powerful Garak engine in a Ruby on Rails framework, this skill enables security teams and developers to perform systematic security assessments, generate PDF reports, and export logs to SIEM systems like Splunk or Syslog, ensuring continuous compliance and model hardening.

Installation

To install this skill, run clawhub install openclaw/skills/skills/adisinghstudent/ai-scanner-garak in your terminal. Ensure your environment has Docker and Docker Compose installed, as the skill utilizes a containerized architecture for isolated scan execution. Post-installation, you must configure your .env file with SECRET_KEY_BASE and POSTGRES_PASSWORD. Use docker compose up -d to spin up the dashboard, which is accessible at http://localhost. Remember to rotate the default administrative credentials immediately after the first successful launch.

Use Cases

This skill is ideal for: 1) Red-teaming AI models prior to production deployment to identify prompt injection vulnerabilities. 2) Routine security auditing for LLM-based applications to maintain compliance with safety standards. 3) SIEM-integrated monitoring where AI security events need to be correlated with broader network security logs. 4) Multi-tenant security environments where different development teams need to scope their scan results by organization.

Example Prompts

1. "Perform a security audit on the target 'Production GPT-4' using all default probe families and notify me when the PDF report is ready."
2. "Scan the 'Internal Chatbot UI' target specifically for prompt injection and data leakage vulnerabilities, then export the logs to our Splunk endpoint."
3. "List all current targets in the system and calculate the average Attack Success Rate (ASR) across the last five scan jobs."

Tips & Limitations

• Tip: Always ensure you have appropriate authorization before scanning external API endpoints to avoid triggering rate limits or security blocks.
• Tip: Utilize the Rails console for complex task scheduling or bulk target creation if the UI becomes a bottleneck.
• Limitation: Browser-based UI targets require careful configuration to ensure the scanner can reliably interact with non-standard DOM elements.
• Limitation: The scanner is resource-intensive; ensure your host server has sufficient memory and CPU allocation when running large, multi-probe scan sets.