About SkillGuard
An AI-powered security audit platform for autonomous agents and skills.
Important Disclaimer
SkillGuard is an independent project and is **not associated with Google**. We utilize Google's Gemini 3 Pro and Flash models for deep code analysis, but the badges and certifications are generated by our own algorithms.
AI models can make mistakes. A "Platinum" or "Obsidian" score does not guarantee 100% security. Always review code manually before deploying critical agents.
How It Works
1. Static Analysis
We scan the repository for sensitive files (API keys, .env), dangerous capabilities (shell access, network), and insecure coding patterns using regex and AST parsing.
2. Deep Audit (Gemini)
We select the most critical files and feed them to **Gemini 3 Pro**, asking it to identify logic flaws, prompt injection risks, and supply chain vulnerabilities that static tools miss.
The Badge System
Agents are assigned a tier based on their Safety Score, which combines static analysis results and AI-detected vulnerabilities.
Security Tiers
Gemini Certified Safe
The Gemini Certified Safe badge is awarded only to agents that achieve **Platinum, Diamond, or Obsidian** status.
This indicates that both our static analysis and the Gemini 3 Pro model found zero critical or high-risk vulnerabilities, and the agent follows security best practices.
Extreme Danger
Click to view details and understand why these patterns are dangerous.
The Extreme Danger badge triggers when our system detects active attempts to bypass security analysis or alter the underlying intentions of the agent.
Agents with this badge should be treated with extreme caution.