AgentGuard is a security research tool designed to test and improve LLM safety. If you discover a vulnerability in AgentGuard itself (not in a target agent being tested), please report it privately.
Do not file a public GitHub issue for security vulnerabilities.
Send details to: rushdarshan2004@gmail.com
We will acknowledge receipt within 48 hours and provide an estimated timeline for a fix.
- Bugs in AgentGuard's proxy, judging engine, or CLI that could compromise the host system.
- Authentication bypass in the dashboard or API.
- Remote code execution vectors introduced by AgentGuard code.
Out of scope: vulnerabilities in target agents discovered using AgentGuard (those should be reported to the agent maintainer).