agentguard

AgentGuard is a security tool for developers deploying AI agents, acting as a runtime guard that blocks malicious skills, prevents data leaks, and protects secrets using 24 detection rules and a trust registry. It is designed for teams building autonomous AI systems who need to enforce safety policies in real time. The project is interesting because it addresses the emerging vulnerability of agent-based workflows, where a single compromised skill or prompt injection could expose sensitive data or hijack system actions.