Leandro: AI SRE Agent with Prompt Injection Sandbox

Leandro is an open-source AI SRE agent that watches Kubernetes clusters, diagnoses pod incidents autonomously, and communicates findings over Google Chat. Built with a security-first architecture, it runs inside a NixOS VM on libvirt/KVM with default-deny egress via nftables, a tinyproxy allowlist for outbound HTTP, and a hard tool denylist specifically designed to block prompt injection attacks. Since the agent reads attacker-influenceable text like pod logs and cluster events by design, every capability is fenced by an explicit, auditable boundary, making it a practical reference implementation for secure AI agent deployment in production infrastructure.

#AISecurity #Kubernetes #PromptInjection #SRE #ApplicationSecurity

https://github.com/SoulKyu/leandro