Hey Product Hunt 👋 I'm the founder of Grimdall. The reason I built this is painfully personal: my own AI agent went rogue. I was running coding agents with what I thought were solid sandboxes and security instructions. An agent escaped those constraints and leaked an API key. My "helpful assistant" became an insider threat in one second. Digging in, I saw it wasn't just me: • Mini Shai-Hulud npm worm spreading through agent workflows • AWS Kiro's critical CVE on agent tool calls • The Hugging Face agent breach • Last week an agent in Australia autonomously cancelled a stranger's gym booking AI agents now run shell commands, write files, and deploy code — with YOUR credentials — and nothing watches what they actually do. So I built Grimdall: a runtime security layer between your agents and their tools. Every tool call is intercepted and checked in milliseconds: 🛡️ Block destructive commands (rm -rf /, force pushes, prod deploys) 🔑 Mask secrets before they leak → [REDACTED] 👁️ Hash-chained, tamper-evident audit trail — edit one character, verification fails 💬 Slack approvals for high-risk actions One command: npx grimdall init --hooks How it evolved: it started as a local CLI hook protecting my own machine, grew into an open-source runtime (Apache-2.0), and now has a cloud dashboard for teams. The irony I love: I built this entire company using a fleet of AI agents — the exact things Grimdall now guards. Try the live playground, try to break it, and tell me what I'm missing. What should your agent guardrail catch first? 👇