I Bypassed a "Secured" AI Agent in 62 Seconds — So I Built the Tool That Catches It

We scanned an AI agent that had keyword security filters. Crucible bypassed every single one using authority impersonation. Grade F. 217 vulnerabilities. Here's exactly what happened — and how to test