Runtime security enforcement for AI coding agents — policy hooks, audit receipts, threat hunting, and 15 MCP security tools.