Open-source supervisor that sits underneath AI coding agents at the OS level. It scores security-relevant syscalls against 18 filters before the kernel runs them: safe calls proceed, ambiguous ones freeze for review, dangerous ones are denied. MPL-2.0, local.
