Verdict: A Working Risk Classifier for Agentic Software Factories

Coding agents can now run for hours without a human reading a line of what they produce. This post is the design spec for Verdict, a small library I'm building that gives an agentic coding loop a stru