AI Pulse · Under the Radar

A four-day-old GitHub repo called Ratchet has picked up 429 stars and is climbing at 105 a day, which tells you something real is happening. Its entire purpose is to check whether an AI agent actually followed the rules you gave it in files like CLAUDE.md or AGENTS.md. Not whether it produced output. Whether it obeyed the constraints.

This matters because most teams building agents right now assume compliance. You write a prompt or a ruleset, the agent runs, you check the result. Ratchet flips that. It treats the agent like an employee you need to audit. Did it skip a step? Did it call an API you told it not to touch? Did it fabricate a citation? The repo is bare bones, but the idea is landing because people are hitting the same wall. Agents drift, hallucinate, or ignore instructions under load, and you don’t always catch it until a customer does.

This is turning into its own tooling category. Verify compliance, don’t just trust the output. It mirrors the dispatch gates and proof-of-done checks we build into systems like the Omni Command Centre, where every agent action logs what it was supposed to do and what it actually did. You need that audit trail when you’re running agents in production, not just prototyping in a notebook.

The speed of adoption, 105 stars a day for a repo with no marketing, suggests this isn’t a nice-to-have anymore. If you’re deploying agents that touch real workflows, real data, or real customers, you need a way to prove they did what you told them to do. Ratchet is one answer. The broader shift is that agent governance is no longer theoretical. It’s infrastructure.