AI Pulse · Under the Radar

Two developers who have never met shipped the same tool on the same day. Both ratchet and scopey went live on July 31st to solve a problem most people running AI coding agents have hit but few have named out loud: the agent starts refactoring your authentication layer when you asked it to fix a button.

This is scope drift. You give Claude Code or Cursor a narrow task, and three minutes later it is rewriting files you did not mention. The old fix was to catch it in review, roll back, and try again. These tools hook into the agent session in real time and stop the drift before it writes code. If the agent tries to touch a file outside the boundary you set, the tool blocks it and logs the attempt.

The fact that two builders, working independently, converged on the same enforcement pattern on the same day is the real signal here. ratchet picked up 406 stars and is adding 380 a day. scopey has 88 stars and 49 a day. Neither repo is polished. Both are proof that the problem is common enough that people want a fix now, not later.

What this means if you run agents in production

If you are using AI to write or refactor code at any scale, you have probably lost an hour to scope drift. These tools are early, but the pattern matters. Real-time enforcement is starting to look like table stakes for any system that lets an agent touch a codebase without a human watching every line. This is exactly the kind of guardrail we build into the Omni Command Centre, where you need agents to stay inside defined boundaries across multiple workflows, not just code.

Neither tool is production-ready yet, but the convergence tells you the category is real. If you are running agents that touch anything important, scope drift is not a quirk. It is a risk you can now measure and block. More on scopey here.