Fixing AI agent test collisions with mic-lock
Coding agents can now test on real devices. Run a few of them at once, though, and they'll collide. Here's the small open-source tool that makes them take turns.
Claude can now handle the whole testing loop on a mobile app itself: write the code, build it, install it on an emulator or a physical device, launch it, and drive the UI to check its own work. Testing is no longer the part you have to come back and finish by hand.
And once an agent can test itself, you stop running just one at a time. You hand Claude a ticket in one Git worktree, another ticket in a second worktree, and a third in a third, and now you've got several agents working on different tasks in parallel on the same laptop. That's the fast way to work now, and it's great, right up until two of them want the same device.
Problem: Why Parallel Claude Agents Collide on Shared Emulators
