AI Agent Tests All Pass. The Bug Is Between Tasks.

Your AI coding agent splits a feature into tasks, and each task gets its own test. Task 2 checks Task 1, Task 4 checks Task 3, and they all pass.