A couple of days ago, I was fixing a payment callback. I had just narrowed the problem down to three files in the main thread when I suddenly remembered something else: the tests weren’t written yet.
My old habit would have been to start a new Claude Code session and explain everything from scratch: the project structure, what I’d changed, which legacy logic must not be touched. Ten minutes later, the actual work still hadn’t started.
This time I tried something different. In the existing session, I typed:
/subtask Based on the current changes, write a set of unit tests for the payment callback
Then I kept working on the code in the main thread.
A few minutes later, the test suggestions came straight back into the main session. It knew which files we’d already looked at and where the boundaries of this change were. I didn’t have to explain the background again.
