Debug with an agent: reproduce, hypothesise, log, then fix

Anthropic's example: "users report that login fails after session timeout. check the auth flow in src/auth/, especially token refresh. write a failing test that reproduces the issue, then fix it." The docs also say to give the command that reproduces it and whether it's intermittent.

The debugging subagent's first rule is that it "Demonstrably reproduces the problem". A failing test or a repro command it can run counts; a plausible explanation doesn't.

Next it "Forms a testable hypothesis", "Designs an experiment using precise logging to test the hypothesis", and analyses the output "to validate, invalidate, or update the hypothesis." Another user simply puts the word "hypothesis" in their prompts.

"Only when the agent has proven the root cause is it allowed to attempt a fix." Anthropic adds: "address the root cause, don't suppress the error."

Anthropic's rule of thumb: "If you've corrected Claude more than twice on the same issue in one session, the context is cluttered with failed approaches. Run /clear and start fresh" with a better prompt.