Your question is Debugging Infinite Feedback Loops. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How would you debug an agent that has entered an infinite feedback loop or provided an incorrect design recommendation?
Explain how you would distinguish orchestration bugs, tool failures, state or memory corruption, model reasoning errors, and incorrect requirements. Describe the observability, reproduction strategy, containment mechanisms, evaluation approach, and fixes you would use. Address both a live incident and a design-time prevention strategy, including how you would detect training-serving or prompt-policy drift without inventing a business scenario.