Your Agent SDK loop for the "automate repetitive tasks" feature sometimes stops halfway through a job. The loop ends as soon as a response contains a text block, but Claude often writes text such as "Next I'll run the migration script" in the same response as a tool_use block. How should the loop decide whether to continue?

Answer: D — Continue while stop_reason is "tool_use", appending each tool result to the conversation, and terminate when stop_reason is "end_turn".

The loop's control flow should follow stop_reason: run the requested tools and append their results while it is "tool_use", and stop at "end_turn". (a) and (b) parse natural-language signals to decide when to stop, which section 6 names as an anti-pattern, and a model can write such a phrase early or leave it out. (c) uses an arbitrary iteration cap as the main stopping mechanism, so short jobs keep running and long jobs get cut off.