stop_reason, tool_use, tool_result — the loop every agent runs

AI wrote an agent loop that checks stop_reason == "tool_use" to CONTINUE, but never checks for end_turn to STOP. Without an iteration cap, this would run forever in real life — here, the fake model returns end_turn after one round, and the loop spins past it because the break condition is wrong. Fix line 13 so the loop exits on end_turn.

Expected output:

done