Continue a response stopped by the output limit

Goal

Detect `max_tokens` termination and continue generation without duplicating or silently truncating the result.

Steps

1. Inspect stop_reason from client.messages.create; continue only when it equals max_tokens, not end_turn. [source09]
2. Append the assistant response and a continuation prompt to messages, then request the next segment with a bounded retry count. [source09]
3. Join segments and test boundary cases where output ends mid-sentence or exactly at a section boundary; surface a clear incomplete status on exhaustion. [source09]

How to verify

Force a short max_tokens limit; confirm continuation runs only after stop_reason=max_tokens, stops at end_turn, and reports incomplete when the retry cap is reached.

Troubleshooting

If a segment repeats or gets appended after end_turn, gate continuation on stop_reason and keep a bounded retry counter with an explicit incomplete result.