PRODUCTION SCENARIO
A service must hand its caller a complete summary. A Claude response comes back with stop_reason max_tokens, and the text ends midway through the second paragraph.
How should the service handle this response?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Raise max_tokens and re-request, or continue from where the text stopped
max_tokens means generation hit the output ceiling, so the documented fix is a larger ceiling or a continuation request. A truncation notice is the right pattern when a caller tolerates partial output, and this one does not. Echoing the assistant content back unchanged is the handling for pause_turn, where a server-tool loop hit its iteration limit. Shortening the requested summary changes the deliverable and leaves this response still truncated.