add asyncresume() and syncresume_sync() helpers onMaxTurnsExceeded that replay one final turn with an optional user prompt, reusing the prior agent, context, and a copy of the originalRunConfig
extendRunResult andRunErrorDetails to retain the effective RunConfig, so guardrail/tool settings, tracing metadata, and other knobs survive into a resume attempt
ensure resume turns disable further tool calls by forcingtool_choice="none" while still respecting any other model settings (e.g., temperature) that the caller supplied
cover the new helpers with unit tests for async/sync flows, verify model-settings preservation, and document the flow via aexamples/basic/max_turns_resume.py sample that demonstrates recovering after a hard turn limit

thomas-lab-openai added2 commits

November 4, 2025 15:08

draft

518914a

updated naming & example

931336a

thomas-lab-openai requested a review fromrm-openai

November 6, 2025 17:20

thomas-lab-openai added2 commits

November 6, 2025 18:27

lint

09113b6

fix typecheck

77ee3eb

seratch added enhancement

New feature or request

feature:core labels

Nov 6, 2025

seratch mentioned this pull request

Nov 7, 2025

Can this generate answer after max turns? Is there any settings to this?#2054

Open

seratch marked this pull request as draft

November 14, 2025 05:31

Copy link

Member

seratch commentedNov 14, 2025

You’ve assigned@rm-openai as a reviewer, so he may come up with other suggestions, but personally, I feel adding the methods to the exception class this way may not be a clean design. I would add some kind of helper that handles the max exceed error and resumes.

Aside from the implementation details, I’m not yet sure whether it helps when the max turns are exceeded. Even if you resume it, it might hit the max turns again.