feat: add on_turn_end callback for inter-turn state changes#3589
feat: add on_turn_end callback for inter-turn state changes#3589Oxygen56 wants to merge 1 commit into
Conversation
Adds on_turn_end callback to RunHooksBase, AgentHooksBase, and RunConfig, firing after each agent turn to allow state inspection and dynamic modification. The callback is invoked at all turn boundaries (handoffs and run-again steps) in both the standard and streaming execution paths. - RunHooksBase.on_turn_end: lifecycle hook for run-level turn completion - AgentHooksBase.on_turn_end: per-agent lifecycle hook for turn completion - RunConfig.on_turn_end: standalone callback via TurnEndData - TurnEndData: dataclass carrying agent, context, and turn number Closes openai#2671
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 6c17b31eac
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| For example, you can use this to add a system prompt to the input. | ||
| """ | ||
|
|
||
| on_turn_end: OnTurnEndCallback | None = None |
There was a problem hiding this comment.
Adding on_turn_end here shifts every existing public RunConfig positional argument after call_model_input_filter: callers that currently pass tool_error_formatter, session_settings, reasoning_item_id_policy, sandbox, or later fields positionally will now bind those values to the wrong field. This violates the repo's public API positional-compatibility rule for exported dataclasses and breaks existing patterns covered by tests/test_source_compat_constructors.py; append the new optional field after the current fields or add a compatibility layer.
Useful? React with 👍 / 👎.
|
This PR is stale because it has been open for 10 days with no activity. |
|
Thanks for working on this. I do not think we should continue with this approach. The released This callback only receives the agent and context, so it does not itself modify the next model input. The current implementation also shifts existing I am going to close this PR. |
2 participants
Adds
on_turn_endcallback for inspecting/modifying agent state between turns.Changes
RunHooksBaseandAgentHooksBase(lifecycle.py)on_turn_endlifecycle hook method that fires after each turn completesRunConfig(run_config.py)TurnEndDatadataclass carrying the current agent, run context, and turn numberOnTurnEndCallbacktype alias for the callback signatureon_turn_endfield toRunConfigas an optional callbackTurn loop wiring (run.py, run_internal/run_loop.py)
on_turn_endat all turn boundaries in both the standard and streaming execution paths:NextStepHandoff(after handoff, before next agent starts)NextStepRunAgain(after tool execution, before next model call)Use cases
Closes #2671