Handoff breaking changes in v0.6.0 break compatibility with non-OpenAI models

[dast-draftwise]

Hi,

I recently upgraded openai-agents from a version prior to v0.6.0 to the latest release and ran into issues caused by the handoff breaking changes introduced here:
https://github.com/openai/openai-agents-python/releases/tag/v0.6.0

While the new handoff mechanism works well with OpenAI models that are aware of and can properly interpret handoffs, it breaks existing agent workflows when using non-OpenAI models (e.g., Claude). These models don’t natively understand the new handoff semantics, which means handoffs no longer work without a manual workaround.

Is there any plan to support:

• an opt-in flag, compatibility mode, or alternative API that preserves the pre-0.6.0 handoff behavior for non-OpenAI models, or
• a first-class abstraction that allows non-OpenAI models to participate in handoffs without requiring manual intervention?

Without this, the handoff feature effectively becomes OpenAI-model-only, which makes it difficult to maintain multi-model agent workflows and gives the impression that non-OpenAI support may gradually degrade over time.

I’d appreciate clarification on the intended direction here and whether backward-compatible or model-agnostic handoff support is something you’re considering.

Thanks!

[ihower]

@dast-draftwise You can manually disable the new nested handoff behavior by setting nest_handoff_history=False to restore the previous behavior.

This was already discussed and decided in #2211. In the upcoming v0.7.0 release, nest_handoff_history will default to False (see PR #2272 for details).

[seratch]

As @ihower mentioned, please set nest_handoff_history=False for now. The upcoming v0.7.0 will switch the default behavior back to the one of v0.5.x.

[dast-draftwise]

Thank you for the feedback - it solved it.
Though to be fair, given how impactful this is to many people, it shouldnt been enabled in the first place.
Happy to see that will be disabled by default.