SDK leaks internal reasoning into client-facing response (adds response_reasoning: inside JSON output)

[mubashirdev]

The OpenAI Agents TypeScript SDK is returning internal reasoning inside the client-facing response field.

Instead of keeping reasoning separate, the SDK appends text like response_reasoning: directly into the response string, which is shown to end users.

Expected
• response contains only user-facing content.
• Internal reasoning is either:
• excluded, or
• returned in a separate non-user-facing field.

Actual
{ "response": "User-facing message...\nresponse_reasoning:\ninternal reasoning text" }

Impact
• Leaks internal/system reasoning to end users
• Breaks structured/JSON responses
• Forces client-side sanitization

[seratch]

Thank you for reporting this. We have received a few similar pieces of feedback about the response text, and it seems to be an issue on the model endpoint side rather than this SDKs logic. We will continue coordinating with the relevant teams and address anything needed on this SDK side if required.