Responses API bridge returns reasoning output items without summary and with output_text content

Finding: In an external report against LiteLLM v1.106.0, reasoning output items returned by the Responses API chat-completions bridge fail openai SDK ResponseReasoningItem validation, because summary is missing and content parts are typed output_text rather than reasoning_text (https://github.com/BerriAI/litellm/issues/45552). Reported environment: LiteLLM v1.106.0, Docker deployment, with the repro calling litellm.responses() directly. Models in the repro were anthropic/claude-sonnet-5-5 and deepseek/deepseek-v4-pro, with and without reasoning effort high (issue body). Reported symptom: the reasoning item in output has no summary field. The reporter says both the OpenAI Responses API and the Open Responses spec require summary, so an item with nothing to summarize should carry an empty array. The reporter also says the item has role and phase, which belong only on message items, and that content parts use output_text. Validation errors listed were summary: Field required and content.0.type expecting reasoning_text. In the third case (Claude with no reasoning parameter), the thinking block was empty, so the item had empty content and only the summary error appeared (issue body). Suspected location: the reporter believes the item is built with GenericResponseOutputItem, a message item type, in _extract_reasoning_output_items in litellm/responses/litellm_completion_transformation/transformation.py, and that streaming is affected because the final response.completed event is built the same way. This is the reporter's hypothesis, not a maintainer-confirmed cause. A commenter without maintainer association repeated this explanation and said they would emit a spec-shaped item with a regression test (https://github.com/BerriAI/litellm/issues/45552#issuecomment-6080672057). Fix status: the same commenter later said a fix is up as #45608 (https://github.com/BerriAI/litellm/issues/45552#issuecomment-6084853206). The PR was not supplied or inspected here. Its merge and any release containing it are unknown, and the issue is still open in the supplied data. Practical implication: clients that strictly parse bridged reasoning items with the openai SDK ResponseReasoningItem model may hit validation errors on this version. No workaround is given in the source. Limitations: this is an external report only and was not reproduced by KFA. Whether other versions are affected is unknown. Maintainer confirmation is absent.