Responses API streaming parity: missing response.output_text.done event

[robinnarsinghranabhat]

Description

The LlamaStack server does not emit the response.output_text.done streaming event which doesn't adhere to official OpenAI Responses behaviour in their gpt models.

The streaming event sequence for text output should be:

output_item.added → content_part.added → output_text.delta (xN) → output_text.done → content_part.done → output_item.done

LlamaStack currently skips output_text.done — the event type is defined in openai_responses.py but never emitted in streaming.py.

How to reproduce

from openai import OpenAI

client = OpenAI(base_url="http://127.0.0.1:8321/v1", api_key="fake")
events = list(client.responses.create(
    model="ollama/gpt-oss:20b",  # or any registered model
    input="What is 2 + 2?",
    stream=True,
))

types = [e.type for e in events]
print("response.output_text.done" in types)  # False — should be True

Expected vs actual

Expected (matches OpenAI ground truth with gpt-5.1):

...
response.output_text.delta (xN)
response.output_text.done          ← contains final accumulated text
response.content_part.done
...

Actual (LlamaStack server):

...
response.output_text.delta (xN)
                                   ← output_text.done MISSING
response.content_part.done
...

Impact

Apart from divergence from official responses behaviour, Clients relying on response.output_text.done as a signal to capture the final accumulated text (without manually accumulating deltas) will not receive it.