/v1/responses streaming omits response.content_part.done and sends empty output_text.done.text

[mikemolinet]

What happened?

POST /v1/responses streaming violates the OpenAI Responses API SSE lifecycle spec (https://platform.openai.com/docs/api-reference/responses-streaming) in two ways:

1. response.content_part.done is never emitted. Per the spec, the event sequence for a text content part is content_part.added → output_text.delta* → output_text.done → content_part.done → output_item.done. The current implementation skips content_part.done.
2. response.output_text.done.text is always "". The spec requires this event to carry the final accumulated content, but index.js:956-964 hardcodes text: "".

Both bugs live in the same SSE emitter at index.js:955-988.

Steps to reproduce

1. Run the proxy (npm start).
2. Send a streaming Responses API request: POST /v1/responses with stream: true.
3. Collect the emitted SSE events.
4. Observe: (a) no response.content_part.done event appears, and (b) response.output_text.done has text: "" regardless of actual output.

Expected behaviour

The SSE stream should match the spec: a response.content_part.done event is emitted between output_text.done and output_item.done, with part.text equal to the accumulated output text. response.output_text.done.text should also be the accumulated content.

Request / response (if applicable)

curl -N -X POST http://127.0.0.1:4010/v1/responses \
  -H 'content-type: application/json' \
  -d '{"model":"gpt-4o","stream":true,"input":"Say hello."}'

Expected in the stream (abbreviated):

event: response.output_text.delta
data: {"type":"response.output_text.delta", ..., "delta":"Hello"}

event: response.output_text.done
data: {"type":"response.output_text.done", ..., "text":"Hello"}            # currently text:""

event: response.content_part.done                                          # currently MISSING
data: {"type":"response.content_part.done", ..., "part":{"type":"output_text","text":"Hello","annotations":[]}}

event: response.output_item.done
data: {"type":"response.output_item.done", ...}

opencode-llm-proxy version

1.6.1

Runtime and OS

Reproduces on Node.js >= 20 and Bun >= 1.0, any OS — pure JS logic (SSE event sequence).

Provider / model

Any. Bug is at the proxy's SSE emission layer, independent of downstream provider.

---

I have a fix ready — accumulate delta tokens at the streaming call site, populate output_text.done.text, and emit the missing content_part.done event with the accumulated text. Happy to open a PR once the approach is confirmed.