fix: close Open Responses conformance gaps in schema, vision input, a…

[alez007]

…nd SSE termination

Vision content parts (input_image) in Responses message input were silently dropped in translation to chat messages; adds required ResponseObject fields (completed_at, truncation, service_tier, etc.) with spec-correct defaults, backfills echoed tools/text/temperature/top_p to match OpenAI's effective values, omits encrypted_content when unset, and emits the SSE [DONE] sentinel on clean stream completion.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[Copilot]

Pull request overview

Closes OpenAI Responses API conformance gaps by aligning response schema defaults and streaming termination semantics, and by ensuring vision (input_image) parts in Responses message input are preserved when translating to chat messages.

Changes:

• Add spec-aligned ResponseObject fields/defaults (e.g., completed_at, penalties, truncation/service tier) and ensure completed_at is only set on completed terminal responses.
• Preserve vision content parts during Responses→Chat translation and reject unsupported content-part types.
• Emit the SSE data: [DONE]\n\n sentinel on clean streaming completion and extend tests to cover these behaviors.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/test_vllm_responses.py	Asserts [DONE] sentinel on clean stream completion and absence on failed streams.
tests/test_responses_utils.py	Adds coverage for build_response_from_parsed and completed_at behavior.
tests/test_responses_streaming.py	Validates completed_at presence/absence across streaming event envelopes and incomplete termination.
tests/test_responses_adapter.py	Adds coverage for image content translation, required response fields/defaults, echoed tool backfilling, and encrypted_content omission.
modelship/openai/utils/responses.py	Sets completed_at for non-streaming completed responses.
modelship/openai/protocol/responses/streaming.py	Sets completed_at only on terminal completed streaming envelope.
modelship/openai/protocol/responses/schemas.py	Adds schema fields/defaults and omits encrypted_content when unset via serializer.
modelship/openai/protocol/responses/adapter.py	Preserves image parts in message content, backfills effective defaults in response echo, and normalizes echoed tools.
modelship/infer/base_infer.py	Emits SSE [DONE] sentinel after clean Responses stream completion.

Comments suppressed due to low confidence (1)

modelship/openai/protocol/responses/adapter.py:212

• For text parts, _content_to_chat appends {"type": "text", "text": p.get("text")} without validating that text is a string. Because the request schema doesn’t type-check input items, a non-string text can make the later "".join(...) raise TypeError, turning a bad request into a 500. Validate (or skip) non-string text and raise UnsupportedResponsesFeatureError instead.

        ptype = p.get("type")
        if ptype in _CONTENT_TEXT_TYPES:
            parts.append({"type": "text", "text": p.get("text")})
        elif ptype in _CONTENT_IMAGE_TYPES:

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.