fix(sse): default reasoning summary for effort-only Responses requests

[rushsinging]

Problem

A Chat-Completions client (e.g. an OpenAI-format coding agent) can only express reasoning through the top-level reasoning_effort hint — it has no way to request a reasoning summary. When such a request is routed to an openai-compatible provider node running with apiType: responses, openaiToOpenAIResponsesRequest promotes reasoning_effort to the Responses API's reasoning.effort but never sets a summary.

As a result the Responses upstream returns an empty reasoning summary, so the downstream chat client sees no thinking stream at all (only encrypted reasoning that never surfaces). The generic DefaultExecutor path lacks the summary/include injection that the Codex executor already performs via ensureCodexReasoningSummary.

Fix

In openaiToOpenAIResponsesRequest, on the effort-only path, default:

• reasoning.summary: "auto"
• include: ["reasoning.encrypted_content"]

so the summary actually streams back to the client. This generalizes what the Codex executor already does to any openai-compatible → Responses translation.

Scope is deliberately narrow:

• Only the effort-only path (a chat client that sent top-level reasoning_effort). An explicit reasoning object from a Responses-shaped client is preserved untouched.
• reasoning_effort: "none" is left without a summary.
• A client-provided include is preserved (idempotent).

Tests

tests/unit/translator-openai-responses-req.test.ts and tests/unit/openai-responses-reasoning-effort.test.ts:

• effort-only request → reasoning: { effort, summary: "auto" } + include: ["reasoning.encrypted_content"]
• reasoning_effort: "none" → no summary, no include
• existing "preserves explicit reasoning objects" and "does not inject include when caller did not set one" behavior stays unchanged

All 53 tests in the two suites pass; typecheck:core is clean; lint / any-budget gates pass.

[gemini-code-assist]

Code Review

This pull request updates the translator logic to default a reasoning summary (summary: "auto") and include the encrypted reasoning content (reasoning.encrypted_content) when converting effort-only Chat Completion requests to Responses API requests. This ensures that the upstream streams thinking back to the chat client. It also updates existing tests and adds a new test to verify that a reasoning summary is not defaulted when reasoning_effort is set to "none". There are no review comments, and I have no feedback to provide.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[diegosouzapw]

Thanks for the clean, narrowly-scoped fix — verified locally: all 53 tests in the two touched suites pass, and reverting just the translator change makes 2 of them fail exactly as expected (empty summary), confirming the test genuinely proves the bug. Mirrors the Codex executor's ensureCodexReasoningSummary pattern well. No changes requested; this looks merge-ready.