[reonokiy]

Summary

Fix the stale SDK compatibility assertion left after #15: max_output_tokens=32 is accepted as a discarded compatibility hint, so the rejection case must use zero instead.

• Exercise null, 1, 32, and u64::MAX through the official Python SDK for standard and Lite models, with both streaming and non-streaming responses.
• Assert the complete fixture response, usage, and preserved extra field, including streamed text, rather than merely successful HTTP status. A hint of 1 must not truncate the response.
• Preserve existing zero, negative, fraction, and string rejection coverage; add boolean and unsigned-64-bit-overflow rejection cases.
• No gateway behavior or dependency changes.

Validation

• Passed: uv sync --locked --group dev (pinned OpenAI SDK 3.16.0).
• Passed: python -m compileall -q tests/sdk and git diff --check.
• Passed: pytest collection for the two changed modules with a two-model fixture configuration: 194 cases collected.
• Not run locally: Rust tests, gateway-backed SDK execution, official CLI fixtures, and source audit, because this execution environment has no Cargo/Rust toolchain. The existing PR workflow runs these checks; collection alone is not an execution pass.
• Live subscription E2E not run; this test-only change targets local fixtures.

The existing workflow may publish its normal pr-N test image after checks pass. No merge or production rollout is part of this change.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-10T09:35:54.384317Z	e717cd4	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.