lsm / neokai / 25138392846

82%

push

github

[codex] Fix Responses bridge compatibility with ChatGPT Codex endpoint (#1720)

* [codex] Fix OpenAI Responses bridge compatibility with ChatGPT Codex endpoint

The ChatGPT Codex endpoint (chatgpt.com/backend-api/codex) rejects several
parameters that the official OpenAI Responses API accepts. Three fixes:

- Add `store: false` (required by the ChatGPT Codex endpoint)
- Introduce `chatgptCodexCompat` mode for oauth sessions: omit
  `max_output_tokens` and `parallel_tool_calls` which the endpoint rejects
- Retry with full history when the endpoint rejects `previous_response_id`
  (the continuation optimisation is silently skipped rather than erroring)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix: always consume continuation entries after previous_response_id fallback

When the fallback is triggered, continuation was set to undefined before
consumeContinuation ran, leaving stale call-id timer entries to expire by
TTL instead of being cleaned up immediately.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

7670 of 10003 branches covered (76.68%)

Branch coverage included in aggregate %.

14 of 31 new or added lines in 1 file covered. (45.16%)

58884 of 82825 relevant lines covered (71.09%)

272.19 hits per line