diff options
Diffstat (limited to 'docs')
| -rw-r--r-- | docs/oauth-provider-auth.md | 13 |
1 files changed, 11 insertions, 2 deletions
diff --git a/docs/oauth-provider-auth.md b/docs/oauth-provider-auth.md index a29f13b..9f5c7fe 100644 --- a/docs/oauth-provider-auth.md +++ b/docs/oauth-provider-auth.md @@ -554,8 +554,17 @@ it is the OpenAI **Responses API** at items, `reasoning`, `include: ["reasoning.encrypted_content"]`, `store: false`; headers `Authorization: Bearer`, `chatgpt-account-id`, `OpenAI-Beta: responses=experimental`, `originator: codex_cli_rs`). That is a -new wire dialect, implemented as the `openai_responses` provider style (see -below), distinct from `openai_chat`. +new wire dialect, implemented as the **`openai_responses` provider style** +(`provider_openai_responses.zig` + `openai_responses_json.zig`), distinct from +`openai_chat`. The default config ships a commented `[providers.codex]` / +`[auth.openai_codex]` example. The serializer maps system→`instructions`, +user/assistant text→message items, tool calls→`function_call` / +`function_call_output` items, and tools→flat function entries; the stream +parser handles `response.output_text.delta`, +`response.reasoning_summary_text.delta`, `response.output_item.added/done` + +`response.function_call_arguments.delta`, and `response.completed` (usage). +Known limitation: encrypted reasoning items are requested via `include` but not +persisted across turns, so multi-turn reasoning continuity is approximate. ### Verification status |