Remove detached delivery from review/start (#57667)

## What changed

`review/start` now supports only inline reviews on the supplied thread and returns that thread's id as `reviewThreadId`. Clients can omit `delivery`, send `null`, or specify `"inline"`; `"detached"` is no longer supported.

For a separate review thread, call `thread/start` followed by `review/start`. To preserve conversation history with custom review instructions, use `thread/fork` followed by `turn/start`.

Remove detached review execution and its app-server dependency on `codex-agent-extension`, and update the protocol schemas and generated clients. Preserve `ReviewDelivery.inline` as a Python enum member.

## Testing

Add a Python regression test verifying that inline review delivery parses as `ReviewDelivery.inline` and round-trips to the original JSON.

GitOrigin-RevId: ed16c5271c3315ae3afcb3ee78883ed421394b4d

sdk/python/scripts/update_sdk_artifacts.py

@@ -547,6 +547,9 @@ def _normalized_schema_bundle_text(schema_dir: Path) -> str:
                 required.remove(field)
         for name, definition in definitions.items():
             if isinstance(definition, dict):
+                if name == "ReviewDelivery":
+                    # Preserve ReviewDelivery.inline when only one delivery mode remains.
+                    definition["x-enum-field-as-literal"] = False
                 if name == "CodexErrorInfo":
                     _normalize_open_ended_enum(definition)
                 _flatten_string_enum_one_of(definition)

sdk/python/src/openai_codex/generated/v2_all.py

@@ -4135,7 +4135,6 @@ class ResponsesApiWebSearchAction(
 
 class ReviewDelivery(Enum):
     inline = "inline"
-    detached = "detached"
 
 
 class UncommittedChangesReviewTarget(BaseModel):
@@ -9460,7 +9459,7 @@ class ReviewStartParams(BaseModel):
     delivery: Annotated[
         ReviewDelivery | None,
         Field(
-            description="Where to run the review: inline (default) on the current thread or detached on a new thread (returned in `reviewThreadId`). Detached delivery is deprecated and emits `deprecationNotice`. Use `thread/start` followed by an inline review for a separate review thread."
+            description="Reviews run inline on the supplied thread. Omit this field or pass `inline`. To review on a separate thread, start a thread first and use its id."
         ),
     ] = None
     target: ReviewTarget
@@ -12565,11 +12564,7 @@ class ReviewStartResponse(BaseModel):
         populate_by_name=True,
     )
     review_thread_id: Annotated[
-        str,
-        Field(
-            alias="reviewThreadId",
-            description="Identifies the thread where the review runs.\n\nFor inline reviews, this is the original thread id. For detached reviews, this is the id of the new review thread.",
-        ),
+        str, Field(alias="reviewThreadId", description="The thread id supplied to `review/start`.")
     ]
     turn: Turn
 

sdk/python/tests/test_client_rpc_methods.py

@@ -23,6 +23,8 @@ from openai_codex.generated.v2_all import (
     PlanType,
     ReasoningEffort,
     ReasoningEffortOption,
+    ReviewDelivery,
+    ReviewStartParams,
     ThreadForkParams,
     ThreadListParams,
     ThreadQueueChangedNotification,
@@ -39,6 +41,14 @@ from openai_codex.types import ThreadSource
 ROOT = Path(__file__).resolve().parents[1]
 
 
+def test_inline_review_delivery_round_trips_as_an_enum() -> None:
+    wire = {"threadId": "thread-1", "target": {"type": "uncommittedChanges"}, "delivery": "inline"}
+    params = ReviewStartParams.model_validate(wire)
+
+    assert params.delivery is ReviewDelivery.inline
+    assert params.model_dump(mode="json", by_alias=True) == wire
+
+
 @pytest.mark.parametrize(
     ("model", "fields"),
     [