Skip to content

Agent Zero AI framework

Fix Codex OAuth utility responses

Alessandrof6dd6128d9ad
diff --git a/plugins/_oauth/AGENTS.md b/plugins/_oauth/AGENTS.md
index cdb3f9d3e..0a26286e1 100644
--- a/plugins/_oauth/AGENTS.md
+++ b/plugins/_oauth/AGENTS.md
@@ -47,6 +47,7 @@
- Codex Responses proxy requests must translate the legacy top-level `reasoning_effort` field to `reasoning.effort`; an explicit native `reasoning` field takes precedence.
- Codex Responses proxy defaults for reasoning effort, reasoning summary, and text verbosity come from the `codex` plugin config; explicit native request values take precedence.
- Codex request shaping tightens an already-advertised native `response` tool to a strict required `text` schema; it must not add tools omitted by the framework tool policy.
+- Non-streaming Codex proxy responses must retain completed SSE output items when the final `response.completed` envelope omits them.
- OAuth providers without upstream Responses support must set `a0_api_mode: chat`; native Responses providers rely on the default, since a local proxy route alone does not prove upstream support.
## Work Guidance
diff --git a/plugins/_oauth/helpers/codex.py b/plugins/_oauth/helpers/codex.py
index 9559f8f90..dfb5b67ef 100644
--- a/plugins/_oauth/helpers/codex.py
+++ b/plugins/_oauth/helpers/codex.py
@@ -794,6 +794,7 @@ def collect_completed_response(response: requests.Response) -> dict[str, Any]:
latest_error: Any = None
text_pieces: list[str] = []
latest_usage: dict[str, Any] | None = None
+ completed_items: dict[int, dict[str, Any]] = {}
for event in iter_sse_events(response):
data = event.get("data")
if not data:
@@ -808,6 +809,11 @@ def collect_completed_response(response: requests.Response) -> dict[str, Any]:
latest_error = parsed
continue
text_pieces.extend(extract_sse_text_deltas(parsed, event.get("event", "")))
+ if (parsed.get("type") or event.get("event")) == "response.output_item.done":
+ output_index = parsed.get("output_index")
+ item = parsed.get("item")
+ if isinstance(output_index, int) and isinstance(item, dict):
+ completed_items[output_index] = item
usage = parsed.get("usage")
if isinstance(usage, dict):
latest_usage = usage
@@ -815,6 +821,16 @@ def collect_completed_response(response: requests.Response) -> dict[str, Any]:
if isinstance(candidate, dict):
latest_response = candidate
+ if (
+ latest_response is not None
+ and completed_items
+ and not latest_response.get("output")
+ ):
+ latest_response = dict(latest_response)
+ latest_response["output"] = [
+ completed_items[index] for index in sorted(completed_items)
+ ]
+
if text_pieces:
text = "".join(text_pieces)
if latest_response is not None:
diff --git a/tests/test_oauth_codex.py b/tests/test_oauth_codex.py
index 9df1576be..56b98993f 100644
--- a/tests/test_oauth_codex.py
+++ b/tests/test_oauth_codex.py
@@ -521,6 +521,42 @@ def test_extract_sse_text_deltas_ignores_final_done_text():
)
+def test_collect_completed_response_restores_native_output_items():
+ item = {
+ "id": "msg_1",
+ "type": "message",
+ "status": "completed",
+ "content": [
+ {
+ "type": "output_text",
+ "annotations": [],
+ "logprobs": [],
+ "text": "Hello",
+ }
+ ],
+ "role": "assistant",
+ }
+
+ class FakeResponse:
+ encoding = "utf-8"
+
+ def iter_content(self, chunk_size=8192, decode_unicode=True):
+ del chunk_size, decode_unicode
+ yield (
+ 'data: {"type":"response.output_item.done","output_index":0,'
+ f'"item":{json.dumps(item)}}}\n\n'
+ ).encode()
+ yield (
+ b'data: {"type":"response.completed",'
+ b'"response":{"id":"resp_1","output":[]}}\n\n'
+ )
+
+ assert codex.collect_completed_response(FakeResponse()) == {
+ "id": "resp_1",
+ "output": [item],
+ }
+
+
def test_collect_completed_response_falls_back_to_text_deltas():
class FakeResponse:
encoding = "utf-8"