fix: retry auxiliary calls without unsupported temperature

agent/auxiliary_client.py

@@ -2796,6 +2796,19 @@ def _build_call_kwargs(
     return kwargs
 
 
+def _is_unsupported_temperature_error(exc: Exception) -> bool:
+    """True when a provider rejects the temperature request parameter.
+
+    Some OpenAI-compatible endpoints (notably Codex Responses-compatible
+    gateways) reject ``temperature`` with messages like
+    ``Unsupported parameter: temperature`` instead of the newer
+    ``unsupported_parameter`` code.  Auxiliary callers frequently pass a
+    conservative temperature for generic models, so retry once without it.
+    """
+    msg = str(exc).lower()
+    return "temperature" in msg and ("unsupported" in msg or "not support" in msg)
+
+
 def _validate_llm_response(response: Any, task: str = None) -> Any:
     """Validate that an LLM response has the expected .choices[0].message shape.
 
@@ -2958,6 +2971,19 @@ def call_llm(
             client.chat.completions.create(**kwargs), task)
     except Exception as first_err:
         err_str = str(first_err)
+        if _is_unsupported_temperature_error(first_err) and "temperature" in kwargs:
+            kwargs.pop("temperature", None)
+            try:
+                return _validate_llm_response(
+                    client.chat.completions.create(**kwargs), task)
+            except Exception as retry_err:
+                # If the temperature retry also hits a payment or connection
+                # error, fall through to the fallback chain below.
+                if not (_is_payment_error(retry_err) or _is_connection_error(retry_err)):
+                    raise
+                first_err = retry_err
+                err_str = str(first_err)
+
         if "max_tokens" in err_str or "unsupported_parameter" in err_str:
             kwargs.pop("max_tokens", None)
             kwargs["max_completion_tokens"] = max_tokens
@@ -3222,6 +3248,19 @@ async def async_call_llm(
             await client.chat.completions.create(**kwargs), task)
     except Exception as first_err:
         err_str = str(first_err)
+        if _is_unsupported_temperature_error(first_err) and "temperature" in kwargs:
+            kwargs.pop("temperature", None)
+            try:
+                return _validate_llm_response(
+                    await client.chat.completions.create(**kwargs), task)
+            except Exception as retry_err:
+                # If the temperature retry also hits a payment or connection
+                # error, fall through to the fallback chain below.
+                if not (_is_payment_error(retry_err) or _is_connection_error(retry_err)):
+                    raise
+                first_err = retry_err
+                err_str = str(first_err)
+
         if "max_tokens" in err_str or "unsupported_parameter" in err_str:
             kwargs.pop("max_tokens", None)
             kwargs["max_completion_tokens"] = max_tokens

tests/agent/test_auxiliary_client.py

@@ -863,6 +863,79 @@ def test_500_not_connection(self):
         assert _is_connection_error(err) is False
 
 
+class TestUnsupportedTemperatureRetry:
+    """Auxiliary calls retry once without temperature when endpoints reject it."""
+
+    def test_sync_call_retries_without_temperature(self):
+        client = MagicMock()
+        client.base_url = "https://chatgpt.com/backend-api/codex"
+        response = MagicMock()
+        client.chat.completions.create.side_effect = [
+            Exception("HTTP 400: Error code: 400 - {'detail': 'Unsupported parameter: temperature'}"),
+            response,
+        ]
+
+        with patch(
+            "agent.auxiliary_client._get_cached_client",
+            return_value=(client, "gpt-5.5"),
+        ), patch(
+            "agent.auxiliary_client._resolve_task_provider_model",
+            return_value=("auto", "gpt-5.5", None, None, None),
+        ), patch(
+            "agent.auxiliary_client._validate_llm_response",
+            side_effect=lambda resp, _task: resp,
+        ):
+            result = call_llm(
+                task="flush_memories",
+                messages=[{"role": "user", "content": "hello"}],
+                temperature=0.3,
+                max_tokens=5120,
+            )
+
+        assert result is response
+        assert client.chat.completions.create.call_count == 2
+        first_kwargs = client.chat.completions.create.call_args_list[0].kwargs
+        second_kwargs = client.chat.completions.create.call_args_list[1].kwargs
+        assert first_kwargs["temperature"] == 0.3
+        assert "temperature" not in second_kwargs
+        assert second_kwargs["max_tokens"] == 5120
+
+    @pytest.mark.asyncio
+    async def test_async_call_retries_without_temperature(self):
+        client = MagicMock()
+        client.base_url = "https://chatgpt.com/backend-api/codex"
+        response = MagicMock()
+        client.chat.completions.create = AsyncMock(side_effect=[
+            Exception("HTTP 400: Error code: 400 - {'detail': 'Unsupported parameter: temperature'}"),
+            response,
+        ])
+
+        with patch(
+            "agent.auxiliary_client._get_cached_client",
+            return_value=(client, "gpt-5.5"),
+        ), patch(
+            "agent.auxiliary_client._resolve_task_provider_model",
+            return_value=("auto", "gpt-5.5", None, None, None),
+        ), patch(
+            "agent.auxiliary_client._validate_llm_response",
+            side_effect=lambda resp, _task: resp,
+        ):
+            result = await async_call_llm(
+                task="session_search",
+                messages=[{"role": "user", "content": "hello"}],
+                temperature=0.3,
+                max_tokens=5120,
+            )
+
+        assert result is response
+        assert client.chat.completions.create.await_count == 2
+        first_kwargs = client.chat.completions.create.call_args_list[0].kwargs
+        second_kwargs = client.chat.completions.create.call_args_list[1].kwargs
+        assert first_kwargs["temperature"] == 0.3
+        assert "temperature" not in second_kwargs
+        assert second_kwargs["max_tokens"] == 5120
+
+
 class TestKimiTemperatureOmitted:
     """Kimi/Moonshot models should have temperature OMITTED from API kwargs.