fix(agent): propagate api_mode to vision provider resolution

resolve_vision_provider_client() computed resolved_api_mode from config
but never passed it to downstream resolve_provider_client() or
_get_cached_client() calls, causing custom providers with
api_mode: anthropic_messages to crash when used for vision tasks.

Also remove the for_vision special case in _normalize_aux_provider()
that incorrectly discarded named custom provider identifiers.

Fixes #8857

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

agent/auxiliary_client.py

@@ -71,13 +71,13 @@ _PROVIDER_ALIASES = {
 }
 
 
-def _normalize_aux_provider(provider: Optional[str], *, for_vision: bool = False) -> str:
+def _normalize_aux_provider(provider: Optional[str]) -> str:
     normalized = (provider or "auto").strip().lower()
     if normalized.startswith("custom:"):
         suffix = normalized.split(":", 1)[1].strip()
         if not suffix:
             return "custom"
-        normalized = suffix if not for_vision else "custom"
+        normalized = suffix
     if normalized == "codex":
         return "openai-codex"
     if normalized == "main":
@@ -1603,7 +1603,7 @@ _VISION_AUTO_PROVIDER_ORDER = (
 
 
 def _normalize_vision_provider(provider: Optional[str]) -> str:
-    return _normalize_aux_provider(provider, for_vision=True)
+    return _normalize_aux_provider(provider)
 
 
 def _resolve_strict_vision_backend(provider: str) -> Tuple[Optional[Any], Optional[str]]:
@@ -1686,6 +1686,7 @@ def resolve_vision_provider_client(
             async_mode=async_mode,
             explicit_base_url=resolved_base_url,
             explicit_api_key=resolved_api_key,
+            api_mode=resolved_api_mode,
         )
         if client is None:
             return "custom", None, None
@@ -1710,7 +1711,8 @@ def resolve_vision_provider_client(
                 # Use provider-specific vision model if available, otherwise main model.
                 vision_model = _PROVIDER_VISION_MODELS.get(main_provider, main_model)
                 rpc_client, rpc_model = resolve_provider_client(
-                    main_provider, vision_model)
+                    main_provider, vision_model,
+                    api_mode=resolved_api_mode)
                 if rpc_client is not None:
                     logger.info(
                         "Vision auto-detect: using active provider %s (%s)",
@@ -1734,7 +1736,8 @@ def resolve_vision_provider_client(
         sync_client, default_model = _resolve_strict_vision_backend(requested)
         return _finalize(requested, sync_client, default_model)
 
-    client, final_model = _get_cached_client(requested, resolved_model, async_mode)
+    client, final_model = _get_cached_client(requested, resolved_model, async_mode,
+                                             api_mode=resolved_api_mode)
     if client is None:
         return requested, None, None
     return requested, client, final_model

tests/agent/test_auxiliary_named_custom_providers.py

@@ -58,6 +58,10 @@ class TestNormalizeVisionProvider:
         assert _normalize_vision_provider("beans") == "beans"
         assert _normalize_vision_provider("deepseek") == "deepseek"
 
+    def test_custom_colon_named_provider_preserved(self):
+        from agent.auxiliary_client import _normalize_vision_provider
+        assert _normalize_vision_provider("custom:beans") == "beans"
+
     def test_codex_alias_still_works(self):
         from agent.auxiliary_client import _normalize_vision_provider
         assert _normalize_vision_provider("codex") == "openai-codex"
@@ -229,3 +233,22 @@ class TestResolveVisionProviderClientModelNormalization:
         assert provider == "zai"
         assert client is not None
         assert model == "glm-5.1"
+
+
+class TestVisionPathApiMode:
+    """Vision path should propagate api_mode to _get_cached_client."""
+
+    def test_explicit_provider_passes_api_mode(self, tmp_path):
+        _write_config(tmp_path, {
+            "model": {"default": "test-model"},
+            "auxiliary": {"vision": {"api_mode": "chat_completions"}},
+        })
+        with patch("agent.auxiliary_client._get_cached_client") as mock_gcc:
+            mock_gcc.return_value = (MagicMock(), "test-model")
+            from agent.auxiliary_client import resolve_vision_provider_client
+
+            provider, client, model = resolve_vision_provider_client(provider="deepseek")
+
+        mock_gcc.assert_called_once()
+        _, kwargs = mock_gcc.call_args
+        assert kwargs.get("api_mode") == "chat_completions"