feat(runtime): 适配推理模式开关

apps/api/internal/httpapi/chat_completions_mode_test.go

@@ -138,6 +138,27 @@ func TestWriteCompatibleTaskResponseReturnsJSONWhenStreamIsFalse(t *testing.T) {
 	}
 }
 
+func TestWriteCompatibleTaskResponseMapsInvalidParameterToBadRequest(t *testing.T) {
+	executor := &fakeTaskExecutor{
+		runErr: &clients.ClientError{
+			Code:      "invalid_parameter",
+			Message:   "reasoning_effort must be one of: none, minimal, low, medium, high, xhigh",
+			Retryable: false,
+		},
+	}
+	req := httptest.NewRequest(http.MethodPost, "/api/v1/chat/completions", nil)
+	recorder := httptest.NewRecorder()
+
+	writeCompatibleTaskResponse(context.Background(), recorder, req, executor, "chat.completions", "gpt-test", store.GatewayTask{ID: "task-test"}, &auth.User{}, false, false)
+
+	if recorder.Code != http.StatusBadRequest {
+		t.Fatalf("status=%d want=%d body=%s", recorder.Code, http.StatusBadRequest, recorder.Body.String())
+	}
+	if !strings.Contains(recorder.Body.String(), "invalid_parameter") {
+		t.Fatalf("response should include invalid_parameter code: %s", recorder.Body.String())
+	}
+}
+
 func TestWriteCompatibleTaskResponseReturnsSSEWhenStreamIsTrue(t *testing.T) {
 	executor := &fakeTaskExecutor{
 		deltas: []clients.StreamDeltaEvent{{Text: "hel"}, {Text: "lo"}},
@@ -284,15 +305,22 @@ type fakeTaskExecutor struct {
 	streamCalls  int
 	deltas       []clients.StreamDeltaEvent
 	output       map[string]any
+	runErr       error
 }
 
 func (f *fakeTaskExecutor) Execute(context.Context, store.GatewayTask, *auth.User) (runner.Result, error) {
 	f.executeCalls++
+	if f.runErr != nil {
+		return runner.Result{}, f.runErr
+	}
 	return runner.Result{Output: f.output}, nil
 }
 
 func (f *fakeTaskExecutor) ExecuteStream(_ context.Context, _ store.GatewayTask, _ *auth.User, onDelta clients.StreamDelta) (runner.Result, error) {
 	f.streamCalls++
+	if f.runErr != nil {
+		return runner.Result{}, f.runErr
+	}
 	for _, delta := range f.deltas {
 		if err := onDelta(delta); err != nil {
 			return runner.Result{}, err

apps/api/internal/httpapi/handlers.go

@@ -1306,7 +1306,7 @@ func scopeForTaskKind(kind string) string {
 
 func statusFromRunError(err error) int {
 	switch {
-	case clients.ErrorCode(err) == "bad_request" || clients.ErrorCode(err) == "cloned_voice_expired" || clients.ErrorCode(err) == "cloned_voice_unavailable" || clients.ErrorCode(err) == "cloned_voice_platform_unavailable" || clients.ErrorCode(err) == "unsupported_operation" || clients.ErrorCode(err) == "invalid_proxy":
+	case clients.ErrorCode(err) == "bad_request" || clients.ErrorCode(err) == "invalid_parameter" || clients.ErrorCode(err) == "cloned_voice_expired" || clients.ErrorCode(err) == "cloned_voice_unavailable" || clients.ErrorCode(err) == "cloned_voice_platform_unavailable" || clients.ErrorCode(err) == "unsupported_operation" || clients.ErrorCode(err) == "invalid_proxy":
 		return http.StatusBadRequest
 	case clients.ErrorCode(err) == "cloned_voice_not_found":
 		return http.StatusNotFound

apps/api/internal/httpapi/openapi_models.go

@@ -190,7 +190,7 @@ type TaskRequest struct {
 	Stream     bool          `json:"stream,omitempty" example:"false"`
 	RunMode    string        `json:"runMode,omitempty" example:"simulation"`
 	MaxTokens  int           `json:"max_tokens,omitempty" example:"512"`
-	// ReasoningEffort 推理深度，OpenAI-compatible 请求字段；开放字符串，取值随 provider 和模型能力而定，常见值为 none、minimal、low、medium、high、xhigh，也可配置 max 等供应商自定义值。
+	// ReasoningEffort 推理强度，OpenAI-compatible 请求字段；仅支持 none、minimal、low、medium、high、xhigh。供应商自定义取值由网关按平台适配。
 	ReasoningEffort     string  `json:"reasoning_effort,omitempty" example:"medium"`
 	Size                string  `json:"size,omitempty" example:"1024x1024"`
 	Duration            int     `json:"duration,omitempty" example:"5"`
@@ -216,7 +216,7 @@ type ChatCompletionRequest struct {
 	Messages    []ChatMessage `json:"messages"`
 	Temperature float64       `json:"temperature,omitempty" example:"0.7"`
 	MaxTokens   int           `json:"max_tokens,omitempty" example:"512"`
-	// ReasoningEffort 推理深度，OpenAI-compatible 请求字段；开放字符串，取值随 provider 和模型能力而定，常见值为 none、minimal、low、medium、high、xhigh，也可配置 max 等供应商自定义值。
+	// ReasoningEffort 推理强度，OpenAI-compatible 请求字段；仅支持 none、minimal、low、medium、high、xhigh。供应商自定义取值由网关按平台适配。
 	ReasoningEffort string `json:"reasoning_effort,omitempty" example:"medium"`
 	Stream          bool   `json:"stream,omitempty" example:"false"`
 	RunMode         string `json:"runMode,omitempty" example:"simulation"`