fix(canvas): 在线 Agent - 系统模型接入工具任务路由

backend/internal/service/provider.go

@@ -35,11 +35,17 @@ type canvasGenerationInput struct {
 	TextHistory     []providerTextMessage  `json:"textHistory"`
 	Mask            *providerMedia         `json:"mask"`
 	Metadata        map[string]interface{} `json:"metadata"`
+	AgentRequests   *agentToolRequests     `json:"agentRequests"`
 	ImageCapability *ImageCapabilityConfig `json:"-"`
 	StreamText      bool                   `json:"-"` // 分镜请求使用上游 SSE 保活；最终结构仍在流结束后统一校验。
 	VideoCapability *VideoCapabilityConfig `json:"-"`
 }
 
+type agentToolRequests struct {
+	Responses      map[string]interface{} `json:"responses"`
+	ChatCompletion map[string]interface{} `json:"chatCompletion"`
+}
+
 type providerTextMessage struct {
 	Role    string `json:"role"`
 	Content string `json:"content"`
@@ -247,6 +253,9 @@ func (s *Service) processCanvasGenerationTask(ctx context.Context, userID string
 	case "image":
 		return runImageTask(ctx, input)
 	case "text":
+		if input.AgentRequests != nil {
+			return runAgentToolTask(ctx, input)
+		}
 		result, taskErr := runTextTask(ctx, input)
 		if taskErr == nil && promptTemplateOperation != "" {
 			taskErr = validatePromptTemplateResult(promptTemplateOperation, result)
@@ -261,6 +270,78 @@ func (s *Service) processCanvasGenerationTask(ctx context.Context, userID string
 	}
 }
 
+func runAgentToolTask(ctx context.Context, input canvasGenerationInput) (map[string]interface{}, error) {
+	request := input.AgentRequests.ChatCompletion
+	path := "/chat/completions"
+	protocol := "chat-completion"
+	if input.Config.InterfaceType == string(model.ChannelInterfaceOpenAIResponse) {
+		request = input.AgentRequests.Responses
+		path = "/responses"
+		protocol = "responses"
+	}
+	if request == nil {
+		return nil, errors.New("画布 Agent 工具请求缺少协议参数")
+	}
+	body := cloneStringAnyMap(request)
+	body["model"] = input.Config.Model
+	var payload map[string]interface{}
+	if err := postJSON(ctx, input.Config, path, body, &payload); err != nil {
+		return nil, err
+	}
+	return parseAgentToolPayload(payload, protocol)
+}
+
+func parseAgentToolPayload(payload map[string]interface{}, protocol string) (map[string]interface{}, error) {
+	if err := validateTextPayload(payload); err != nil {
+		return nil, err
+	}
+	result := map[string]interface{}{"mode": "text", "text": "", "toolCalls": []interface{}{}}
+	if protocol == "responses" {
+		result["text"] = firstNonEmptyString(stringField(payload, "output_text"), extractResponseText(payload))
+		calls := make([]interface{}, 0)
+		for _, value := range interfaceSlice(payload["output"]) {
+			item, _ := value.(map[string]interface{})
+			if stringField(item, "type") != "function_call" {
+				continue
+			}
+			calls = append(calls, map[string]interface{}{"id": firstNonEmptyString(stringField(item, "call_id"), stringField(item, "id")), "type": "function", "function": map[string]interface{}{"name": stringField(item, "name"), "arguments": stringField(item, "arguments")}})
+		}
+		result["toolCalls"] = calls
+		return result, nil
+	}
+	choices := interfaceSlice(payload["choices"])
+	if len(choices) == 0 {
+		return nil, errors.New("画布 Agent 接口没有返回 choices")
+	}
+	choice, _ := choices[0].(map[string]interface{})
+	message, _ := choice["message"].(map[string]interface{})
+	result["text"] = stringField(message, "content")
+	if reasoning := firstNonEmptyString(stringField(message, "reasoning_content"), stringField(message, "reasoning")); reasoning != "" {
+		result["reasoning"] = reasoning
+	}
+	calls := make([]interface{}, 0)
+	for _, value := range interfaceSlice(message["tool_calls"]) {
+		item, _ := value.(map[string]interface{})
+		function, _ := item["function"].(map[string]interface{})
+		calls = append(calls, map[string]interface{}{"id": stringField(item, "id"), "type": "function", "function": map[string]interface{}{"name": stringField(function, "name"), "arguments": stringField(function, "arguments")}})
+	}
+	result["toolCalls"] = calls
+	return result, nil
+}
+
+func interfaceSlice(value interface{}) []interface{} {
+	items, _ := value.([]interface{})
+	return items
+}
+
+func cloneStringAnyMap(value map[string]interface{}) map[string]interface{} {
+	cloned := make(map[string]interface{}, len(value)+1)
+	for key, item := range value {
+		cloned[key] = item
+	}
+	return cloned
+}
+
 type styleExecutionPlanDocument struct {
 	SchemaVersion   int    `json:"schemaVersion"`
 	ProfilePresetID string `json:"profilePresetId"`

backend/internal/service/provider_test.go

@@ -107,6 +107,59 @@ data: [DONE]
 	}
 }
 
+func TestParseAgentToolPayloadSupportsChatCompletions(t *testing.T) {
+	result, err := parseAgentToolPayload(map[string]interface{}{
+		"choices": []interface{}{map[string]interface{}{
+			"message": map[string]interface{}{
+				"content": "准备读取画布",
+				"tool_calls": []interface{}{map[string]interface{}{
+					"id": "call-1",
+					"function": map[string]interface{}{"name": "canvas_get_state", "arguments": `{}`},
+				}},
+			},
+		}},
+	}, "chat-completion")
+	if err != nil {
+		t.Fatalf("parseAgentToolPayload() error = %v", err)
+	}
+	if result["text"] != "准备读取画布" {
+		t.Fatalf("text = %v", result["text"])
+	}
+	calls, _ := result["toolCalls"].([]interface{})
+	if len(calls) != 1 {
+		t.Fatalf("toolCalls = %#v", result["toolCalls"])
+	}
+	call, _ := calls[0].(map[string]interface{})
+	function, _ := call["function"].(map[string]interface{})
+	if call["id"] != "call-1" || function["name"] != "canvas_get_state" || function["arguments"] != `{}` {
+		t.Fatalf("tool call = %#v", call)
+	}
+}
+
+func TestParseAgentToolPayloadSupportsResponses(t *testing.T) {
+	result, err := parseAgentToolPayload(map[string]interface{}{
+		"output": []interface{}{
+			map[string]interface{}{"type": "message", "content": []interface{}{map[string]interface{}{"type": "output_text", "text": "开始操作"}}},
+			map[string]interface{}{"type": "function_call", "call_id": "call-2", "name": "canvas_apply_ops", "arguments": `{"ops":[]}`},
+		},
+	}, "responses")
+	if err != nil {
+		t.Fatalf("parseAgentToolPayload() error = %v", err)
+	}
+	if result["text"] != "开始操作" {
+		t.Fatalf("text = %v", result["text"])
+	}
+	calls, _ := result["toolCalls"].([]interface{})
+	if len(calls) != 1 {
+		t.Fatalf("toolCalls = %#v", result["toolCalls"])
+	}
+	call, _ := calls[0].(map[string]interface{})
+	function, _ := call["function"].(map[string]interface{})
+	if call["id"] != "call-2" || function["name"] != "canvas_apply_ops" || function["arguments"] != `{"ops":[]}` {
+		t.Fatalf("tool call = %#v", call)
+	}
+}
+
 func TestPostStreamingTextSetsStreamHeaders(t *testing.T) {
 	t.Setenv("CANVAS_ALLOW_PRIVATE_UPSTREAMS", "true")
 	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {