fix(tools): fix ToolInvokeMessage Union type parsing issue (#31450)

Co-authored-by: qiaofenglin <qiaofenglin@baidu.com>

api/core/tools/entities/tool_entities.py

@@ -130,7 +130,7 @@ class ToolInvokeMessage(BaseModel):
         text: str
 
     class JsonMessage(BaseModel):
-        json_object: dict
+        json_object: dict | list
         suppress_output: bool = Field(default=False, description="Whether to suppress JSON output in result string")
 
     class BlobMessage(BaseModel):
@@ -144,7 +144,14 @@ class ToolInvokeMessage(BaseModel):
         end: bool = Field(..., description="Whether the chunk is the last chunk")
 
     class FileMessage(BaseModel):
-        pass
+        file_marker: str = Field(default="file_marker")
+
+        @model_validator(mode="before")
+        @classmethod
+        def validate_file_message(cls, values):
+            if isinstance(values, dict) and "file_marker" not in values:
+                raise ValueError("Invalid FileMessage: missing file_marker")
+            return values
 
     class VariableMessage(BaseModel):
         variable_name: str = Field(..., description="The name of the variable")
@@ -234,10 +241,22 @@ class ToolInvokeMessage(BaseModel):
 
     @field_validator("message", mode="before")
     @classmethod
-    def decode_blob_message(cls, v):
+    def decode_blob_message(cls, v, info: ValidationInfo):
+        # 处理 blob 解码
         if isinstance(v, dict) and "blob" in v:
             with contextlib.suppress(Exception):
                 v["blob"] = base64.b64decode(v["blob"])
+
+        # Force correct message type based on type field
+        # Only wrap dict types to avoid wrapping already parsed Pydantic model objects
+        if info.data and isinstance(info.data, dict) and isinstance(v, dict):
+            msg_type = info.data.get("type")
+            if msg_type == cls.MessageType.JSON:
+                if "json_object" not in v:
+                    v = {"json_object": v}
+            elif msg_type == cls.MessageType.FILE:
+                v = {"file_marker": "file_marker"}
+
         return v
 
     @field_serializer("message")

api/core/workflow/nodes/agent/agent_node.py

@@ -494,7 +494,7 @@ class AgentNode(Node[AgentNodeData]):
 
         text = ""
         files: list[File] = []
-        json_list: list[dict] = []
+        json_list: list[dict | list] = []
 
         agent_logs: list[AgentLogEvent] = []
         agent_execution_metadata: Mapping[WorkflowNodeExecutionMetadataKey, Any] = {}
@@ -568,13 +568,18 @@ class AgentNode(Node[AgentNodeData]):
             elif message.type == ToolInvokeMessage.MessageType.JSON:
                 assert isinstance(message.message, ToolInvokeMessage.JsonMessage)
                 if node_type == NodeType.AGENT:
-                    msg_metadata: dict[str, Any] = message.message.json_object.pop("execution_metadata", {})
+                    if isinstance(message.message.json_object, dict):
-                    llm_usage = LLMUsage.from_metadata(cast(LLMUsageMetadata, msg_metadata))
+                        msg_metadata: dict[str, Any] = message.message.json_object.pop("execution_metadata", {})
-                    agent_execution_metadata = {
+                        llm_usage = LLMUsage.from_metadata(cast(LLMUsageMetadata, msg_metadata))
-                        WorkflowNodeExecutionMetadataKey(key): value
+                        agent_execution_metadata = {
-                        for key, value in msg_metadata.items()
+                            WorkflowNodeExecutionMetadataKey(key): value
-                        if key in WorkflowNodeExecutionMetadataKey.__members__.values()
+                            for key, value in msg_metadata.items()
-                    }
+                            if key in WorkflowNodeExecutionMetadataKey.__members__.values()
+                        }
+                    else:
+                        msg_metadata = {}
+                        llm_usage = LLMUsage.empty_usage()
+                        agent_execution_metadata = {}
                 if message.message.json_object:
                     json_list.append(message.message.json_object)
             elif message.type == ToolInvokeMessage.MessageType.LINK:
@@ -683,7 +688,7 @@ class AgentNode(Node[AgentNodeData]):
                 yield agent_log
 
         # Add agent_logs to outputs['json'] to ensure frontend can access thinking process
-        json_output: list[dict[str, Any]] = []
+        json_output: list[dict[str, Any] | list[Any]] = []
 
         # Step 1: append each agent log as its own dict.
         if agent_logs:

api/core/workflow/nodes/datasource/datasource_node.py

@@ -301,7 +301,7 @@ class DatasourceNode(Node[DatasourceNodeData]):
 
         text = ""
         files: list[File] = []
-        json: list[dict] = []
+        json: list[dict | list] = []
 
         variables: dict[str, Any] = {}
 

api/core/workflow/nodes/tool/tool_node.py

@@ -244,7 +244,7 @@ class ToolNode(Node[ToolNodeData]):
 
         text = ""
         files: list[File] = []
-        json: list[dict] = []
+        json: list[dict | list] = []
 
         variables: dict[str, Any] = {}
 
@@ -400,7 +400,7 @@ class ToolNode(Node[ToolNodeData]):
                         message.message.metadata = dict_metadata
 
         # Add agent_logs to outputs['json'] to ensure frontend can access thinking process
-        json_output: list[dict[str, Any]] = []
+        json_output: list[dict[str, Any] | list[Any]] = []
 
         # Step 2: normalize JSON into {"data": [...]}.change json to list[dict]
         if json: