Structured logging was losing critical data during JSON serialization when BaseModel classes contained fields typed as base classes (like BaseChatMessage) but holding subclass instances (like TextMessage).

Problem

When PydanticJsonMessageSerializer serialized messages, it used model_dump_json() without serialize_as_any=True, which only preserved fields defined in the base class schema. Subclass-specific fields like TextMessage.content and TextMessage.type were being lost:

# Before fix - data loss
text_msg = TextMessage(content="Hello, world!", source="Agent")
group_msg = GroupChatMessage(message=text_msg)
json_data = group_msg.model_dump_json()
# Result: {"message": {"id": "...", "source": "Agent", "created_at": "..."}}
# Missing: content and type fields

Solution

Added SerializeAsAny annotations to BaseModel fields containing BaseChatMessage or BaseAgentEvent types:

• GroupChatStart.messages: List[SerializeAsAny[BaseChatMessage]]
• GroupChatMessage.message: SerializeAsAny[BaseAgentEvent | BaseChatMessage]
• TaskResult.messages: Sequence[SerializeAsAny[BaseAgentEvent | BaseChatMessage]]
• GroupChatAgentResponse.response: SerializeAsAny[Response]
• GroupChatTeamResponse.result: SerializeAsAny[TaskResult]

Result

Now all subclass data is preserved during serialization:

# After fix - complete data preservation
json_data = group_msg.model_dump_json()
# Result: {"message": {"id": "...", "source": "Agent", "created_at": "...", "content": "Hello, world!", "type": "TextMessage"}}
# All fields preserved!

This fix ensures structured logging captures complete message information for better observability and debugging, with no breaking changes to existing APIs.

Fixes #6855.

💬 Share your feedback on Copilot coding agent for the chance to win a $200 gift card! Click here to start the survey.