[darshjoshi]

run_stream() and run_async_stream() fail at the end of the stream when the output schema has a field alias. run() works fine with the same schema.

Both streaming methods rebuild the final response like this:

self.output_schema(**last_response.model_dump())

model_dump() keys the data by field name, but a field declared with alias= only accepts its alias on input. So the whole response streams fine and then fails validation. The reply also never gets added to history.

Repro with a real model (gpt-4o-mini through Instructor):

class Reply(BaseIOSchema):
    """A short reply."""
    chat_message: str = Field(..., alias="chatMessage", description="The reply text")

agent = AtomicAgent[BasicChatInputSchema, Reply](AgentConfig(client=instructor.from_openai(openai.OpenAI()), model="gpt-4o-mini"))

On main:

run(): Reply(chat_message='Hello there!')
run_stream() raised ValidationError ['1 validation error for Reply', 'chatMessage', "  Field required [type=missing, input_value={'chat_message': 'Hello there!'}, input_type=dict]"] history: 1
run_async_stream() raised ValidationError

With this PR:

run(): Reply(chat_message='Hello there!')
run_stream(): Reply(chat_message='Hello there!') history: 2
run_async_stream(): history: 2 Reply(chat_message='Hello there!')

The fix validates the dumped data with model_validate(..., by_name=True), so field names are accepted alongside aliases. Schemas without aliases behave exactly as before. by_name is available from pydantic 2.11, which matches the current floor.

I added a regression test for each streaming method. Both fail on main and pass here. black --check, flake8 and pytest --cov=atomic_agents atomic-agents (623 passed) are green.

[Eigenwise]

Reproduced the sync and async alias failures against current main. Both pass with this change, including history recording. Combined core suite, Black and Flake8 pass. Thanks.