Repository navigation
Python: [Bug]: Duplicate arguments in declaration-only function call when streamed #6973
Description
Description
When streaming an agent with a declaration-only tool, the arguments of the function call are duplicated when looking at the message contents.
Code Sample
import asyncio
from agent_framework import Agent, FunctionTool
from agent_framework.foundry import FoundryChatClient
from azure.identity import AzureCliCredential
from pydantic import BaseModel, Field
class WeatherInput(BaseModel):
location: str = Field(description="The location to get the weather for.")
async def main() -> None:
client = FoundryChatClient(
credential=AzureCliCredential(),
)
agent = Agent(
client=client,
name="WeatherAgent",
instructions="You are a helpful weather agent. Use the get_weather tool to answer questions.",
tools=[
FunctionTool(
name="get_weather", description="Get the weather for a given location.", input_model=WeatherInput
)
],
)
response_stream = agent.run("What's the weather like in Seattle?", stream=True)
outputs = await response_stream.get_final_response()
print(outputs.messages[0].contents[0].arguments) # {"location":"Seattle"}{"location":"Seattle"}
if __name__ == "__main__":
asyncio.run(main())Error Messages / Stack Traces
Package Versions
agent-framework-core: 1.10.0, agent-framework-foundry: 1.10.0
Python Version
Python 3.13.13
Additional Context
No response
Metadata
Metadata
Labels
Type
Projects
- Status
Activity
github-actions commentedon Jul 7, 2026
🤖 Automated triage reproduction notes (agent-authored — trust but verify)
Agent analysis
Repro: FunctionInvocationLayer._stream() in packages/core/agent_framework/_tools.py (around line 2821) re-yields function_call contents that were already yielded from the inner stream (line 2774), causing _process_update in _types.py (line 1910) to concatenate arguments via +=. The fix should prevent the duplicate yield of function_call contents for declaration-only tools in the streaming path — either by not yielding the extra update when the results are already function_call type, or by not including function_call items in result["function_call_results"] at line 2379. Minimal repro: create a MockBaseChatClient with FunctionInvocationLayer, set streaming_responses with function_call chunks for a declaration-only FunctionTool, call get_response(stream=True), and check final_response arguments.
python/packages/core/tests/core/test_declaration_only_streaming_duplication.py1.10.01.10.0fix: avoid duplicating arguments in streamed declaration-only tool ca…
fix: avoid duplicating arguments in streamed declaration-only tool ca…
fix: avoid duplicating arguments in streamed declaration-only tool ca…
fix: avoid duplicating arguments in streamed declaration-only tool ca…