Description
Agents configured with additional_properties lose those properties when executed through a handoff workflow.
The same agent exposes its configured properties during a direct run, but middleware sees an empty dictionary when the agent runs through HandoffBuilder.
This breaks middleware and integrations that use agent metadata for tenant information, tracing tags, feature flags, or other application-specific configuration.
Expected behavior
Agents participating in a handoff workflow should retain their configured additional_properties.
Actual behavior
The original agent retains its properties, but the agent executed by the handoff workflow has no additional properties.
Code Sample
import asyncio
from collections.abc import Awaitable, Mapping, Sequence
from typing import Any
from agent_framework import (
Agent,
AgentContext,
BaseChatClient,
ChatOptions,
ChatResponse,
Message,
)
from agent_framework.orchestrations import HandoffBuilder
seen: list[dict[str, Any]] = []
class FakeClient(BaseChatClient[ChatOptions[Any]]):
model = "fake"
def _inner_get_response(
self,
*,
messages: Sequence[Message],
stream: bool,
options: Mapping[str, Any],
**kwargs: Any,
) -> Awaitable[ChatResponse]:
async def respond() -> ChatResponse:
return ChatResponse(
messages=[Message("assistant", ["done"])]
)
return respond()
async def observe(context: AgentContext, call_next) -> None:
seen.append(dict(context.agent.additional_properties))
await call_next()
async def main() -> None:
coordinator = Agent(
client=FakeClient(),
name="coordinator",
additional_properties={
"tenant": "contoso",
"trace_tag": "triage",
},
middleware=[observe],
require_per_service_call_history_persistence=True,
)
specialist = Agent(
client=FakeClient(),
name="specialist",
require_per_service_call_history_persistence=True,
)
await coordinator.run("hello")
print("Direct run:", seen)
seen.clear()
workflow = (
HandoffBuilder(
participants=[coordinator, specialist],
termination_condition=lambda conversation: any(
message.role == "assistant"
for message in conversation
),
)
.with_start_agent(coordinator)
.build()
)
await workflow.run("hello")
print("Configured:", coordinator.additional_properties)
print("Handoff run:", seen)
asyncio.run(main())
Actual output:
Direct run: [{'tenant': 'contoso', 'trace_tag': 'triage'}]
Configured: {'tenant': 'contoso', 'trace_tag': 'triage'}
Handoff run: [{}]
Expected handoff output:
Handoff run: [{'tenant': 'contoso', 'trace_tag': 'triage'}]Error Messages / Stack Traces
No exception is raised. The agent metadata is silently discarded.
Package Versions
agent-framework-orchestrations: 1.0.2
Python Version
Python 3.12.10
Additional Context
Issue #5173 previously reported that handoff execution discarded function middleware. PR #5220 fixed middleware preservation, but additional_properties remain unpreserved. This report concerns a separate public agent configuration field.
Description
Agents configured with
additional_propertieslose those properties when executed through a handoff workflow.The same agent exposes its configured properties during a direct run, but middleware sees an empty dictionary when the agent runs through
HandoffBuilder.This breaks middleware and integrations that use agent metadata for tenant information, tracing tags, feature flags, or other application-specific configuration.
Expected behavior
Agents participating in a handoff workflow should retain their configured
additional_properties.Actual behavior
The original agent retains its properties, but the agent executed by the handoff workflow has no additional properties.
Code Sample
import asyncio from collections.abc import Awaitable, Mapping, Sequence from typing import Any from agent_framework import ( Agent, AgentContext, BaseChatClient, ChatOptions, ChatResponse, Message, ) from agent_framework.orchestrations import HandoffBuilder seen: list[dict[str, Any]] = [] class FakeClient(BaseChatClient[ChatOptions[Any]]): model = "fake" def _inner_get_response( self, *, messages: Sequence[Message], stream: bool, options: Mapping[str, Any], **kwargs: Any, ) -> Awaitable[ChatResponse]: async def respond() -> ChatResponse: return ChatResponse( messages=[Message("assistant", ["done"])] ) return respond() async def observe(context: AgentContext, call_next) -> None: seen.append(dict(context.agent.additional_properties)) await call_next() async def main() -> None: coordinator = Agent( client=FakeClient(), name="coordinator", additional_properties={ "tenant": "contoso", "trace_tag": "triage", }, middleware=[observe], require_per_service_call_history_persistence=True, ) specialist = Agent( client=FakeClient(), name="specialist", require_per_service_call_history_persistence=True, ) await coordinator.run("hello") print("Direct run:", seen) seen.clear() workflow = ( HandoffBuilder( participants=[coordinator, specialist], termination_condition=lambda conversation: any( message.role == "assistant" for message in conversation ), ) .with_start_agent(coordinator) .build() ) await workflow.run("hello") print("Configured:", coordinator.additional_properties) print("Handoff run:", seen) asyncio.run(main()) Actual output: Direct run: [{'tenant': 'contoso', 'trace_tag': 'triage'}] Configured: {'tenant': 'contoso', 'trace_tag': 'triage'} Handoff run: [{}] Expected handoff output: Handoff run: [{'tenant': 'contoso', 'trace_tag': 'triage'}]Error Messages / Stack Traces
Package Versions
agent-framework-orchestrations: 1.0.2
Python Version
Python 3.12.10
Additional Context
Issue #5173 previously reported that handoff execution discarded function middleware. PR #5220 fixed middleware preservation, but
additional_propertiesremain unpreserved. This report concerns a separate public agent configuration field.