I have a root agent that I run via App so that I can handle its tool errors using ReflectAndRetryToolPlugin.
If I try to run evals for that agent, the evals use the root_agent directly, rather than via App

To Reproduce
Create a simple setup of an agent with a failing tool and an app using it, then create some evalset for it. The exception will not be handled

# Create the orchestrator agent with the AgentTools
root_agent = Agent(
    name="SomeAgent",
    model=ORCHESTRATOR_LLM_MODEL,
    instruction=prompt,
    tools=[
        some_failing_tool
    ],
)

app = App(
    root_agent=root_agent,
    name="app",
    plugins=[
      ReflectAndRetryToolPlugin(
            max_retries=3,
            throw_exception_if_retry_exceeded=False,  # Return message instead of raising
        ),
    ],
)

Expected behavior
I expect the evals to use the app rather then the agent if available

Desktop (please complete the following information):

• OS: macOs
• Python version(python -V): 3.13
• ADK version(pip show google-adk): 1.22

Model Information:

• Are you using LiteLLM: Yes/No: No
• Which model is being used(e.g. gemini-2.5-pro): gemini-2.5-pro

Additional context
Take a look at

It seems to me that LocalEvalService is using the root_agent directly