Callbacks seemingly not working

[TommyBark]

Hello, I am following Tutorial ( https://google.github.io/adk-docs/get-started/tutorial/#step-5-adding-safety-input-guardrail-with-before_model_callback ) and I am not sure if I am encountering bug or if it's intended but following logs are generated using the provided code.

This is the agent definition:

weather_agent_team = Agent(
    name="weather_agent_v5", 
    model=root_agent_model,
    description="The main coordinator agent. Handles weather requests and delegates greetings/farewells to specialists.",
    instruction="You are the main Weather Agent coordinating a team. Your primary responsibility is to provide weather information. "
    "Use the 'get_weather' tool ONLY for specific weather requests (e.g., 'weather in London'). "
    "You have specialized sub-agents: "
    "1. 'greeting_agent': Handles simple greetings like 'Hi', 'Hello'. Delegate to it for these. "
    "2. 'farewell_agent': Handles simple farewells like 'Bye', 'See you'. Delegate to it for these. "
    "Analyze the user's query. If it's a greeting, delegate to 'greeting_agent'. If it's a farewell, delegate to 'farewell_agent'. "
    "If it's a weather request, handle it yourself using 'get_weather'. "
    "For anything else, respond appropriately or state you cannot handle it.",
    tools=[
        get_weather_stateful
    ],
    sub_agents=[greeting_agent, farewell_agent],
    output_key="last_weather_report",
    before_model_callback=block_keyword_guardrail,
)

Now these are the logs I get when testing callbacks:

You: hello!
--- Callback: block_keyword_guardrail running for agent: weather_agent_v5 ---
--- Callback: Inspecting last user message: 'hello!...' ---
--- Callback: Keyword not found. Allowing LLM call for weather_agent_v5. ---
--- Tool: say_hello called with name: there ---
Agent: Hello, there!

You: hello!
# Here, triggering the same subagent, callback is not triggered again
--- Tool: say_hello called with name: there ---
Agent: Hello, there!


You: BLOCK
# Here, using blocked word, callback is not triggered and also choice of subagent is weird.
--- Tool: say_hello called with name: there ---
Agent: Hello, there!


You: what is the weather in New York?
# Here callback gets triggered twice and in both cases reads a wrong user message (this is just a bug in guardrail fnc implementation)
--- Callback: block_keyword_guardrail running for agent: weather_agent_v5 ---
--- Callback: Inspecting last user message: 'For context:...' ---
--- Callback: Keyword not found. Allowing LLM call for weather_agent_v5. ---
...
Agent: The weather in New york is sunny with a temperature of 25°C.

What is the reason for callbacks not being triggered until I prompt the agent with a prompt that asks for the tool call?

I am thinking that it had something to do with execution order and the fact that the Greeting agent itself doesn't have before_model_callback but that is surely not the case as the Root agent's LLM is supposed to be triggered first before any of the sub-agent's LLMs right?

[boyangsvl]

It seems that you have only set the callback on weather_agent_team agent, so it's not applied when other agents are replying. You will need to set the callback on all the agents you want it to apply.