I am designing a multi agent app using google adk. The agent should accept a SQL script and emit few details from the script. Like source, stage, target details. The common table expression and other details. In my design I wrapped all the extractor inside Parallel agent. Similarly I wrapped the initial agent which accepts the SQL file and the parallel agent inside a sequential agent. When the file is passed on during the initial interaction itself, my agent is working fine. If the user just send a "Hi" message to activate the agent. My agent ends the interaction with emitting hallucinated output. I have clearly mentioned in the instruction of the first agent that it should wait for the input from the user. But its not respecting it. Even setting the state variable by using before_model_callback in the first agent and validating the value in the second agent prompt doesn't work. Kindly share your suggestion.