MALFORMED_RESPONSE during Tool Use

[GinkoJack911]

Issue Description: MALFORMED_RESPONSE during Tool Use

Model Information

My Agent uses the following configuration:

• Model: gemini-3.1-pro-preview
• Planner: BuiltInPlanner()
• Thinking Level: high

The Problem

The agent occasionally outputs a thinking process but fails to generate a final response or tool call. In my architecture, the expected behavior after the thinking phase is to invoke transfer_to_agent to return control to the root_agent.

The "thought" logs indicate the model clearly knows it should call the tool; it repeatedly confirms that it needs to initiate a transfer_to_agent call. However, the final output payload is empty and ends in an error:

• toolCalls: []
• toolDefinitions: {}
• "text": ""
• "thought": true
• finish_reason: MALFORMED_RESPONSE

Is this caused by the model (thinking_level, planner, or the model version itself), or is there another underlying cause?

[llalitkumarrr]

Hello @GinkoJack911 ,

Could you please let us know which version you are currently using? If you are not on the latest version, we recommend updating and testing again with google-adk==1.27.2.

If the problem persists after updating, please share your observations. Additionally providing a reproducible code snippet would greatly help us analyze and investigate the issue further.