Worked on improving the stability of AgentTool output in the google/adk-python repository, focusing on scenarios where summarization is disabled. Addressed a UI issue by ensuring that textual output is always emitted, even when skip_summarization is set to True. Implemented robust handling for non-string results by serializing them to JSON and appending them as text parts to the response event. Enhanced reliability by adding comprehensive regression tests that cover string, JSON-string, and structured outputs. Utilized Python, JSON, and unit testing to ensure consistent tool behavior and prevent UI hangs, contributing to more maintainable and predictable output handling.