容易造成超出模型token的问题，能不能提供改进adk上下文相关的问题

I'm developing a multi-agent system similar to Meta's "Manus," and encountering issues with the model's distorted tool-calling capabilities, primarily related to context. We eagerly hope the official team can provide improvements to the ADK's context-related methods. Thank you

Can you be more specific about your question? For example, what is the setup and what models/tools are you using?

Move to discussion/question as it's a question.