When in agent mode, minimize unnecessary interruptions.

未指定任何文件、测试或入口点。首先复现 agent-mode 的三种中断类型，然后追踪应用切换、授权提示和使用限制的处理方式；要视为完成，需要有经 maintainer 批准的范围，以及针对已优先处理的中断变更的明确行为。

Currently, in agent mode, there are the following types of interruptions:

1. If switched to another application, the agent will forcibly switch back to Xcode after modifying a piece of code.
2. Seek user authorization permission.
   1. For example, by running a command through the terminal.
3. Exceed the limit.
   1. The number of tool calls exceeds the maximum limit.
   2. Your prompts have exceeded the limit. Please open a new tab to continue.

For the first type of issue, I hope it can be improved so that after making modifications, it doesn’t switch to Xcode. That way, I can use this computer for other tasks while it’s working.

For the second category, I would like to add an additional "Agree" option so that I don't have to keep clicking "Agree" each time.

For the last category, I’d like to ask: do all models have the same restrictions? If not, could you indicate which model has a broader range? Also, if the process isn’t completed, isn’t it likely to continue automatically? Why do I need to get involved? After all, if the prompt really does reach its maximum length, you could simply prompt me to open a new tab then—it wouldn’t be too late. In my view, it’s quite easy to reach the limit anyway.

Here are my suggestions. The most important improvement I hope for is the first one, because the interference it causes is really overwhelming. If it's not improved, I can only let the computer run there, and all I can do is sit and watch, unable to do anything else. That's because it keeps switching to Xcode nonstop.