What's the proper way to stub tool calls?

[AlexVPopov]

Hey folks,

Let's say one has this tool:

class MyTool < RubyLLM::Tool
  def initialize(my_arg)
    super()
    @my_arg = my_arg
  end
  
  def execute(input)
    MyService.new(@my_arg).call(input)
  end 
end

What's the proper way to stub this? Obviously, I could stub the service:

my_service = instance_double(MyService)
expect(MyService).to receive(:new).with(...).and_return(my_service)
expect(my_service).to receive(:call).with(...).and_return(...)

What I don't like is that this goes into the implementation details of the tool. I was hoping I could do:

my_tool = instance_double(RubyLLM::Tool::MyTool)
expect(RubyLLM::Tool::MyTool).to receive(:new).with(...).and_return(my_tool)
expect(my_tool).to receive(execute).with(...).and_return(...)

However, this fails, because under the hood the stub receives other methods, such as name, parameters etc.. Of course, I could stub all of those, but that breaks encapsulation and is prone to failure in the future (if the implementation changes).

How do you stub tool calls, folks?