Prompt caching could be more efficient if big files are read into context

[manueloverride]

App Version 3.16.0
API Provider Google Gemini
Model Used 2.5 pro

I have a design mode in which the task of the role is to:

• concatenate all current design documents into one huge file.
  

• read that file into context using read_file (300k tokens)
  

Then I interact with the mode by asking questions or giving tasks. However, the first few turns are more expensive than they could be, because apparently the big file doesn't get added to the prompt cache immediately.