I need to think about doing it that way as I think ISIS normally does some internal rescaling rather than outputting in ADU counts so the final spectra may not subtract correctly. (you would need to test this) What I was proposing is to produce a single pre processed spectrum image manually with the dark, flat correction and the sky image subtracted already, ready for processing in ISIS without any further dark, flat bias correction or background subtraction to produce the spectrum profile but ISIS may then complain if it is not given any master dark or bias frames.
It seems this might not be as straightforward in ISIS as I thought. Perhaps putting the question on ARAS might be a better bet. Francois Teyssier there would know how to do it in practise.