After a number of attempts at using one-shot scripts for grading, the computer scientist will begin to mull over methods of designing a more robust system requiring a shorter development and testing cycle per assignment. In the author's experience, the most obvious steps in grading are those of identifying what files are actually a useful part of a student submission, compiling the student's submission, and writing out the testing report for the student. Since all or nearly all programming submissions will involve these steps, these are the most likely first targets for a more reusable framework for CAA.
If these tasks are automated in terms of a library to be incorporated into a larger, single-use program for grading an assignment, flexibility and transparency are kept very high, but the efficiency of this approach is questionable