I guess the best would be to start small, and avoiding changing any code before you have some basic testing to verify the work done inside the black box, and when that is in place, consider refactoring the code so that individual modules / subs can be verified.

To do that, you could start verifying the data the script is supposed to work on is what they are supposed to be. Then that the result is what it should be for a given set of input data.

How to do that is quite dependent on what the scripts do, and where they get the data and where they put the result. It's greatly simplified if the datasource and result locations are parameters to the script. If it's not, you might have to alter the scripts a litle to be able to give alternate params for test runs, but in such a way that other callers are not affected.

You need to tell more about the spesifics...