I cannot tell exactly from the post or the linked project what type of data these scripts are designed for, but I am going to assume that "Scripts involved in our workflow for detecting CNVs from WGS data using read depth-based methods" refers to short-read Illumina sequencing.
Simulations can be helpful, but they are usually imperfect. Nowadays we have long-read data available. I would suggest obtaining some long-read data for which short-read data is also available, doing CNV detection on both datasets independently, and comparing the results.
The problem with using long-read data is that any differences you find will be conflated with the biases of the technology. So, if you find a difference, then this approach will not tell you if the differences are due to the biases of the technologies, or due to bugs in the scripts. If you use simulations, then this conflation of biases remains.
Short of re-writing the pipeline, the only other suggestion I have is to borrow from software testing. Try constructing a few minimal examples that might also test some corner cases and see if the scripts behave exactly as expected.