This is a very labor-intensive solution to the arguments problem, but here goes.

Edit each of the scripts to respond to a set input. For instance "mytool -h" might return a specification of the arguments that the tool accepts and perhaps a description of their effects that you could parse automatically. Perhaps you could use xml. You would want to look out for optional inputs, required inputs, modes, etc.

Since it sounds like you have a lot of existing scripts, you might want to just write a quick perl wrapper for each of the existing tools that could return this description if called with no arguments, or runs the intended tool with whatever arguments are supplied. New scripts that you write could then support this self-specification protocol and do away with the wrapper. Just point your program at the directory of wrappers and it would be able to automatically determine what sort of arguments each tool needs.

Using this method, you wouldn't need some sort of database that would need to be maintained seperately, the tool itself (or its wrapper) would advertise its interface.