In the real world situation that you've got, I'd write tests to cover everything that you've documented, plus corner cases and places where you know your code might be a bit dodgy, and stop when you think you've covered everything. Then when you find a bug, you write a test to cover it and then change the code to fix it - and also make sure you still pass all your old tests.

You might find some of the code coverage modules on the CPAN useful - if they find code that's not covered when you run your tests, that's a sure sign that you need more tests.