|Think about Loose Coupling|
Think before refactoring!
Immediately after reading Perl Best Practices, and before, after reading Perl Medic, and countless other other books, I've wanted to go back and rewrite my code.
Luckily, I have version control, and was able to roll back everything that I broke. Don't change things just for change's sake. I know, I'm a bad one to preach on this one, as I've done 'optimizations' and 'clarifications' and such to the projects that I've worked on way too many times. Yes, it might be a one line fix, but it might introduce other subtle errors, as you found, and it may be something that you might not catch for days, weeks, or even months down the road. At that point, you can end up losing significant time from trying to pretty something up.
So, I'd say that the two rules that you should learn from Perl Best Practices apply to more than just Perl -- version control, and a good library of tests.
Now, there might be times when it's worth fixing up the documentation (I'm going to be handing off this project to someone else, and I want to make sure everything's good), or optimization (the processes is taking more than 2 seconds, and that's unacceptable).
I still cringe every time I look at some of my old code. Instead of risking breaking it, by trying to clean it up, I prefer to just go in, and make sure there's a comment at the top reflecting how long ago it was written, so if anyone starts giving me crap for it, I can point to that.
I've started working in a few of the tips from the book into my work, but if I went back and tried fixing every bad script I've written, it might take me a year, as I deal with cleaning up new messes that I'd make.
Update: Ovid's right -- I've been changing more than one thing at a time, and I haven't been running tests after every change. I guess this is one of the times when having too many tests can be a problem ... I don't mind running 5 min of tests for an install, but it's damned annoying for every 10 sec modification.