A two-step approach for constructing a questionnaire / rating scale is pretty common practice in psychology and test construction. The first step involves getting a list of item candidates, and the best items (according to criteria, including statistic metrics, that fit your specific goals) make it into the final version.
If you can find an existing list that has been tested and fits your tasks well, it's often preferable to use that instead of creating your own. There can be a lot of work involved in creating valid and reliable tests, and for proven tests this has already been done.
To create a good list from scratch it's often desirable to use user input. That way you can find vocabulary that closely matches the user's mental model, and make your items clear to understand. The alternative here is that you decide by yourself on which items should be included, which has its own advantages and disadvantages.
I would have thought a totally open ended approach would be just as valid, ie ask users to list top five tasks without any list whatsoever and then construct a prioritised list...?
I wouldn't ask users to list just the top five, but ask them to name as many tasks as they can think of. If you ask for only 5, you'll find a lot of duplicates among the answers, and you'll need to ask an enormous group of users in order to get to 100 distinct items.