I noticed that a simple, open ended task with a lot of time to solve it gives you a nice idea. I'm not sure what kind of task would be applicable for you, but I do a simple database manipulation in python. Some people will just do the minimum that works (or sometimes doesn't), but some will send the whole git repository, show that they did tdd, generate documentation, handle logging/reporting, etc. It becomes obvious quickly who do you talk to.
These are done at home before the interview - I don't think we need to worry about anyone getting outside help to get a perfect answer. With the amount of bad/mediocre answers the onesI'm looking for tend to really stand out. And issues can be verified at the interview.
The trick is to invite people to go as far as they want instead of giving strict boundaries. And lots of time. If it should take an hour to complete, give out a couple of days before the interview anyway. If the answer is really good, the interviewed person will be able to talk for hours about the details.