As far as I understand it there are multiple levels to this. One of the mechanisms is that the cells don't necessarily know where to stop or where to fold - but as everything is being done in decentralised manner the cells reach boundaries of other structures and this will then "shape" the structure. It's an extremely efficient data compression where the instructions many times don't include specific measures or counts (e.g. of fingers) but are more about just timing or symmetry breaking. Imagine construction workers being told to just keep making a wall until they can't build anymore because they will hit another wall (and the other wall is being built with similar instructions). The limits can be spatial, chemical, probably bioelectric...

One aspect I've learned about is what I wrote in a different comment:

Cells in different locations produce different chemical markers. These markers spread out through the tissue and create a sort of gradient. By observing the gradient, a cell can know where it is (or at least what it's supposed to do). At least that's the explanation as I've understood it.

Michael Levin is working on figuring that out, check out his work on bioelecteicity