scikit-learn's commit log tends to use prefixes on commit messages that don't seem to be documented anywhere. (The developers' guide doesn't say anything about commit messages; it does say every issue and PR should have tags, but non-priveleged users are not permitted to add them.)

The listing over at numpy is useful (http://docs.scipy.org/doc/numpy/dev/gitwash/development_workflow.html), but doesn't correspond to what is used here (e.g. BUG vs FIX, MAINT/STY vs COSMIT).

At the same time, these should be explained: is their purpose to encourage atomic commits (i.e. don't do ENH and TST in the same commit)? or for analytics (well, the current informality won't make that much use)? or is it an irrelevant hangover from related projects (even so, n00bs should have some explanation so they don't spend too long confused)?

The meaning of WIP and MRG prefixes on PR titles similarly seems to be unexplained. (It might also be useful if the developers' guide introduced the acronym PR.)