My take on this, FWIW, is that any methods in the under-test API - setUp, tearDown, test_* and anything registered via addCleanup should all support the same protocol as much as possible, whatever it is.

That is, raising a skip in setUp should skip the test. raising a skip in tearDown should skip the test, and raising a skip from a cleanup should skip the test.

This is complicated by the case where some code is called after exceptions- teardown and cleanups. Thats fairly straight forward: errors are higher precedence than failure than skips than anything which resolved as a pass.