I realize this may be splitting hairs, but pedantically there's nothing in GitLab CI's model that requires shell; it is, as best I can tell, 100% docker image based. The most common setup is to use "script:" (or its "before_script:" and "after_script:" friends) but if you wanted to write your pipeline job in brainfuck, you could have your job be { image: example.com/brainfuckery:1, script: "" } and no shell required[1]

1: although TIL that the "script:" field itself is actually required in GLCI https://docs.gitlab.com/ci/yaml/#script