It was an interesting experiment but Bitbucket pipelines is quite limited and testing it is hard. Issues like #1312 scare me a bit.

I'm dreading adding support for GitHub actions, which has a completely different sort of ecosystem, including externalized repos. It is quite dangerous from a security perspective (zizmor would need to be integrated into the test suite to provide assurance that the files usethis generates are secure, but even then it feels like a weak guarantee).

I'm considering the trade-off of continuing to support CI config architecturally versus giving up on it and keeping usethis focused on pure Python config, and leaving CI for the bespoke ways that people like to configure it.

In a commercial context, CI often uploads results to internal servers etc. so it will almost never be fully abstracted in the way that usethis envisages.

Most tools run through pytest and prek/pre-commit so the extra benefit of adding CI config seems very low compared to the complexity of what I'm needing to build/maintain. And it's really quite orthogonal to the main objectives of this project.

It's also a maintenance burden, as schemas change, I have to bump them regularly e.g. #412 #926, #1299, etc.

R usethis has "questioning" lifecycle marked for CI config and doesn't have it for GitHub. And I don't think it tries to do what usethis Python is trying to do right now in terms of dynamic population of the CI config file.