We're teaching a course at ETH Zurich [1] where --besides the actual payload of solving partial differential equations (PDEs) on GPUs-- we put a lot of emphasis on "tools". Thus students learn how to use git and submit their homework via pushing to a repo of theirs on github, we teach testing and continuous integration, writing documentation, running code on a cluster, etc. In their final project, again submitted as a GitHub repo, they need to make use of all of theses skills (and of course solve some PDEs).

Note that excellent work in this space is done by the Software Carpentry project which exists since 1998 [2].

[1] https://pde-on-gpu.vaw.ethz.ch/ [2] https://software-carpentry.org/