Partially related with #1196 it would be nice if all job scripts automatically included set -e or set -o errexit to make sure that the job stops at the first failure (which then in #1196 could not be silently ignored, addressing the concern of the second bullet point).
I know I can use @rfm.run_before or making my own test class to avoid repeating that all the time, yet I think this would be a useful default to have.