To be good at programming, a lot of knowledge is needed, but "accidental". From practical ones like how to use git, to conceptual ones like cache performance mental model. It's perfectly possible that git is designed with a different CLI, or the common cache line size being 512 bytes. Mathematicians usually don't care about these things, since they are accidental. So they are bad at writing programs that's far away from math.

It's a completely different story when they are writing programs about math. If the tool is good enough, i.e. allowing them to express math ideas in familiar terms, mathematicians are very good at writing math programs. As can be observed in Lean and mathlib.