[w0m]

I'd just do it over a Docker mount (or equivalent) to keep it a bit more lightweight. Can keep the LLM running local; and teach it how to test/debug via instruction files.