Comment by buremba

Since prompt caching won't work across different models, how is this approach better than dropping a PR for the other harnesses to review?