I forked ribbon, the small usage band above the Claude Code prompt. Mine follows me between accounts, shows my usage before I've typed anything, and says something useful when I run a local model. It can also shrink to just the numbers, in grey, without the dots.

I tested 2 public ambition skills blind on my own work. On the first 6 tasks neither scored higher than no skill. A shorter rewrite did. Here's how I measured it and what it doesn't prove.