Cuts your token usage by around 75% by making Claude respond like a technically precise caveman. Drops articles, filler words, and pleasantries while keeping all the technical accuracy intact. You get three intensity levels: lite keeps full sentences, full goes classic caveman with fragments, and ultra abbreviates everything down to arrows and single words. There's also a wenyan mode that does the same thing in classical Chinese if that's your thing. It stays on until you explicitly turn it off, which is great for long debugging sessions where you're burning through context. The auto-clarity feature is smart, it drops the caveman voice for security warnings or destructive operations where ambiguity could cause real problems.
npx skills add https://github.com/juliusbrussee/caveman --skill caveman-es