This is exactly what it sounds like: Claude drops all the filler words, articles, and pleasantries to cut token usage by about 75%. It stays on until you explicitly turn it off, which is smart because otherwise it would drift back to verbose mode after a few exchanges. The format is brutally efficient (think "Bug in auth middleware. Token expiry check use < not <=. Fix:") but it keeps technical accuracy intact and includes a sensible exception for security warnings or destructive operations where clarity matters more than brevity. If you're burning through tokens on a long debugging session or working with a tight context window, this will stretch your budget. Just be ready for responses that sound like a very smart caveman wrote your code review.
npx -y skills add vinvcn/mattpocock-skills-zh-cn --skill caveman --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
juliusbrussee/caveman
mattpocock/skills
shadcn/improve
obra/superpowers
forrestchang/andrej-karpathy-skills
vercel-labs/skills