Caveman

Drops your token usage by about 75% by stripping articles, filler words, and pleasantries while keeping technical accuracy intact. Once you trigger it with "caveman mode" or similar, it stays on across all responses until you explicitly turn it off. Works well for long troubleshooting sessions or when you're burning through context windows on repetitive tasks. The persistence is the real feature here, you don't need to remind it every turn. It makes an exception and switches back to clear language for security warnings or destructive operations, then resumes. Reads a bit like terse shell output, which honestly fits the vibe of deep technical work better than chattiness anyway.

Install to Claude Code

npx -y skills add mattpocock/skills --skill caveman --agent claude-code