Caveman: open-source tool cuts AI coding agent output by 65%

Caveman is an open-source skill that makes AI coding agents answer in short, blunt sentences without filler. In the author's benchmark on the Claude API, output tokens dropped by 65% on average, though savings ranged from 87% on explanations to 22% on refactoring.

• Install with one command: npx skills add JuliusBrussee/caveman
• Works with 30+ agents: Claude Code, Codex, Gemini CLI, Cursor, Copilot
• Modes /caveman lite, full, ultra, off plus commit, review, stats commands
• Optional local proxy cut input tokens by about 33% across 54 runs