Caveman
Token-cutting skill and local proxy that makes coding agents terse
What it is
Caveman shortens what AI coding agents write and read: a skill makes replies terse while leaving code, commands, and errors verbatim, and a local proxy compresses logs, JSON, test output, and web pages before the model sees them. It also adds per-request model selection and reusable agent scripts to lower cost. Developers running Claude Code, Codex, Gemini CLI, OpenCode, and other agents install it with one npx command to cut token spend on their own keys.