Caveman is a lightweight skill and plugin that makes AI coding agents answer in an intentionally terse, compressed style to reduce output-token usage. It keeps code, commands, and error messages intact while removing conversational filler and shortening explanations, so developers can receive essentially the same technical guidance with less generated text.

The project supports Claude Code, Codex, Gemini CLI, Cursor, Windsurf, Cline, Copilot, and many other agent environments. Users can choose different levels of brevity and switch modes when needed, making Caveman useful for developers who work with coding agents heavily and want lower token consumption, faster reading, and more compact responses without changing the underlying model or coding capability.