AI Briefing75% Fewer Tokens
Key point
Caveman for Claude Code and Codex compresses answers to cut output tokens by about 75%.
Details
Caveman is a skill/plugin for Claude Code and Codex that compresses answers in a "caveman" style, cutting output tokens by about 75%. It keeps the core technical content while trimming away the filler.
The modes are finely divided.
- lite / full / ultra / 文言文: intensity levels for saying the same content more briefly
- caveman-commit: generates short commit messages
- caveman-review: one-line code review
- caveman-help: quick-reference help
- caveman-compress: a compression tool that reduces input tokens read per session by about 46%
In the README benchmarks, across multiple prompts the average went from 1214 → 294 tokens, a 65% reduction, with the reduction ranging from 22% to 87%. However, the Caveman mode itself only reduces output tokens and does not touch the tokens used in the thinking process.
Installation is done with a single line. It supports Claude Code, Gemini CLI, Codex, Cursor, Windsurf, Cline, and Copilot, and provides automatic session activation for Claude Code, Gemini CLI, and repo-local Codex.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.