Cut AI Coding Assistant Token Use by 75% with Caveman’s Minimalist Output
Caveman is an open‑source plugin for AI coding assistants that removes redundant phrasing, cutting output tokens by up to 75% and speeding responses threefold, while preserving code blocks, error messages, and technical terms, and offering multiple intensity levels and specialized commands to streamline development workflows.
Problem
AI programming assistants often add filler words, polite phrases, and other redundant text, which inflates token consumption, slows response time, reduces reading efficiency, and lacks flexible verbosity control.
Project Overview
Caveman (the “caveman mode”) is an open‑source skill/plugin compatible with Claude Code, Codex, Gemini CLI, Cursor, Windsurf, Copilot, Cline and other assistants. Its goal is “extreme concise output while keeping technical precision”, delivering the same technical information with far fewer tokens.
Core Features
1. Minimalist Output Mode
Automatically strips filler words, articles, hedging phrases, etc., reducing output tokens by about 75 % while keeping code blocks, technical terms, error messages, and Git commit/PR descriptions unchanged (“caveman not stupid”).
2. Adjustable Intensity Levels
Lite – concise, retains full syntax, suitable for readable communication.
Full – default, removes articles and uses short fragments.
Ultra – telegram‑style abbreviations for maximum token saving.
Wenyan – classical Chinese style, also with Lite/Full/Ultra sub‑modes.
3. Specialized Skills
caveman-commit– generates minimal Conventional Commits (≤50 characters) focusing on “why” the change was made. caveman-review – produces a single‑line code review comment without filler. caveman-compress – compresses input files (e.g., project docs) by ~46 % while keeping a backup. caveman-help – displays all modes, skills, and commands.
Value Proposition
Cost reduction: ~75 % fewer output tokens and ~46 % fewer input tokens lower API expenses.
Speed boost: response time improves roughly threefold.
Accuracy preservation: 100 % of technical content retained; studies show short answers can raise accuracy by 26 % on certain benchmarks.
Fun factor: “caveman” style adds humor to code reviews.
Quick Start
Installation requires a single command per assistant; examples:
# macOS / Linux / WSL
bash <(curl -s https://raw.githubusercontent.com/JuliusBrussee/caveman/main/hooks/install.sh)
# Windows PowerShell
irm https://raw.githubusercontent.com/JuliusBrussee/caveman/main/hooks/install.ps1 | iexInstallation commands per assistant
Claude Code –
claude plugin marketplace add JuliusBrussee/caveman && claude plugin install caveman@cavemanCodex – clone the repository, place it in /plugins, and search “Caveman”.
Gemini CLI –
gemini extensions install https://github.com/JuliusBrussee/cavemanCursor – npx skills add JuliusBrussee/caveman -a cursor Windsurf – npx skills add JuliusBrussee/caveman -a windsurf Copilot – npx skills add JuliusBrussee/caveman -a github-copilot Cline – npx skills add JuliusBrussee/caveman -a cline Other assistants – npx skills add JuliusBrussee/caveman (auto‑detect).
Basic usage
Activate mode with any of:
/caveman
talk like caveman
caveman mode
less tokens please
Deactivate with:
stop caveman
normal mode
Switch intensity:
/caveman lite
/caveman full
/caveman ultra
/caveman wenyan
/caveman wenyan-lite
/caveman wenyan-ultra
Specialized skills usage examples:
Generate commit: /caveman-commit One‑line review: /caveman-review Compress file: /caveman:compress filename Help:
/caveman-helpRepository
https://github.com/JuliusBrussee/caveman
AI Architecture Path
Focused on AI open-source practice, sharing AI news, tools, technologies, learning resources, and GitHub projects.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
