Save Money & Tokens: Installing and Using Caveman in VS Code, Antigravity, and Claude Code
If you build software using AI coding tools like Claude Code, Cursor, Codex, or VS Code extensions, output token costs can accumulate rapidly. Conversational fluff ("Sure! I would be happy to help you with that..."), long explanatory intros, and filler text consume bandwidth, increase response times, and consume API budgets.
Created by Julius Brussee, Caveman (juliusbrussee/caveman) is an open-source AI
skill that strips conversational filler, instructing your AI assistant to communicate in direct, highly
concise phrases while keeping code, diffs, terminal commands, and technical logic exact.
Why Use Caveman?
- Up to 65% Token Savings: Significantly lowers output token usage on prose, explanations, and reviews.
- Faster Response Latency: Generating fewer output tokens means your editor or terminal yields results much faster.
- Zero Loss in Code Quality: Code blocks, stack traces, and commands remain unaltered. Only conversational fluff is removed.
1. One-Line Auto-Installer (Recommended)
The fastest way to install Caveman across all AI agents configured on your system (including local VS Code extensions, CLI tools, and agent environments) is the global universal install script:
macOS / Linux / WSL:
curl -fsSL https://raw.githubusercontent.com/JuliusBrussee/caveman/main/install.sh | bashWindows (PowerShell):
irm https://raw.githubusercontent.com/JuliusBrussee/caveman/main/install.ps1 | iex2. Platform-Specific Setup Instructions
Option A: Claude Code CLI
If you run Anthropic's official claude CLI terminal agent, you can install Caveman directly via
the plugin marketplace:
claude plugin marketplace add JuliusBrussee/caveman
claude plugin install caveman@cavemanOnce installed, Caveman activates automatically on your next prompt in Claude Code.
Option B: VS Code (Cursor, Windsurf, Copilot, Cline, Continue)
To register Caveman across VS Code extensions and AI coding environments using the global skills registry, run:
npx skills add JuliusBrussee/cavemanAlternatively, for individual VS Code extensions that support custom instructions or prompt templates (e.g.,
Cursor .cursorrules, Cline, or Copilot instructions), clone the repository and add it to your
local environment:
git clone https://github.com/JuliusBrussee/caveman.gitOption C: Antigravity & Web AI Clients
For custom web clients, API proxy layers, or Antigravity agents, load the Caveman rule definition into your system prompt context or prompt modifier stack. You can trigger it natively by appending the rule set or issuing the instruction:
Type /caveman or add instruction: "Talk like a caveman. Drop filler words, keep technical answers precise, preserve code."3. Compression Modes & Usage Commands
Caveman supports multiple verbosity levels depending on how concise you want your AI output to be:
/caveman lite– Removes unnecessary conversational filler while maintaining standard sentence structures./caveman full– Default mode. Direct shorthand ("New object ref created each render. Wrap in useMemo.")./caveman ultra– Extreme telegraph-style compression for maximum speed and token reduction.normal mode– Restores standard conversational output.
Before and After Comparison
Standard AI Output:
"Sure! I would be happy to help you fix that issue. The reason your React component is re-rendering is likely because you are creating a new object reference on every render cycle. When you pass an inline object as a prop, React's shallow comparison sees it as a new object, which triggers a re-render. I recommend using useMemo to memoize it."
Caveman AI Output:
"Inline object prop creates new reference every render. Triggers shallow comparison re-render. Wrap
object in useMemo."
Written by A.M. Rinas · Published August 8, 2026
