caveman
🪨 why use many token when few token do trick — Claude Code skill that cuts 65% of tokens by talking like caveman
Caveman is an AI agent skill and proxy that dramatically reduces token usage by compressing agent responses and input data into ultra-concise "caveman-speak" while preserving technical accuracy. The skill component cuts agent output tokens by 50-65% by eliminating verbose preambles and using terse phrasing, while the optional proxy compresses logs, JSON, diffs, and other input data by 30-90% before the agent reads them. It works with 30+ AI coding agents including Claude Code, Codex, Cursor, Windsurf, Gemini CLI, and Aider through simple installation commands. The compression never touches code, commands, file paths, or error messages—only prose—and all originals are recoverable.
Key Features
Use Cases
- 01Reducing AI coding session costs by 30-65% without changing code quality
- 02Compressing verbose test output and logs before feeding them to agents
- 03Writing terse commit messages and code review comments
- 04Shrinking agent memory files (CLAUDE.md, etc.) while keeping structure intact
- 05Analyzing months of agent history to identify biggest token sinks
- 06Running long debugging sessions with extensive log analysis at lower cost
Related Skills
View moresuperpowers
An agentic skills framework & software development methodology that works.
skills
Skills for Real Engineers. Straight from my .agents directory.
skills
Public repository for Agent Skills
ponytail
Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
caveman — FAQ
What is Caveman and how does it reduce tokens?+
Caveman is a skill and proxy for AI coding agents that cuts token usage by 50-65% by making agents respond in ultra-concise 'caveman-speak' (removing verbose preambles) and by compressing input data like logs and JSON before the agent reads them. Technical content like code, commands, and errors are never compressed.
How do I install Caveman?+
For the skill only, run 'npx skills add JuliusBrussee/caveman -g' then type '/caveman' in your agent. For the full proxy, run 'npm install -g @caveman-ai/cli && caveman setup --install' then 'caveman <agent-name>' to wrap your agent. No account or API key required—it runs entirely locally.
Which AI agents and clients does Caveman work with?+
Caveman works with 30+ agents including Claude Code, Codex, Cursor, Windsurf, Cline, Gemini CLI, Aider, Kilo Code, Qwen Code, opencode, Hermes Agent, OpenClaw, and Pi. The skill works universally; the proxy has native wrappers for the listed agents.
Is Caveman free and does it require API keys?+
The skill is MIT licensed and free forever. The CLI is MIT and the runtime proxy is BSL-1.1 (source-available, free for first-party use, converts to Apache-2.0 in 2030 or after 4 years). No Caveman account or API key needed—your existing provider credentials pass through unchanged.
Does Caveman reduce quality or break code?+
No. JetBrains tested it on 86 real coding tasks and found 'no detectable quality change' (p=0.82). Caveman never shortens code, commands, paths, or error messages—only the prose around them. Security warnings and confirmations return to full sentences automatically.
What are the prerequisites to use Caveman?+
You need Node.js 22.13+ for installation and an existing AI coding agent (Claude Code, Codex, etc.). The proxy runs locally on your machine and requires no cloud service, account creation, or additional API keys beyond what your agent already uses.
How do I install caveman?+
Open the source repository on GitHub and follow its README. caveman is a skill — MCP Agents Market links you directly to the official repo.
Is caveman free?+
caveman is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.