Skip to main content
affpapa.org
RU EN
skill CX Codex

caveman-token-optimizer — Claude Code skill: shorter agent replies

A skill that makes AI agents answer in caveman-speak, dropping pleasantries and filler. It cuts roughly 65% of output tokens while code and technical terms stay exact.

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓ 
npx skills add reason-machines/trending-skills --skill caveman-token-optimizer -a claude-code
Open on GitHub

What caveman-token-optimizer does

The skill strips greetings, hedging and articles out of an agent's replies. Code blocks, variable names, technical terms and error messages come through unchanged. The description puts the average saving at about 65% of output tokens, with some tasks reaching 87%.

Four intensity levels are available: lite keeps grammar intact, full drops articles and uses fragments, ultra goes telegraphic. The level sticks until you change it or the session ends. It runs in Claude, Codex, Cursor, GitHub Copilot, Gemini CLI and OpenCode.

One thing to know before installing: only output tokens shrink. Reasoning tokens are untouched, so the bill drops less than the benchmark table suggests. And if you need a normal PR description or commit message, turn the mode off first, because it compresses those too.

What it helps with

  • cut output token spend on long sessions
  • remove filler and pleasantries from agent replies
  • get a short answer without losing exact technical terms
  • switch between compressed and normal replies mid-session
  • reproduce the token benchmarks on your own tasks

Example prompts for the agent

  • /caveman full
  • /caveman lite
  • /caveman off

How to install caveman-token-optimizer

1Skills CLI (Vercel)

npx skills add reason-machines/trending-skills --skill caveman-token-optimizer -a claude-code

2Manually: put the files in place

~/.claude/skills/caveman-token-optimizer/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/caveman-token-optimizer/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about caveman-token-optimizer

How do I install caveman-token-optimizer in Claude?

Run npx skills add reason-machines/trending-skills --skill caveman-token-optimizer -a claude-code. The skill file lands at ~/.claude/skills/caveman-token-optimizer/SKILL.md.

What does caveman-token-optimizer do?

It makes the agent reply in compressed caveman-speak, dropping pleasantries and articles. Code, technical terms and error messages stay exact, and output tokens drop by roughly 65%.

What intensity levels does the skill offer?

Four: lite for brevity with grammar intact, full for fragments without articles, ultra for telegraphic style. Set one with /caveman plus the level name, and it holds until the session ends.

Similar skills

All skills for Agent setup: memory, context, orchestration →
token-optimizer D4kooo A procedure set for OpenClaw agents: isolate background tasks, run local RAG, and consolidate context with Reset & Summarize before it balloons. CX Codex +1 495installs caveman Shawnchee Enables an ultra-terse mode that strips pleasantries, narrations, and filler words to save context tokens on every request. CX Codex +1 1,100installs prompt-optimizer getsentry A skill that rewrites system and agent prompts against a set of eval cases instead of gut feeling. Useful when a prompt exists but behaves inconsistently. CX Codex +1 2,300installs skill-optimizer mcollina A skill for people who write and edit skills: it shows why a model ignores yours and whether it makes output worse. CX Codex +1 1,400installs cost-compact-context ruvnet A skill that wraps getTokenOptimizer().getCompactContext() for token cost analysis. Retrieves compacted context from ReasoningBank, reports tokens saved via bridge measurements, and exposes agent-booster availability. CX Codex +1 726installs rtk akillness Installs and wires RTK (Rust Token Killer) so shell command output reaches your agent in compact form. Useful when logs and diffs eat the context window. CX Codex +1 671installs

From the same set42

weixin-agent-sdk This skill wires any AI backend into WeChat through the weixin-agent-sdk TypeScript framework: QR login plus a message loop. CX Codex +1 1,100installs ghostling-libghostty-terminal A skill that walks through building a minimal terminal emulator on the libghostty-vt C library, with Raylib handling windows and rendering. CX Codex +1 1,100installs web-access-claude-skill Lets the agent pick between search, page fetch and driving your real Chrome. Useful when a site sits behind a login or renders on JS. CX Codex +1 965installs agent-flow-visualization Installs the Agent Flow VS Code extension, which draws Claude agent orchestration as a live node graph: tool calls, branches, subagents, timings. CX Codex +1 887installs pump-analyzer-solana A static front end for watching Pump.fun token launches on Solana: a WebSocket feed, price and volume charts, and alerts on your own criteria. CX Codex +1 837installs codex-session-patcher Finds AI refusals in Claude CLI, Claude and OpenCode session files and replaces them so you can resume the conversation. CX Codex +1 798installs wildworld-dataset A skill for the WildWorld dataset: 108M+ frames from an ARPG, 450+ actions, per-frame annotations. Useful when preparing data for world modeling. CX Codex +1 782installs kordoc-korean-document-parser This skill turns Korean HWP 5.x, HWPX and PDF files into Markdown and structured blocks. Useful when you need text, tables or form fields out of a document. CX Codex +1 774installs

Project sponsors

Companies that keep this analytics open
Gold sponsor
Nutra.Club

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor