Pathwise AcademyOnline & In-Person AI ClassesIdea to BuildCreators Club · Ages 10–17Marketing with AI · AdultsCorporate WorkshopsHackathons for SchoolsSingaporeHong Kong
Pathwise AcademyAI classes for kids & adults
Find a class →Log in
← All free guides
Building appsBeginner · about 20 minFree · no signup

Test whether the Caveman skill saves you tokens

Caveman is a free, open-source skill that makes AI coding agents answer in short caveman-style lines. Here's how to try it and measure whether it actually saves you anything.

By Alan Brown · Works with Claude Code

Jump to the prompt ↓

🪨 Why use many token when few token do trick? This free skill makes Claude talk like a caveman to save you money.

  • 🧠 It's called Caveman. It's free and open source (Apache-2.0) on GitHub with over 100,000 stars: github.com/JuliusBrussee/caveman
  • ✂️ It makes your AI coding agent drop the filler. In the repo's example, a normal answer took 63 tokens and the caveman answer took 20, for the same fix.
  • 💸 Tokens are what you pay for, so fewer words out can mean a smaller bill and a plan that goes further.
  • 🔒 Code, commands and numbers stay exact, and it switches back to full sentences for security warnings.
  • ⚙️ One line installs it, and it works with Claude Code, Codex and 30+ other agents.

Context & limitations

The repo has advertised a 65% cut in output tokens. JetBrains tested it on 86 real coding tasks and measured 8.5% fewer output tokens, with no measurable drop in quality (blog.jetbrains.com/ai/2026/07/speak-to-ai-agents-like-cavemen-tosave-tokens/). Your saving depends on your work, so test it on your own tasks.

Do this yourself

I want to test the Caveman skill (https://github.com/JuliusBrussee/caveman) on my own coding work and see if it really saves me tokens. First, check the repo's README for the current install command for ___ (my agent: Claude Code, Codex, Cursor or another), and tell me what it will change on my machine before I run it. Then help me run a fair test: 1. Ask me for 5 small, real tasks from my project (a bug fix, a refactor, a test, an explanation, a config change). One task at a time. 2. For each task, run it once normally and once with /caveman on, with the same model and the same starting point. 3. Record the output tokens (or reply length if my tool doesn't show tokens) for both runs in a simple table. 4. Check that both versions of the code work, and point out any place where the caveman reply left out something important. At the end, tell me the average saving, whether quality changed, and whether I should keep it on, use it only for some tasks, or turn it off. If my plan charges per request instead of per token, tell me that shorter answers may save me nothing.

You pay for every word your AI coding agent writes back. Shorter replies can mean a smaller bill, but how much you save depends on your work, so it's worth testing on your own tasks before you trust the headline number.

Want to build this with a teacher?

Pathwise Academy runs small, hands-on AI classes in Singapore for ages 10–17 and adults.