Ponytail
Coding Workflows
Makes Claude write the smallest code that works, cutting tokens and cost by about half.
Makes Claude say less and read less, cutting token use by up to 75% with the same answers.
Why many token when few do trick. Caveman makes Claude answer in a terse, answer-first voice: one idea per sentence, no greeting, no recap. Code, commands, paths and error messages stay untouched, character for character.
Version 4 is six parts in one install. Say less shrinks replies (a 63-token answer becomes 20). Read less compresses logs, JSON and test output before Claude reads them, measured at a third fewer input tokens per session. Auto picks the model and effort for each task, and Blocks keeps the scripts Claude writes so it can rerun them instead of rewriting them.
It has been cited by Adobe Research, A/B tested by JetBrains with no measurable quality loss, and hit number one on Hacker News and GitHub Trending.
Run npx @caveman-ai/cli@latest in your terminal (needs Node 22.13 or newer). It finds Claude Code and your other agents, shows what it will change and switches caveman on. Inside Claude Code use /caveman, /ultracave or /megacave for progressively shorter replies.
/caveman why does my React component re-render on every keystroke?Anyone on a Claude plan who hits usage limits, or who pays per token.
Why it made the list
It is the easiest way to stretch a Claude Pro or Max plan further without losing accuracy.
Coding Workflows
Makes Claude write the smallest code that works, cutting tokens and cost by about half.
Productivity
Forces Claude to lead with the answer, number the steps and skip the padding.
Coding Workflows
A template that gives Claude full context up front so features ship with fewer re-prompts.
Productivity
A directory of 1000+ Claude skills plus a plugin that lets Claude act in 1000+ apps.