Radar / AI coding tools / Caveman
Caveman: the viral agent skill that makes coding agents talk in telegraphic fragments
An agent skill plus local proxy that forces coding agents to answer in short, plain fragments and shrinks the context they read, cutting token use while keeping code and commands byte-exact.
Why it matters
The joke that became infrastructure: 120,000+ GitHub stars, 626,000+ installs, a v4.1.0 release on 8 Nov, and GitHub Trending #1 again this week, with JetBrains and Adobe Research both putting it to the test.
What you could build with it
A startup running coding agents on metered API plans could install the skill across its team so every agent reply stays terse, cutting token spend on long sessions without changing the underlying model or workflow.
Does it hold up?
Widely adopted and genuinely usable: one-command install across 30+ agents, but real savings are modest — JetBrains measured 8.5% fewer output tokens in an independent test — versus the 65% headline the project's own Honest Numbers page now walks back.
Built with Caveman
- Caveman Review: The Claude Code Skill That Cuts 65% of Tokensarticle · Independent review re-checked against the repo on 12 Nov: v4.1.0, 119,000+ stars, and the honest note that the project's own benchmark shows a smaller median saving than the headline.
- Caveman Skill Token Savings: What v4.0 Installs and Which Numbers Hold Uparticle · Evidence audit of the 65% claim against the project's own Honest Numbers page, the JetBrains A/B test (8.5% fewer output tokens), and the Adobe CAVEWOMAN paper.
- Caveman Skill for Claude Codearticle · Catalog entry tracking 626,578 installs and 120k stars as of 14 Nov, with install instructions and the skill's coverage across agents.
Learn more
First spotted on github: source.
More AI coding tools
Chrome DevTools MCP: give your coding agent a real browserOfficial MCP server from the Chrome team that lets coding agents drive a live Chrome — console, network, traces…coding · JEV 0.82Orca: the agent development environment for running coding-agent fleetsStably AI's open-source desktop app for running many CLI coding agents side by side — Claude Code, Codex, Cursor and…coding · JEV 0.78Open Code Review: Alibaba's hybrid AI code reviewer, now past 50,000 starsAlibaba's open-source code review CLI (Apache 2.0) pairs deterministic rules with an LLM agent to review git diffs…coding · JEV 0.77Astryx: Meta's agent-ready design system, open sourceMeta open-sourced the React 19 + StyleX design system behind 13,000 internal apps: 150+ accessible components, themes…coding · JEV 0.76GitHub Spec Kit — spec-driven development for coding agentsGitHub's open-source toolkit that gives any coding agent a structured spec, plan, tasks, implement workflow, so specs…coding · JEV 0.75shadcn/improve: audit with your best model, execute with your cheapestAn MIT-licensed Agent Skill that audits any codebase across nine categories with parallel subagents, vets its own…coding · JEV 0.75
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.