JuliusBrussee/caveman

Strip away the polished prose and what remains is raw instruction. This open-source Claude Code skill rewires how the AI communicates internally, replacing verbose, grammatically complete responses with compressed caveman-style output. The result is a claimed 65 percent reduction in token consumption per session, which translates directly into lower API costs and faster response cycles for anyone running Claude Code heavily throughout a workday. The setup is minimal. You install the skill, and Claude starts responding in stripped-down language that still conveys the essential meaning but ditches filler words, pleasantries, and over-explanation. For developers who live inside Claude Code for coding tasks, the savings compound quickly across dozens of daily interactions. The honest reservation is that caveman output can feel jarring in collaborative settings or when you need Claude to generate something readable by non-technical stakeholders. This is purely a solo-developer optimization, not a team tool. That caveat aside, for cost-conscious builders grinding through high-volume coding sessions, the token math is straightforwardly attractive. -> Best for: solo developers using Claude Code daily who want to stretch API budgets without switching models.