caveman Hits 97,000 Stars for a Token-Cutting Skill — and Its README Publishes the JetBrains Test That Drops the Claim From 65% to 8.5%

GitHub·high signalXBlueskyLinkedInCopy link

caveman is an MIT-licensed skill that makes coding agents drop prose filler while leaving code, commands, and errors byte-for-byte identical, installable in one command across 30+ agents with six compression levels. The headline 65% figure comes from 10 chat-style prompts with a 22–87% range; the README also carries an independent JetBrains test across 86 real coding tasks that measured 8.5% savings, because most agentic output is code, not narration. It further discloses that the skill only shrinks output tokens, adds ~1–1.5k input tokens per turn, and can be net-negative on terse workloads — an unusually honest accounting for a repo this viral, and the number builders should plan against is 8.5%, not 65%.