Tip: The `caveman` Skill Claims 65% Output-Token Savings by Making Your Agent Drop Prose, Not Precision

GitHub·medium signalXBlueskyLinkedInCopy link

Caveman (94,167 stars, verified via GitHub API 2026-07-29) is a single skill/plugin for Claude Code, Codex, Gemini, Cursor, Windsurf, Cline, Copilot and 30+ other agents that strips conversational filler from every reply while keeping code, commands, and error strings byte-for-byte exact — the README's worked example compresses a 69-token React re-render explanation to 19 tokens with the same `useMemo` fix. It offers selectable verbosity levels rather than one setting, and the savings are output-side only (input tokens are largely unaffected), which matters if your bottleneck is a subscription output cap rather than context. The 65% figure is the maintainer's own benchmark and has not been independently reproduced, so treat it as directional; the mechanism is plausible and the install cost is one file.