Cut ~75% of output tokens by making Claude respond in terse "caveman speak" while keeping full technical accuracy.
The most viral Claude Code skill ever. Caveman makes Claude respond in extremely terse language, dramatically cutting token usage and speeding up responses. Features: - ~75% output token reduction - ~3x speed increase on most tasks - Multiple intensity levels: Lite, Full, Ultra - Classical Chinese mode for fun - Caveman-commit: terse git commits - Caveman-review: fast code reviews - Caveman-compress: cuts ~46% of input tokens too Works with Claude Code, Codex, Gemini CLI, Cursor, Windsurf, and more. 27,000+ GitHub stars.
Install this skill from the source repository:
git clone or copy-paste into your ~/.claude/skills/ directory).No comments yet. Be the first to share your thoughts!
Manus-style persistent markdown planning, progress tracking, and knowledge storage with session recovery.
Persistent AI memory plugin — auto-captures everything Claude does, compresses it, and injects context into future sessions.
Official Linear integration — create issues, manage projects, update statuses, and search workspaces.