Caveman skills for the agentplugins CLI — ~75% fewer output tokens, full technical accuracy.
A lean integration layer bringing caveman's output compression into the agentplugins workflow. For the full skill system, benchmarks, and install options → caveman.
# install via agentplugins
pnpm dlx agentplugins install agentplugins-caveman
# activate in session
/caveman| Command | What |
|---|---|
/caveman [lite|full|ultra] |
Compress every reply. Levels stick until session end. |
/caveman-commit |
Conventional commits, ≤50 char subject. |
/caveman-review |
One-line PR comments: L42: bug: user null. Add guard. |
/caveman-stats |
Real token usage + savings for this session. |
/caveman-compress <file> |
Rewrite memory files (CLAUDE.md, etc.) into caveman-speak. |
cavecrew-* |
Caveman subagents — ~60% fewer tokens than vanilla. |
- ~75% output reduction — same fix, fewer words.
- Full accuracy — only the mouth shrinks, not the brain.
- Persistent — compress once, save every session.
- Works everywhere — agentplugins CLI + any agent that speaks SKILL.md.
For benchmarks, full install options, and all the detail → caveman.
MIT