Caveman
Terse caveman-style answers and smaller context reads that cut token usage — with an honest ~3% median gain.
About
Makes the agent answer in terse caveman-style fragments and shrinks what it reads (logs, JSON, CSV), auto-picking model and effort to cut token usage. Honest note: the author's own README admits /caveman is only ~3% better than plain "answer concisely" at the median (the big 65% figure is vs no instruction at all), and the CLI telemetry is on by default.
npx @caveman-ai/cli@latestRequires Node 22.13+.
Or install the skill globally:
npx skills add JuliusBrussee/caveman -gPlugin:
/plugin marketplace add JuliusBrussee/cavemanMore in Claude Skills
View all
Anthropic Agent Skills
Anthropic's official skill collection — documents, spreadsheets, design and more.
Claude Red
78 offensive-security skills in 23 categories as drop-in SKILL.md playbooks that turn Claude Code into a red team operator.
Claude Code Templates
Open registry + CLI for Claude Code components — skills, agents, commands, MCPs and more, installable one at a time.