Talk like cave humans
Talk like cave humans
Agents tend to write a lot, from subagents spawning context and MCP requests/outputs to outputs shown to users, which increases token consumption.
The caveman stack counters this by acting at different places within agent runtimes.
Warning
Hooks activation is limited to Claude Code by default, but a mirrored plugin can be used to broaden agent runtime compatibility.
- Upstream: https://github.com/JuliusBrussee/caveman
- What is it: A skill suite for compressed outputs, primarily for chat output, documentation, commits, code review, and hooks to automatically activate caveman output mode.
claude plugin marketplace add kilianpaquier/ai-integration
claude plugin install cave-talk@one-for-allapm marketplace add kilianpaquier/ai-integration
apm install cave-talk@one-for-all -gapm install kilianpaquier/ai-integration/plugins/cave-talk -gnpx skills add kilianpaquier/ai-integration