JuliusBrussee/caveman
Token-saving skill and proxy for coding agents
- Chart
- Not on a chart right now
- Stars
- 121.4k
What it is
Caveman is a skill and proxy for AI coding agents that cuts token usage by shortening agent replies and compressing the data they read. It keeps code, commands, and file paths verbatim while trimming unnecessary prose.
Who it is for
Developers using AI coding agents like Claude Code, Codex, Gemini CLI, or OpenCode who want to reduce token costs.
Why it is trending
It gained 2,258 stars today, reaching over 120,900 total stars, following the recent Caveman 4.1.0 release. Unlike simple conciseness prompts, it combines terse output rules with input file compression and automatic model-and-effort routing.
Main features
- Terse output modes (/caveman, /ultracave, /megacave) that preserve code verbatim
- Input compression for logs, CSV, YAML, JSON, and test output
- Auto mode to pick the model and effort level per request
- Blocks code mode to save and reuse scripts written by the agent
Summarized on November 17, 2026 from the README and repository data. It can be wrong; check the source before relying on it. Report a wrong summary