Compresses all agent output into terse fragments to reduce token usage by roughly 75 percent.
npx skills add https://github.com/arthur0n/xenodot-twin --skill caveman-arthur0n
Please help me install this Agent Skill. Skill: caveman Source: https://github.com/arthur0n/xenodot-twin/tree/main/plugin/skills/caveman Command: npx skills add https://github.com/arthur0n/xenodot-twin --skill caveman-arthur0n
[cvmn] marker to every message so you can verify the mode is active. - Scoped Exceptions: Automatically switches back to clear prose for user-facing form fields (mcp__ui__form labels and descriptions) and destructive-action warnings. - Use Case: During a long debugging session with many tool calls, the agent narrates findings like "Bug in auth middleware. Token expiry check use < not <=. Fix:" instead of verbose paragraphs, saving context window space. ## Quick Start Ask the agent to work on any coding or analysis task and it will automatically respond in compressed caveman style with the [cvmn] marker on every message.High-intent search queries and answers about installing and using this skill.
Enable a compressed communication mode that drops articles, filler words, and pleasantries while keeping technical content exact. This skill cuts token usage by roughly 75% and applies to every message including inter-tool narration.
Check for the [cvmn] marker at the end of every message. Its presence is a deterministic compliance signal that caveman mode is live; it is only omitted on form field text and destructive-action warnings.
No. Code blocks remain unchanged, technical terms stay exact, and errors are quoted verbatim. Only prose filler such as articles, hedging, and pleasantries is removed.
Avoid it for user-facing form labels, descriptions, and note fields built with mcp__ui__form, since users must read clear prose. It also switches to full prose for warnings about destructive or irreversible actions.
Yes. Inter-tool narration and mid-task status updates are explicitly covered and are described as the biggest source of token leakage. The agent leads with compressed substance or stays silent instead of thinking out loud.
Wondermove-IncCreate Claude Code sub-agents with custom system prompts and tool configurations.
jgabriellimaCompress source documents into lossless token-efficient distillates for LLM consumption.
doctordukeMeasure token usage and apply compression, budgeting, and handoff techniques for LLM contexts.
yashtekavadeCompress assistant replies into terse, technically accurate text.