skills.rest
northguild
northguildOFFICIAL
Explore all skills by this author →
GitHub

caveman

Compresses AI responses into terse caveman-style prose while preserving technical accuracy.

6|Updated Mar 10, 2026
Search Tags:#token efficiency#concise output#token compression#classical chinese#response style#caveman mode#prompt behavior
One-click install
npx skills add https://github.com/northguild/gmt --skill caveman-northguild
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: caveman
Source: https://github.com/northguild/gmt/tree/main/.agents/skills/caveman
Command: npx skills add https://github.com/northguild/gmt --skill caveman-northguild

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Long, filler-heavy AI responses waste output tokens and slow down reading. This Skill compresses every model response into terse caveman-style prose, cutting output tokens by roughly 65% while keeping all technical details, code, and error strings exact. ## Core Features & Use Cases - Six intensity levels: lite, full (default), ultra, plus three Classical Chinese (wenyan) variants for character-level compression. - Persistent session mode: once activated, compression applies to every response until explicitly stopped with "stop caveman" or "normal mode". - Auto-clarity fallback: automatically reverts to normal prose for security warnings, irreversible-action confirmations, and ambiguous multi-step sequences. - Use Case: A developer working through a long debugging session invokes /caveman to get dense, fragment-style answers that keep code blocks and error messages verbatim while dropping all filler. ## Quick Start Ask the assistant to talk like caveman or invoke /caveman to enable compressed responses for the rest of the session.

Frequently Asked Questions about caveman

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I make AI responses shorter and use fewer tokens?▼

Invoke /caveman or say "talk like caveman" to enable compressed responses. The mode drops articles, filler, and hedging while keeping code, error strings, and technical terms exact, cutting output tokens by about 65%.

What compression levels does caveman mode support?▼

Six levels: lite (no filler, full sentences), full (default, drops articles), ultra (bare fragments), and three Classical Chinese variants — wenyan-lite, wenyan-full, and wenyan-ultra — achieving up to 80-90% character reduction.

Does caveman mode change code blocks or error messages?▼

No. Code blocks, function names, API names, CLI commands, and exact error strings are always preserved verbatim. Only prose is compressed; technical content stays exact.

How do I turn off caveman mode?▼

Say "stop caveman" or "normal mode" to revert to standard prose. The mode otherwise persists for the entire session and does not drift back on its own.

When does caveman mode revert to normal prose automatically?▼

It auto-reverts for security warnings, irreversible-action confirmations, multi-step sequences where fragments risk misreading, and when you repeat a question. Compression resumes after the clear part is done.

Does caveman mode work in languages other than English?▼

Yes. It preserves the user's dominant language and compresses style, not language. The wenyan levels specifically target Classical Chinese compression, and grammatical particles in other languages are kept where they carry meaning.

Related Skills

Explore More Software Engineering Skills→
delorenjdelorenj

opencode-controller

Manage Opencode sessions, select models, and switch between Plan and Build agents.

Community
Intermediate
khrorekhrore

context-compression

Compress agent session context into structured summaries with anchored merging.

Community
Advanced
m2ai-portfoliom2ai-portfolio
4

context-cost-line-item-analyzer

Decompose agent session token spend into context-driven and output-driven costs.

Official
Advanced
louderthanthunderx1louderthanthunderx1

caveman

Generate ultra-brief technical responses when users request caveman mode.

Community
Intermediate