Caveman Skill Intensity Levels: The Complete Guide (6 Modes Explained)

The Caveman skill offers six distinct intensity levels—lite, full (default), ultra, wenyan-lite, wenyan-full, and wenyan-ultra—that control how aggressively the model compresses prose into terse, caveman-style output.

The Caveman skill trims model responses into minimalist, high-density prose inspired by "caveman" speech patterns. Depending on which Caveman skill intensity level you select, the output ranges from lightly edited professional sentences to extreme telegraphic fragments—or even classical Chinese registers. This guide breaks down each level's behavior, how to invoke them, and where they're defined in the source code.

How Intensity Levels Work

The Caveman skill's intensity system is a rule-based compression engine that applies different rewrite strategies based on your chosen setting. Each level progressively strips filler words, grammatical scaffolding, and syntactic complexity.

The six levels fall into two families:

• Modern English modes: lite, full, ultra
• Classical Chinese (文言文) modes: wenyan-lite, wenyan-full, wenyan-ultra

All six are documented in skills/caveman/README.md and exposed through the /caveman command interface defined in skills/caveman/SKILL.md.

The Six Caveman Intensity Levels Explained

lite: Professional Concision

Removes filler and hedging while preserving complete sentences.

• Cuts phrases like "I think," "perhaps," and "it seems that"
• Output remains grammatically intact and workplace-appropriate
• Best for: Executive summaries, quick briefings where tone matters

full: Balanced Compression (Default)

Drops articles, permits short synonyms, and allows fragmentary phrases.

• This is the default intensity when you invoke /caveman without arguments
• "The quick brown fox" becomes "Quick brown fox"
• Indicated in src/rules/caveman-openclaw-bootstrap.md as the baseline behavior

ultra: Maximum Density

Produces bare fragments, aggressive abbreviations, and arrow-based causality.

• Abbreviations: DB (database), auth (authentication), fn (function), ref (reference)
• Arrows (→, =) replace "causes" or "leads to"
• Grammatical sentences are abandoned entirely
• Best for: Speed-reading dense technical material, personal notes

wenyan-lite: Classical Chinese, Light Touch

文言文 register with moderate compression.

• Applies classical Chinese lexical choices and grammatical structures
• Retains slightly more connective tissue than deeper wenyan modes

wenyan-full: Full Classical Compression

Maximum classical-Chinese density while remaining parseable.

• Heavy use of literary particles (之, 其, 而)
• Parallel structure and extreme ellipsis

wenyan-ultra: Extreme Classical Condensation

Most compressed output in any mode.

• Approaches the density of historical Chinese commentaries
• Requires familiarity with 文言文 conventions

How to Switch Intensity Levels

Invoke any level using the /caveman command with the level name as an argument:

# Default full intensity

/caveman

# Explicit full (same as default)

/caveman full

# Lighter touch

/caveman lite

# Extreme compression

/caveman ultra

# Classical Chinese variants

/caveman wenyan-lite
/caveman wenyan-full
/caveman wenyan-ultra

# Disable and return to normal prose

stop caveman

The command syntax is specified in skills/caveman/SKILL.md, which contains the complete LLM-facing instruction set for parsing these switches.

Auto-Clarity Safeguard

The Caveman skill includes a context-awareness override to prevent dangerous ambiguity. When the model detects:

• Security warnings or safety-critical content
• Multi-step procedural instructions
• Repeated user questions (indicating possible confusion)

…it temporarily exits caveman mode, renders the response in normal prose, then reactivates your selected intensity for subsequent turns. This safeguard is implemented as part of the skill's rule engine referenced in the bootstrap configuration.

Output Comparison: Same Question, Different Intensities

Consider the question: "Why does my React component re-render?"

Intensity	Output
Full	"New object ref each render. Inline object prop = new ref = re-render. Wrap in useMemo."
Ultra	"Inline obj prop → new ref → re-render. useMemo."

The ultra version cut 7 words to 6 tokens while preserving all technical meaning—a compression ratio of roughly 60% without losing actionable guidance.

Where Intensity Levels Are Defined

File	Purpose
skills/caveman/README.md	Documents all six intensity levels with usage examples [Link]
skills/caveman/SKILL.md	Contains the LLM-facing instructions including the /caveman [level] command syntax [Link]
src/rules/caveman-openclaw-bootstrap.md	References the default full intensity and switching mechanism [Link]

These files together form the authoritative specification for the Caveman skill's intensity system.

Summary

• Six Caveman skill intensity levels exist: lite, full (default), ultra, wenyan-lite, wenyan-full, and wenyan-ultra
• Modern English modes compress progressively; wenyan modes apply classical Chinese registers at parallel intensities
• Switch levels with /caveman [level]; disable with stop caveman
• Auto-clarity safeguard pauses caveman mode for safety-critical or confusing contexts
• Source definitions live in skills/caveman/README.md, skills/caveman/SKILL.md, and src/rules/caveman-openclaw-bootstrap.md

Frequently Asked Questions

What is the default Caveman skill intensity level?

The default is full. When you run /caveman without specifying a level, the skill applies standard compression: articles dropped, short synonyms allowed, fragmentary phrases permitted. This default is established in src/rules/caveman-openclaw-bootstrap.md.

Can I use Caveman intensity levels with languages other than English?

Yes—the wenyan-* modes are specifically designed for Classical Chinese (文言文). These apply the same compression philosophy using historical Chinese lexical and grammatical conventions. No other modern languages have dedicated modes in the current implementation.

Does the ultra intensity level ever hurt comprehension?

For technical readers familiar with the domain, ultra typically preserves sufficient signal. However, the auto-clarity safeguard automatically disables caveman mode when it detects potentially dangerous contexts (security warnings, procedural instructions) where ambiguity risks outweigh brevity benefits.

How do I completely turn off the Caveman skill?

Send stop caveman (without the slash prefix). This command exits caveman mode entirely and restores normal prose generation for all subsequent turns until you re-invoke /caveman.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}