What Are the Different Skills Available in the JuliusBrussee/caveman Project?

The JuliusBrussee/caveman project ships seven distinct Claude Code skills—caveman, caveman-stats, caveman-review, caveman-help, caveman-compress, caveman-commit, and cavecrew—that reduce token usage through conversational compression, session analytics, automated code review, file compression, commit generation, and sub-agent delegation.

The JuliusBrussee/caveman repository implements a modular plugin architecture where each skill is declared with a SKILL.md front-matter file stored in the skills/ directory. The plugin loader reads these definitions to expose slash commands and define the contract between Claude and the skill implementation. Understanding the different skills available in the JuliusBrussee/caveman project helps developers maximize efficiency across both input context and output generation.

Available Skills in the JuliusBrussee/caveman Project

caveman: Conversational Compression

The caveman skill transforms Claude's output into compressed "caveman-style" prose by removing filler words, articles, and hedging while preserving exact code, URLs, and technical terms. It supports multiple intensity levels including lite, full (default), ultra, wenyan-lite, wenyan-full, and wenyan-ultra.

Core definition resides in skills/caveman/SKILL.md with usage documentation in skills/caveman/README.md. Trigger via /caveman or natural language cues like "talk like caveman".

caveman-stats: Session Analytics

The caveman-stats skill reads Claude's session log to report actual token usage and estimated savings versus a non-caveman baseline. This provides quantitative validation of the compression benefits.

Definition files: skills/caveman-stats/SKILL.md and skills/caveman-stats/README.md. Invoke with /caveman-stats.

caveman-review: Automated PR Reviews

The caveman-review skill generates one-line PR review comments in the strict format L<line>: <severity> <problem>. <fix>. using emoji severity indicators. It preserves exact line numbers and symbols while minimizing token count.

Configuration lives in skills/caveman-review/SKILL.md and skills/caveman-review/README.md. Trigger via /caveman-review or phrases like "review this PR".

caveman-help: Command Reference

The caveman-help skill displays a quick-reference card showing all caveman modes, sibling skills, and deactivation triggers. This one-shot display never changes the active mode, providing immediate command documentation.

Source files: skills/caveman-help/SKILL.md and skills/caveman-help/README.md. Access via /caveman-help or "caveman help".

caveman-compress: File Compression

The caveman-compress skill reduces large natural-language project-memory files such as CLAUDE.md, notes, and todos into caveman representation. This halves the token load on every session start while keeping headings, code blocks, URLs, and file paths intact.

Implementation defined in skills/caveman-compress/SKILL.md with detailed benchmarks in skills/caveman-compress/README.md. Usage: /caveman-compress <filepath> (e.g., /caveman-compress CLAUDE.md).

caveman-commit: Conventional Commit Generation

The caveman-commit skill analyzes diffs to generate Conventional Commit messages with subjects ≤ 50 characters and bodies ≤ 72 characters. It performs read-only operations—never staging or amending changes—returning only the formatted message string.

Defined in skills/caveman-commit/SKILL.md and skills/caveman-commit/README.md. Invoke via /caveman-commit or trigger phrases like "write a commit".

cavecrew: Sub-Agent Delegation

The cavecrew skill provides a decision-guide for delegating work to specialized sub-agents: cavecrew-investigator, cavecrew-builder, and cavecrew-reviewer. It minimizes context usage by keeping sub-agent output terse through caveman-compressed formatting.

Contract specified in skills/cavecrew/SKILL.md with agent wiring documentation in skills/cavecrew/README.md. Trigger via delegation phrases like "delegate to sub-agent" or "spawn investigator".

How to Invoke Caveman Skills

All skills reside under the top-level skills/ directory and are automatically loaded by the caveman plugin at runtime. The modular design allows combination of skills to maximize token efficiency across both Claude's output and input contexts.

# Activate full-mode caveman output (default)

/caveman

# Switch to lighter compression

/caveman lite

# Switch to ultra-terse mode

/caveman ultra

# Display command cheat-sheet

/caveman-help

# Report current session token usage

/caveman-stats

# Generate one-line PR review comment

/caveman-review

# Create Conventional Commit message from diff

/caveman-commit

# Compress project memory file

/caveman-compress CLAUDE.md

# Delegate to sub-agents (natural language trigger)

delegate to subagent

Summary

• The JuliusBrussee/caveman project provides seven specialized skills that optimize Claude Code interactions for token efficiency.
• Each skill is defined by a SKILL.md front-matter file in the skills/<skill-name>/ directory that exposes slash commands and implementation contracts.
• caveman handles conversational compression with multiple intensity levels, while caveman-compress reduces file-based context windows.
• caveman-stats quantifies token savings, and caveman-review automates PR feedback in ultra-compact format.
• caveman-commit generates formatted commit messages, caveman-help provides command documentation, and cavecrew manages multi-agent workflows.
• Skills can be chained together (e.g., /caveman-compress followed by /caveman-stats) to optimize both input and output token usage.

Frequently Asked Questions

What is the main skill in the JuliusBrussee/caveman project?

The caveman skill is the primary output compression engine that converts Claude's responses into terse "caveman-style" prose. It offers intensity levels ranging from lite to wenyan-ultra and is defined in skills/caveman/SKILL.md. This skill removes filler words while preserving exact code blocks, URLs, and technical terminology.

How does the caveman-stats skill calculate token savings?

The caveman-stats skill reads Claude's session log to compare actual token consumption against a non-caveman baseline estimate. It reports the current session's token usage and calculates estimated savings by comparing compressed output against standard verbose responses. The implementation is specified in skills/caveman-stats/SKILL.md.

Can I use multiple caveman skills simultaneously?

Yes, the modular architecture allows skill composition. For example, you can run /caveman-compress CLAUDE.md to reduce input context size, then use /caveman to compress output, and finally /caveman-stats to verify savings. The skills/ directory structure ensures each skill operates independently while the plugin loader manages their execution context.

What file types does caveman-compress support?

The caveman-compress skill processes large natural-language project-memory files such as CLAUDE.md, notes, and todos. It preserves structural elements like headings, code blocks, URLs, and file paths while compressing prose content, making it ideal for repository documentation and session context files as documented in skills/caveman-compress/README.md.

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