Supported AI Agents for the Caveman Installer: Complete Integration Guide

Caveman supports over 30 AI coding agents including Claude Code, Gemini, Cursor, Codex, Cline, Copilot, and OpenClaw, automatically installing its token-compression skill into each agent's plugin directory via platform-specific scripts.

The Caveman installer, hosted at JuliusBrussee/caveman, is a cross-platform skill deployment system that injects token-compression capabilities into popular AI coding assistants. Whether you use Claude Code, Gemini CLI, or Cursor, the installer detects your agents and configures the Caveman skill to shrink textual output while preserving code and error messages byte-for-byte.

Supported AI Agents and Installation Methods

Caveman functions as a skill/plugin that attaches to AI coding agents to reduce token usage. The installer adapts its deployment method based on each agent's specific architecture.

Claude Code

Install via the Claude plugin marketplace. According to the source code in INSTALL.md, this activates automatically without requiring the /caveman command.

claude plugin marketplace add JuliusBrussee/caveman && claude plugin install caveman@caveman

Gemini CLI

Functions as a Gemini extension. The installer copies the caveman.skill file to the appropriate Gemini extensions directory.

gemini extensions install https://github.com/JuliusBrussee/caveman

Cursor, Windsurf, Cline, and Codex

These agents use the generic skills registry accessed via npx. The -a flag in bin/install.js specifies the target agent.

npx skills add JuliusBrussee/caveman -a cursor

OpenClaw

Requires a different approach. The bin/lib/openclaw.js utility injects a marker block into the agent's workspace (SOUL.md), enabling terse output from the first turn.

curl -fsSL https://raw.githubusercontent.com/JuliusBrussee/caveman/main/install.sh | bash -s -- --only openclaw

Other Supported Agents

The install.sh and install.ps1 scripts detect over 30 additional agents including Copilot, Windsurf, and various VS Code extensions. Each receives the dist/caveman.skill file in its respective plugin directory, with per-agent configuration read from bin/lib/settings.js.

How the Installer Works

The installation process relies on three core components documented in the repository:

  1. Shell wrappers – install.sh (Unix) and install.ps1 (Windows) download the repository and delegate to the JavaScript installer.
  2. Node.js core – bin/install.js parses CLI flags (--only, -a, --uninstall) and resolves each agent's installation path.
  3. Configuration layer – bin/lib/settings.js reads per-agent defaults (compression level, environment overrides) before copying files.

For OpenClaw specifically, bin/lib/openclaw.js appends a marker block to SOUL.md rather than installing a traditional plugin.

Using Caveman Commands Across Agents

Once installed, agents recognize the /caveman command family defined in commands/caveman.md and commands/*.toml.

Toggle compression levels during any session:

/caveman lite
/caveman ultra
/caveman wenyan

View lifetime token savings:

/caveman-stats

Permanently compress memory files (e.g., CLAUDE.md) to reduce input tokens by approximately 46%:

/caveman-compress CLAUDE.md

Sub-Agents and Extensions

Beyond the main skill, Caveman includes sub-agents stored in agents/cavecrew-reviewer.md and related files. These "cavecrew" agents (reviewer, investigator, builder) inherit the same compression logic, allowing entire workflows to remain concise.

For Codex framework integration, .codex/hooks.json registers Caveman as a discoverable skill via npx skills add.

Summary

  • Caveman supports 30+ AI agents including Claude Code, Gemini, Cursor, Codex, Cline, and OpenClaw.
  • The installer uses install.sh or install.ps1 to detect agents and deploy dist/caveman.skill to the appropriate plugin directories.
  • OpenClaw receives special treatment via bin/lib/openclaw.js, which injects a marker into SOUL.md.
  • Compression levels (lite, full, ultra, wenyan) are managed by bin/lib/settings.js and toggled via /caveman commands.
  • Sub-agents in agents/ extend the compression benefits to review and build workflows.

Frequently Asked Questions

How many AI agents does Caveman support?

Caveman explicitly supports over 30 AI coding agents, with the install.sh and install.ps1 scripts capable of detecting and configuring the skill for Claude Code, Gemini, Cursor, Windsurf, Cline, Codex, Copilot, OpenClaw, and approximately 20 additional platforms listed in INSTALL.md.

Is the installation reversible?

Yes. The installer supports an --uninstall flag processed by bin/install.js that removes the caveman.skill file from each detected agent's plugin directory and cleans up any marker blocks added to workspace files like SOUL.md.

How does Caveman integrate with OpenClaw differently?

Unlike standard agents that load the skill as a plugin, OpenClaw integration uses bin/lib/openclaw.js to append a marker block to SOUL.md in the workspace. This marker instructs the agent to use terse output from the first conversation turn, eliminating the need for explicit /caveman activation.

Can I use Caveman with custom AI agents?

While Caveman targets specific supported agents, the caveman.skill file in dist/caveman.skill is a plaintext prompt that can theoretically be loaded into any agent supporting custom instructions. The npx skills add command with the -a flag also allows targeting specific agent configurations if their plugin directories follow standard patterns.

Have a question about this repo?

These articles cover the highlights, but your codebase questions are specific. Give your agent direct access to the source. Share this with your agent to get started:

Share the following with your agent to get started:
curl -s "https://instagit.com/install.md"

Works with
Claude Codex Cursor VS Code OpenClaw Any MCP Client

Maintain an open-source project? Get it listed too →