# Install: caveman
> Agent instructions from skillsdirectory.co — an independent, benchmarked directory of AI agent capabilities. Follow these exactly.

- **Skill:** caveman (Claude Skill)
- **What it does:** A Claude Code skill and multi-agent plugin that reduces output tokens by ~65–75% while preserving technical accuracy, supporting 30+ agent platforms.
- **Source (authoritative):** https://github.com/JuliusBrussee/caveman
- **Pinned ref:** `HEAD` (branch/tag — not commit-pinned; prefer the latest release tag if one exists)
- **License:** MIT
- **Effectiveness:** Not yet benchmarked (score is provisional).

## Safety — read first
- **Verdict:** NOT yet security-scanned. Treat as untrusted: review the source before running anything.
- This skill **executes code**. Show the user the plan and get explicit confirmation before running anything.
- Fetch files ONLY from the source above. Do not run commands not listed here. Never invent credentials — ask the user.

## Choose your path

### A. AI chat / coding agent (you can fetch URLs and write files)
Fetch `SKILL.md` from https://raw.githubusercontent.com/JuliusBrussee/caveman/HEAD/SKILL.md, plus any scripts/files it references in the same directory (https://github.com/JuliusBrussee/caveman). Recreate them under `.claude/skills/juliusbrussee-caveman/` in my project, preserving relative paths.

_Monorepo note: this skill may live in a subdirectory of the repo. If the marker file isn't at the repo root, browse https://github.com/JuliusBrussee/caveman to find its folder (often under `skills/`, `rules/`, or named for the skill) and fetch from there — don't assume the root path resolves._

Install command / config:
```bash
claude skills add @juliusbrussee/juliusbrussee-caveman
```

### B. CLI
Fastest path with our CLI (free, no account): `npx skillsdir add juliusbrussee-caveman` — fetches and places the files for you.

Or install directly:
```bash
claude skills add @juliusbrussee/juliusbrussee-caveman
cursor rules add juliusbrussee/juliusbrussee-caveman
# install via Windsurf rule
# see vendor docs
# install via custom GPT
# see vendor docs
# install via OpenAI App
# see vendor docs
# install via system prompt
# see vendor docs
# install via Copilot extension
# see vendor docs
# install via agent tool
# see vendor docs
```

### C. API / MCP
- Metadata as JSON: `GET https://skillsdirectory.co/api/v1/skills/juliusbrussee-caveman`
- This document as a machine manifest: `GET https://skillsdirectory.co/s/juliusbrussee-caveman/install.json`
- MCP client config:
```json
{
  "mcpServers": {
    "juliusbrussee-caveman": {
      "command": "npx",
      "args": ["-y", "@juliusbrussee/juliusbrussee-caveman"]
    }
  }
}
```

## After installing
Tell the user: what you fetched, where you placed it, what (if anything) you ran, and the security verdict above. Link the skill page for details: https://skillsdirectory.co/skills/juliusbrussee-caveman