Quickstart: create a rig

Turn an existing agent workspace into a portable, shareable rig in five commands.

You have a folder where you've been working with Claude Code. There's an MCP config, some prompts in .claude/, maybe an env file, maybe a data/ folder. You want to share it.

1. Initialise

cd my-agent-workspace
rig init

An interactive wizard inspects your folder and asks a few questions (name, author, description, version). It writes a rig.toml at the root with everything it inferred: env vars from your .env.example, MCP servers from your .mcp.json, tools from your scripts, Claude permissions from .claude/settings.json.

It also writes .claude/skills/rig/SKILL.md and .claude/CLAUDE.md so that Claude Code, running inside this workspace, knows it's a rig and can self-heal if anything's missing for a future user.

Want non-interactive? rig init --plain --author you.

2. Decide what gets shared

By default rig packages configuration only — .claude/, CLAUDE.md, rig.toml, .mcp.json, .env.example, scripts/, your skills/agents/commands folders, and templates. Your actual data, outputs, and private files stay out.

To opt files in:

rig add strategies/**
rig add data/reference/companies.json

These get added to [package].include in rig.toml. Read What gets shared before you publish — there are sensible defaults but it's worth a look.

3. Pack it

rig pack

Produces dist/<name>-<version>.rig.tgz. Inspect what went in:

rig inspect dist/my-agent-workspace-0.1.0.rig.tgz

You'll see the manifest plus the file list. Rig also surfaces warnings if something looks like a secret (private key blocks, .env-shaped strings, email-ish data in bulk).

4. Test the round trip

In another folder:

rig hub use ./dist/my-agent-workspace-0.1.0.rig.tgz --dir /tmp/test-rig
cd /tmp/test-rig
rig install
rig doctor

If it all comes up green, you're ready to publish.

5. Publish

To the Hub:

rig login                # device-flow OAuth, opens browser
rig hub publish

Or to a local file registry, or as a GitHub release — see Publish.