Rig

Portable, runnable agent workspaces.

A rig is a folder that bundles everything an AI coding agent needs — system prompts, MCP servers, tool configs, data, and setup instructions — into a single portable package. Install one and start working in seconds.

npm install -g @rigxyz/cli

rig hub search market-researcher
rig hub use hub:maria/market-researcher
cd market-researcher && claude

Commands

# Rig — the package
init               Initialize a new rig (--no-sync for local-only)
add <pattern>      Add files/patterns to package.include
pack               Package into a .rig.tgz artifact
install            Run setup hooks for the current rig
inspect <source>   Show artifact metadata
doctor             Check workspace readiness

# Workspace — live & shared
share [<email>]    Invite a person (or --ops … for scoped/agent access)
who                Members, invites, scoped access
unshare <target>   Revoke a grant
join <url>         Join someone else's workspace
status [--verbose] Live? syncing? who's on it
disconnect         Stop being a synced workspace (local)

# Hub — distribution
hub publish        Publish this rig to the Hub
hub use <source>   Install a rig
hub update [src]   Update an installed rig
hub search <query> Search Rig Hub
hub list           List a registry

# Account
login / logout     Authenticate with Rig Hub

Sources

Rigs can be installed from multiple sources:

hub:author/name[@ver]       Rig Hub (userig.xyz)
github:user/repo[@ver]      GitHub release
./path.rig.tgz              Local artifact
registry:author/name        Local file registry

How it works

# 1. Start from any agent folder
cd my-agent-workspace

# 2. Initialize — detects MCP servers, env vars, tools, plugins
rig init

# 3. Optionally add files/folders to share
rig add strategies/**
rig add data/reference/companies.json

# 4. Package into a portable artifact
rig pack

# 5. Publish to the Hub
rig login
rig hub publish

# 6. Anyone can install and run it
rig hub use hub:you/my-agent-workspace
rig install
rig doctor

rig init scans your folder and infers setup metadata: MCP server names, required tools (uvx, npx, uv), env vars from config placeholders, plugin hints, permissions (allow/deny rules from .claude/settings.json), install commands, and workspace directories. By default, only configuration files (.claude/**, CLAUDE.md, .mcp.json, etc.) are included — folder contents are not shared unless you opt in via interactive init or rig add.

Permissions declared in [permissions] are written to .claude/settings.json during install, so users don't face permission dialogs they don't understand. Rig merges its rules with existing user permissions (union merge, deny wins).

rig doctor provides a live machine-readable readiness check. Agents like Claude Code use it to self-heal broken workspaces.

Canvas

View a rig's outputs as an interactive UI — briefings, tables, maps, and documents rendered as cards on an infinite canvas. From any rig folder:

npx @rigxyz/canvas

The board is just a file in the rig (.canvas/<name>.board.toml): each card points at a rig file and picks a render primitive, so content stays files-first and the canvas updates live as the rig changes. Installed rigs ship a rig-canvas skill that teaches the agent how to author a board and launch the viewer — ask Claude to "build a UI for this rig."

Hub

Rig Hub is the community registry. Browse, search, and install agent workspaces published by others.

Links

• Rig Hub
• Manifesto
• CLI Docs