Skip to content

The harness package manager for teams

Describe your team's AI-assistant setup once. Everyone installs, checks, and updates it with a single command — reproducibly, and with every change shown before it is written.

Every developer configures their AI coding assistant by hand — skills, permission rules, MCP servers, context files. Left to each machine, that setup diverges: no two teammates end up with the same one, and no one can say who has what.

agent-rigger turns that setup into something a team declares once and everyone applies the same way.

Your team describes its harness — the skills, sub-agents, guardrails, and context that shape the assistant — in a catalog: a plain git repository, versioned like any other code. The tool binary carries no content of its own; the opinion lives in your team’s catalog, not in agent-rigger.

One command installs the declared setup, another audits whether it is still correctly in place, a third brings it up to the catalog’s latest version. The same source produces the same result on every machine, for Claude Code and opencode alike.

Nothing is written before you see a plan of the exact changes and confirm them. Remote content is scanned before it touches disk, and every install records what it changed so it can be undone precisely.

Terminal window
# one-time: point rigger at your team's catalog
agent-rigger init
# see what's available, install it, and audit the result
agent-rigger ls
agent-rigger install
agent-rigger check
# bring every installed artifact up to the catalog's latest version
agent-rigger update

Start the 10-minute tutorial →