CLI reference

Documented against OpenRig 0.7.7. Help text uses "node" where these pages say seat, for a seat's position in the running rig.

What it is for

rig is the one command both people and agents use to operate OpenRig. It is readable by default and most commands take --json for agents. Every command that changes something ends by telling you what happened, the current state and a sensible next step. Every error says what happened, why it is a problem and what to do next. The surface is large and grows from release to release, so the full reference below is built from the CLI's own command tree rather than written by hand, and it names the source commit it was checked against.

How to read it

The full reference is the OpenRig repository's as-built CLI reference, as shipped with the current release and rendered on this site: the full command table. It lists every registered command and option, with the source that implements each one.

Two habits make it useful. rig --help lists the installed top-level commands; read that list before building anything, because the most expensive failure is a script that re-implements a verb. rig <verb> --help is always more current than any page, including this one.

The verbs by job

• Setting up: setup prepares the machine; config, mode and policy set settings, operating posture and the recorded permission policy. See Configuration.
• Seeing it all: tui opens the terminal UI; ui opens the older web UI, in maintenance mode. See The terminal UI.
• Finding out what is true right now: whoami, ps, status, roster, queue whoami, seat status, view, config, health, usage, telemetry.
• Reaching another seat: send, capture, broadcast, transcript, ask, chatroom, walk, terminal. See Messaging.
• Work that outlives a turn: queue, stream, view, project, and acp to hand a row's task to an Agent Client Protocol agent. See Coordination.
• Being woken: watchdog. Running a pipeline: workflow. See Workflows.
• Bringing rigs and seats into and out of existence: up, down, launch, add, remove, expand, shrink, grow, create, fork, archive, unarchive, attach, bind, adopt, discover, release, unclaim, reconcile-session, edge, plus bootstrap and requirements for a spec. See Lifecycle.
• Surviving: snapshot, restore, restore-check, restore-packet, handover, seat, compact-plan, compact, agent-image, startup-proof. See Continuity.
• Work on disk: scope, proof, workspace. See the work tree.
• Knowledge for seats: context, skill, plugin. See Context.
• Other machines and people: host, file, gateway, slack. Accounts and services: auth, provider, env. Driving OpenRig from an MCP agent: mcp. Harnesses beyond Claude Code and Codex, installed as presets OpenRig doesn't test: community-harness.
• When something is broken: doctor, preflight, daemon, start, crash-cart, destroy, heartbeat, parked.

Help uses "node" where these pages say seat, for a seat's position in the running rig; the generated reference quotes help verbatim.

What it does not do

• The reference does not describe effects that were never executed. It lists what is registered and where it is implemented; rig <verb> --help says how to use each one.
• Exit codes and daemon requirements are per command; the general rule is that most commands need a running daemon, and daemon, config, preflight and doctor work without one.

Where it goes next

Each family page above. The Specs pages for spec, specs, agent, bundle, export and import; package is the legacy form.