Agent integration

Rig is designed for agents as well as humans. When Claude Code is running in a rig workspace, it can read the rig's state, understand what's missing, and repair gaps without you having to babysit it.

This section explains exactly what's installed for the agent, what the agent is allowed to do, and how to lean on it.

What rig installs for the agent

Combined, an agent dropped into either an empty folder or an existing rig workspace has enough context to do the right thing.

What the agent can do unsupervised

Read state, propose actions, run safe commands. Specifically:

• Run rig doctor and parse --json.
• Run rig status, rig status, rig hub list, rig hub search.
• Read rig.toml, .rig/setup.json, .rig/instance.toml.
• Create declared directories (mkdir -p data/portfolios).
• Append env values to .env after asking the user for them.

What the agent must ask before doing

• Running [install].commands shell hooks.
• Anything destructive: deletes, --purge, rig disconnect.
• Publishing (rig hub publish).
• Inviting collaborators (rig share).
• Joining a binding (rig join).

What the agent should never touch

• .rig/tap-binding.local.json — device secrets.
• ~/.config/rig/config.json — auth tokens.
• Other people's .env values.

Pages

• The rig skill — what's in user-level vs project-level skills.
• Doctor as live state — the JSON contract.
• Repair workflow — the canonical loop.
• .well-known/rig.json — agent-discoverable site descriptor.