Start

Start

Guided onboarding for new workspaces and "what's next?" for existing ones. Goal: help every workspace build all 5 primary components.

Step 1 — Check workspace

If there's no mug.json in the current directory, the user isn't in a workspace yet:

1. Run mug start — this unpacks the agent kit to ~/.mug/agent-kit/ (CLAUDE.md, skills, docs) and prints the path. Review the materials there for full platform context.
2. Ask if they want to create a new workspace or clone an existing one
3. New → run mug init (ask for a workspace name)
4. Clone → run mug clone
5. After init/clone completes, continue to Step 2

Step 2 — Detect what's built

Scan the workspace to see which of the 5 primary components exist:

• Connectors: any .ts files in connectors/
• Workflows: any .ts files in workflows/
• AI Agents: any folders in agents/ containing agent.json
• Surfaces: any .json files in surfaces/
• Slack App: slack.json exists and has "enabled": true

Step 3 — Route based on state

New workspace (nothing built)

Print the full orientation, then recommend a starting point:

---

What Mug is

Mug is an AI automation platform for building deployed agents, real code workflows, and headless web surfaces for everyday business — integrated with any API, and accessed via email, SMS, and Slack. Everything is built locally using Claude Code, Codex, or Cursor.

What you can build

There are 5 primary components in a Mug workspace:

1. Connectors — sync data from any external API into local SQLite. OAuth, pagination, rate limits, and incremental sync are handled for you. Use /connector to get started.
2. Workflows — scheduled or triggered automation. Query data, call AI, send notifications, approve/reject, call external APIs. Use /workflow to get started.
3. AI Agents — autonomous multi-turn agents with tools, memory, cost controls, and structured output. Reason over synced business data. Use /agents to get started.
4. Surfaces — web forms, portals, and dashboards on branded subdomains. Collect data, display dashboards, handle approvals. Use /form or /portal to get started.
5. Slack App — deploy a fully bespoke Slack app with slash commands, interactive buttons, Home Tab, and AI agent DMs. Use /slack to get started.

Where to start

Most workspaces start with a connector — pull in real data first, then everything else has something to work with. What system does your client use? (QuickBooks, HubSpot, ServiceTitan, Airtable, etc.)

→ Hand off to /connector with the user's chosen API.

---

Existing workspace (has at least one component)

Print a progress checklist, then recommend the next component:

---

Workspace progress

Show each component with ✓ or ○, listing what's built:

✓ Connectors — quickbooks (3 tables)
✓ Workflows — weekly-report, invoice-reminder
○ AI Agents — none yet
○ Surfaces — none yet
○ Slack App — not enabled

Recommended next

Follow the dependency chain to pick the best next component:

1. Connector (if none) — "Start by connecting a data source. What API does your client use?" → /connector
2. Workflow (if no workflows, but has connectors) — "You have data syncing. Build a workflow to do something with it — a scheduled report, a triggered alert, an approval chain." → /workflow
3. AI Agent (if no agents, but has data) — "Build an AI agent that can reason over your synced data — classify, summarize, draft responses, make decisions." → /agents
4. Surface (if no surfaces, but has data or workflows) — "Give people a way to interact — an approval form, a status dashboard, a data portal." → /form or /portal
5. Slack App (if not enabled, but has agents or workflows) — "Connect your agents and workflows to Slack — slash commands, interactive buttons, AI chat." → /slack

If all 5 are built: "This workspace has all 5 primary components. You can add more of any type, or run mug deploy to ship it."

Present the recommendation but let the user choose any component. Hand off to the relevant skill.

---

Available skills for hand-off

• /connector — build a connector for an external API
• /workflow — create a workflow
• /agents — create an AI agent
• /form — create a form surface
• /portal — create a portal surface
• /slack — set up a Slack app
• /notify — add email/SMS notifications to a workflow
• /ai — add AI to a workflow
• /demo — share surfaces with stakeholders

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: mugwork
• Source: mugwork/mug
• License: MIT
• Homepage: https://mug.work

Install and usage instructions live in the source repository linked above.