Menu
Get started with Mug — guided orientation for new workspaces, progress checklist for existing ones. Recommends what to build next.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Build a custom AI agent — autonomous multi-step work with tools, brain memory, skills, and structured output. Scaffolds the agent folder with agent.json, SOUL.md, and skills.
Create a portal surface — tabbed, section-based pages that query workspace data. Sections include tables (with detail pages + actions), stats cards, progress bars, charts, galleries, text blocks, and accordions. Used for dashboards, employee portals, approval inboxes, status trackers.
Create a new workflow — multi-step automation with data queries, AI classification, and notifications. Scaffolds the file and tests locally.
Build a connector for an external API. Walks through the full pipeline — research, discover, gather, verify, scaffold — and wires it into the workspace. Supports read-only sources and bidirectional connectors with write actions.
Run Mug CLI developer workflow commands — dev server, update, deploy, clone, shutdown, workspaces.
Build Slack integrations — configure slack.json, send Block Kit messages, slash commands, interactive buttons, Home Tab, shortcuts, and two-way workflows.
| name | start |
| description | Get started with Mug — guided orientation for new workspaces, progress checklist for existing ones. Recommends what to build next. |
Guided onboarding for new workspaces and "what's next?" for existing ones. Goal: help every workspace build all 5 primary components.
If there's no mug.json in the current directory, the user isn't in a workspace yet:
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.mug init <name> (ask for a workspace name)mug cloneScan the workspace to see which of the 5 primary components exist:
.ts files in connectors/.ts files in workflows/agents/ containing agent.json.json files in surfaces/slack.json exists and has "enabled": truePrint 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:
/connector to get started./workflow to get started./agents to get started./form or /portal to get started./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.
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:
/connector/workflow/agents/form or /portal/slackIf 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.
/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