/setup

Guide the user through a complete OpenClaw setup in three sequential phases: install, config, and credentials. Be conversational, check in after each phase, and skip phases the user indicates are already done.

Command execution rule: Never ask the user to run commands themselves. Use the Bash tool to run all commands directly. Only ask for confirmation before running commands that modify the system (installs, config changes). Show the user the output and explain what it means.

When asking questions, always use AskUserQuestion with structured options:

• Yes/No questions: provide options ["Yes", "No"]
• Single-choice questions: provide a selection list; always include a final option like "Other / enter manually" that accepts free text input
• Multi-select questions: ask one item at a time with checkboxes, OR present a numbered list and ask the user to type the numbers they want (e.g. "1,3,4"), always including an "Other" option

---

Behavior

Parse any arguments passed by the user:

• --skip-install — skip Phase 1
• --skip-config — skip Phase 2
• --skip-credentials — skip Phase 3

If no arguments, run all three phases in order.

---

Phase 1: Installation

Unless --skip-install was passed:

1. Ask (Yes/No): "Is OpenClaw already installed on your system?"

   • Yes → skip to Phase 2.
   • No → continue.

2. Ask (selection list): "What platform are you on?" Options: ["macOS / Linux / WSL2", "Windows (PowerShell)", "Other / enter manually"]

3. Ask (Yes/No): "Ready to run the install command?" then run it via Bash:

   macOS / Linux / WSL2:

   curl -fsSL https://openclaw.ai/install.sh | bash
   

   Windows (PowerShell):

   powershell -Command "iwr -useb https://openclaw.ai/install.ps1 | iex"
   

   Show the output. If it fails, ask (Yes/No): "Want to try the npm fallback instead?" and run:

   npm install -g openclaw@latest && openclaw onboard --install-daemon
   

4. After install, run verification automatically via Bash:

   openclaw doctor
   openclaw status
   

   Show the user the output and flag any errors before proceeding.

---

Phase 2: Configuration

Unless --skip-config was passed:

1. Ask (Yes/No): "Do you want to use the default gateway settings (port 18789, localhost only)?"

   • Yes → confirm the defaults are secure and move on.
   • No → ask what they want to change using a selection list: ["Port number", "Tailscale exposure", "Auth mode", "Other / enter manually"]

2. Walk through DM policy with a selection:

   • Briefly explain: pairing (safe default — only paired devices can DM) vs open (any user can message).
   • Ask (selection list): "Which DM policy do you want?" Options: ["pairing (recommended)", "open", "Tell me more before deciding"]
   • If "Tell me more", explain further then ask again.

3. Ask (multi-select): "Which messaging channels do you want to connect?" Present as a numbered list:

   1. WhatsApp
   2. Telegram
   3. Slack
   4. Discord
   5. Other
   

   Ask: "Enter the numbers of the channels you want (e.g. 1,3), or type channel names if choosing Other." Note which ones they selected — credentials for these will be covered in Phase 3.

4. If any config changes were made, run automatically via Bash:

   openclaw doctor
   

   Show the output and flag any issues.

---

Phase 3: Credentials

Unless --skip-credentials was passed:

1. Start with the AI model API key:

   • Ask (selection list): "Which AI provider are you using?" Options: ["Anthropic (Claude)", "OpenAI (GPT)", "Other / enter manually"]
   • Provide the appropriate env var:
     • Anthropic: export ANTHROPIC_API_KEY=sk-ant-...
     • OpenAI: export OPENAI_API_KEY=sk-...
     • Other: ask the user to provide the env var name and value.
   • Ask (Yes/No): "Ready to run openclaw onboard to store the key persistently?" then run it via Bash:

     openclaw onboard
     

   • Guide the user through the interactive prompts.

2. For each messaging channel the user selected in Phase 2, walk through credentials one at a time. Where credentials require external steps (browser, phone), explain what the user needs to do, then ask (Yes/No): "Done — ready to continue?" before running the next command.

   WhatsApp: Ask (Yes/No): "Ready to pair WhatsApp?" then run openclaw onboard, guide the user to select WhatsApp and scan the QR code with their phone.

   Telegram:

   • Instruct the user to message @BotFather on Telegram → /newbot → copy the token.
   • Ask (Yes/No): "Got the token? Ready to add it?" then run openclaw onboard via Bash.

   Slack:

   • Instruct the user to create an app at https://api.slack.com/apps, enable Socket Mode, add scopes (chat:write, im:history, im:read, app_mentions:read), install to workspace, and copy both tokens (xapp-... and xoxb-...).
   • Ask (Yes/No): "Got both tokens? Ready to add them?" then run openclaw onboard via Bash.

   Discord:

   • Instruct the user to create a bot at https://discord.com/developers/applications, enable Message Content Intent and Server Members Intent, copy the bot token, and invite the bot to their server.
   • Ask (Yes/No): "Got the token? Ready to add it?" then run openclaw onboard via Bash.

   For other channels: ask (Yes/No): "Ready?" then run openclaw onboard via Bash and guide the user through channel selection.

3. After all credentials are set, run a final check automatically via Bash:

   openclaw doctor
   openclaw status
   

---

Completion

Once all phases are done:

• Summarize what was set up (platform, channels connected, model provider).
• Remind the user they can reconfigure at any time with openclaw onboard.
• Ask (selection list): "What would you like to do next?" Options: ["Send a test message on a connected channel", "View openclaw status", "I'm done — nothing else", "Other / enter manually"]
  • "View openclaw status" → run openclaw status via Bash and show output.