Setup

Nagi Setup

Step 0: Language selection

Before proceeding with any other steps in this skill, ask the user which language to continue in using AskUserQuestion. Keep this initial prompt in English because the preferred language is not yet known.

• Question: Which language should I continue in?
• Options: English, 日本語 (Japanese)

Use the selected language for all subsequent user-facing messages and for every further AskUserQuestion prompt in this skill. Do not translate code, file paths, shell commands, or file contents.

Run setup steps automatically. Only pause when user action is required (pasting tokens, configuration choices).

Principle: When something is broken or missing, fix it. Don't tell the user to go fix it themselves unless it genuinely requires their manual action (e.g. pasting a secret token). If a dependency is missing, install it.

UX Note: Use AskUserQuestion for all user-facing questions.

1. Prerequisites (Node.js + pnpm)

Check Node.js and pnpm are available:

node --version   # Must be >= 22
pnpm --version   # Must be installed

• If Node.js missing or too old: install via nodenv (nodenv install 22.x.x && nodenv local 22.x.x), nvm (nvm install 22), or brew (brew install node@22)
• If using nodenv/nvm and switching versions: rm -rf node_modules && pnpm install to rebuild native bindings
• If pnpm missing: corepack enable && corepack prepare pnpm@latest --activate, or npm install -g pnpm

2. Install Dependencies & Build

pnpm install
pnpm build

If build fails, read the error and fix. Common issues:

• Missing native build tools: xcode-select --install (macOS) or sudo apt-get install build-essential (Linux)
• better-sqlite3 build failure: install build tools and retry

3. Docker

Check container runtime:

docker info

• If running: continue
• If installed but not running: open -a Docker (macOS) or sudo systemctl start docker (Linux). Wait 15s, recheck.
• If not installed: AskUserQuestion — install Docker? Then:
  • macOS: brew install --cask docker then open -a Docker
  • Linux: curl -fsSL https://get.docker.com | sh && sudo usermod -aG docker $USER

3b. Choose Agent Type

AskUserQuestion: Which agent runtime do you want to use?

• Claude Code — Anthropic's Claude Code CLI runs inside the container. Auth is proxied from the host (credential proxy). Requires CLAUDE_CODE_OAUTH_TOKEN or ANTHROPIC_API_KEY.
• Open Code — Open-source agent SDK. Supports OpenRouter, Google Gemini, OpenAI, and Anthropic as providers. Requires provider-specific API key.

Remember the choice — it affects the container image build (step 3c) and authentication (step 5).

3c. Build Container Image

Build the Docker image matching the chosen agent type:

Claude Code:

./container/claude-code/build.sh

Verify: docker images nagi-agent

Open Code:

./container/open-code/build.sh

Verify: docker images nagi-agent-opencode

Takes a few minutes on first build (cached afterwards).

4. Deploy

Hand off to the /deploy skill (invoke it via the Skill tool, do not just print "run /deploy"). Tell /deploy to select All targets, and pass through the agent type chosen in step 3b so it can set CONTAINER_IMAGE correctly.

/deploy will guide the user through:

• .env creation, with prompts for agent authentication (CLAUDE_CODE_OAUTH_TOKEN via claude setup-token, or ANTHROPIC_API_KEY, or Open Code provider keys)
• .env channel tokens (Slack / Discord / Asana — multi-select, with Bot/App creation instructions)
• Entry points (host, Claude Code container, Open Code container)
• Group prompt defaults
• Data directories (__data/{ASSISTANT_NAME}/)

When /deploy returns, .env should have at least one channel token plus an agent auth token. If neither is present, ask the user before continuing — pnpm dev in the next step will not produce a usable bot otherwise.

5. Install as a Background Service & Register Main Group

Hand off to /setup-launchd (invoke via the Skill tool, do not just print "run /setup-launchd"). It installs the launchd plist with detected paths and loads the service so the orchestrator starts in the background. macOS only — see Troubleshooting if the host is Linux.

Once the service is loaded, the channel adapters connect and log their channel JIDs. Read them so the user can pick which one to register as the main group:

tail -50 __data/{ASSISTANT_NAME}/logs/nagi-{ASSISTANT_NAME}.log | grep -E 'slack:|discord:|asana:'

Look for log lines like slack:C0AP0BRN50X or discord:1487646521259196426. If nothing shows up yet, wait a few seconds for the adapters to connect and re-tail.

Now register the main group via the /register-channel skill (invoke via the Skill tool, do not run a hand-rolled node -e script). Pre-fill the answers so the user just confirms:

• Action: register (the option labelled 登録する / Register depending on the language picked in /register-channel)
• Channel kind: whichever JID prefix you found in the logs
• Channel ID: the suffix of the JID (e.g. C0AP0BRN50X for slack:C0AP0BRN50X)
• Group name: Main
• folder: main
• trigger: @{ASSISTANT_NAME}
• isMain: true
• requiresTrigger: false

/register-channel runs a reachability check (Slack/Discord/Asana API), creates the DB row, and creates __data/{ASSISTANT_NAME}/groups/main/. The main group has elevated privileges: no trigger required, and it can register additional groups at runtime.

After /register-channel returns, run /nagi-restart so launchd reloads the service and picks up the new group.

6. Dashboard UI (Optional)

Start the web dashboard to monitor agent activity:

pnpm ui:dev    # SPA (port 5174) + API server (port 3001)

Features: Overview stats, Groups, Channels, Sessions (chat viewer), Tasks, Logs, Settings.

7. Verify

Send a message in your main channel and watch the logs via /nagi-logs (or tail directly):

tail -f __data/{ASSISTANT_NAME}/logs/nagi-{ASSISTANT_NAME}.log

Expected behavior:

1. Channel connects (Slack/Discord bot comes online)
2. Message received (log: "Discord message stored" or similar)
3. Container spawned (log: "Spawning container agent")
4. Response sent back to channel

8. Next Steps

After Step 7 succeeds, do not stop silently. Assess the current state and surface follow-up actions via AskUserQuestion (use the language picked in Step 0). Inspect the workspace before asking so you can recommend only what still applies:

• Read deploy/{ASSISTANT_NAME}/.env and check which of SLACK_BOT_TOKEN, DISCORD_BOT_TOKEN, ASANA_PAT are set.
• Check whether a launchd plist exists for this assistant (e.g. ~/Library/LaunchAgents/com.nagi.{ASSISTANT_NAME}.plist on macOS).

Then offer the relevant items below as options. Suggest only what is not already done; do not pad the question with already-completed items.

1. Add more channels — if only one channel token is configured and the user has not said they only want one. Recommend:

   • /add-channel-slack — Slack Socket Mode bot
   • /add-channel-discord — Discord Gateway bot
   • /add-channel-asana — Asana PAT + project polling

2. Agent hooks (optional) — /add-agent-hooks-claude-code or /add-agent-hooks-open-code for PostToolUse / SessionStart notifications during long agent sessions.

Note: Slack defaults to Block Kit Embed (colored left bar) and Discord defaults to Embed rich display, both straight from the template. Only mention /add-channel-slack-block-kit (drop the colored bar) or downgrading to plain text if the user explicitly asks for a less rich look.

Present the still-applicable items as AskUserQuestion options (single-select; include an explicit "Done — exit setup" option). When the user picks one, hand off by suggesting they invoke that slash command — do not run the other skill yourself inside /setup. If they pick "Done", confirm completion and exit.

Troubleshooting

"No channels connected": Check .env has correct tokens. Restart orchestrator.

"Container runtime is required but failed to start": Docker isn't running. Start it and retry.

No response to messages: Check group is registered in DB. Check trigger pattern matches. Main group doesn't need trigger prefix.

Container fails: Check __data/{ASSISTANT_NAME}/groups/main/logs/container-*.log for details. Ensure the Docker image is built (nagi-agent:latest for Claude Code, nagi-agent-opencode:latest for Open Code).

"SLACK_BOT_TOKEN not set": Tokens must be in deploy/{ASSISTANT_NAME}/.env, not in environment variables. The project-root .env is not loaded.

Linux: Currently unsupported. The maintained startup path uses macOS launchd (/setup-launchd). pnpm dev exists in the codebase and may work for self-driven Linux setups, but there is no official Linux flow.