setup

Guide the user through tlive setup. Execute in order, showing each result:

1. Run tlive status. If the command is missing → tell the user to install the engine first: npm i -g tlive, then re-run this command.

2. Read the output: an idle daemon is fine (sessions auto-start it); what matters is whether channels says (none).

3. Already configured (channels present)? Don't just say "all set" — run the verification tour instead, one check at a time with its result: a. plugins: lines show the bundled version with a ✓ for every installed vendor (a mismatch → tlive setup --hooks-only). b. Have the user send the bot a message and confirm the reply arrives (proves the inbound path, not just the config). c. Ask the user to check the bot's command menu shows /mute /trust /safe /mode /help (a stale client cache → close and reopen the chat). d. Desktop notifications: explain one fires per thing that needs you to act — a pending approval, or the idle "waiting for your input" nudge — and tlive never touches it again afterward; it just ages into the notification centre like any other app's, so that's a log, not a live status board (open the dashboard for what's waiting right now). A finished turn notifies the desktop too, after the same grace as the IM card; tool failures stay on IM, since they block nobody. There is no separate on/off switch: use the OS's Do Not Disturb to silence them temporarily, or tlive mode off to stop tlive entirely. e. Hand out the dashboard via tlive url; suggest opening it once. f. Offer optional next steps: add another channel (continue below), or stop here. Then skip to step 8 unless the user wants a new channel.

4. If no channel (or adding one): ask whether the user wants Telegram, Feishu, or both, and collect credentials per platform:

   • Telegram: bot token (create via @BotFather) + chat id (send the bot a message and read getUpdates, or the user already knows it)
   • Feishu: appId + appSecret (self-built app on the open platform with im message permissions)

5. Read ~/.tlive/config.json (may be absent or partial) and MERGE the new fields in (preserve existing ones):

   { "allowedSenders": [],
     "adapters": {
       "telegram": { "token": "<token>", "chatIdAllowList": ["<chatId>"] },
       "feishu": { "appId": "<appId>", "appSecret": "<secret>" } } }
   

   (Only write the platforms the user chose; validate the JSON.)

6. tlive stop (if running) then tlive start, then tlive status to confirm the configured platforms appear under channels.

7. Have the user send the bot a test message; hand out the dashboard address via tlive url.

8. Offer remote approval. tlive defaults to notify (watch + notify only — it never holds a tool call, so a fresh install can't hang a workflow). Ask whether the user wants to Allow/Deny tool calls from their phone; if yes, run tlive mode full (holds each tool call for a remote answer, in parallel with the local prompt — first answer wins; revert any time with tlive mode notify). If they only want monitoring, leave it in notify. If they also want sub-agent approvals on their phone (e.g. they're about to step away), that's tlive mode all — flag the trade: a held sub-agent has no terminal dialog until the window ends, so it only pays off when nobody is at the keyboard (tlive mode full goes back).

9. If status shows the Codex companion as off or degraded, explain what it means (codex missing from PATH / nothing answering on the app-server socket — see ~/.tlive/codex-appserver.log); Codex approvals stay local-only until it is running. There is no trust step to perform, and no restart either: tlive keeps checking, so installing codex clears it on its own.