Slack Notifications Setup (Sentinel)

Sentinel pings the user on Slack when a colony stalls and lets them reply from Slack to keep it going. It needs a Slack app with two tokens:

• a Bot User OAuth Token (xoxb-…) — stored under credential id slack, used to send alerts and call the Slack API.
• an App-Level Token (xapp-…) with connections:write — stored under credential id slack_app, used by Socket Mode to receive the user's reply.

You do the whole thing end to end: reuse-or-create the app, read the tokens, store them with the sentinel_setup tool (the same API the desktop connector uses), create/select the channel via the Slack API, and turn Sentinel on — without making the user paste anything.

Activate hive.browser-automation first — this skill assumes you know the lifecycle rules (the bridge attaches to the user's own Chrome; never launch or kill a browser), the screenshot + viewport-fraction coordinate workflow, and hive-browser interact.

Before you start — check what already exists

1. Run sentinel_setup({"action": "status"}). If it shows both slack bot and slack app already configured, the app and tokens are already in place — skip Steps 1–4 entirely and go to Step 5 (channel). Don't recreate anything.
2. Confirm the user is signed into the right Slack workspace: run hive-browser navigate https://api.slack.com/apps --json in the terminal. If it shows a sign-in screen, ask them to log into the workspace they want alerts in. If they have several workspaces, ask which one.
3. Reuse before create. Scan the apps list for an app named Hive Sentinel. If it exists, open it and skip to Step 2 (app-level token) / Step 3 (install) — do not create a second app. Only create one if none exists.

Step 1 — Create the app from a JSON manifest

The manifest sets the name, bot scopes (including channel management), Socket Mode, and event subscriptions in one shot.

1. Click Create New App → From a manifest, pick the workspace, click Next.
2. Switch to the JSON tab (not YAML — the browser type tools strip newlines and break YAML).
3. Inject the manifest with hive-browser evaluate (run in the terminal) using CodeMirror's API — see "Filling the manifest editor" below for the exact, non-doubling method. Use this manifest verbatim:

{
  "display_information": { "name": "Hive Sentinel" },
  "features": { "bot_user": { "display_name": "Hive Sentinel", "always_online": true } },
  "oauth_config": {
    "scopes": {
      "bot": [
        "chat:write",
        "chat:write.public",
        "channels:read",
        "channels:history",
        "channels:manage",
        "groups:read",
        "groups:history",
        "users:read"
      ]
    }
  },
  "settings": {
    "event_subscriptions": { "bot_events": ["message.channels", "message.groups"] },
    "socket_mode_enabled": true,
    "org_deploy_enabled": false,
    "token_rotation_enabled": false
  }
}

channels:manage is included up front so you can create the channel by API in Step 5 without ever reinstalling. Valid manifest keys only: socket_mode_enabled (not socket_mode), and there is no org_domains field.

4. With the editor showing a single copy and no "can't translate" error, click Next, then Create.

Filling the manifest editor (JSON tab — do this, nothing else)

Slack's manifest editor is CodeMirror 5, and it validates from React state, not the DOM. Two traps, both avoided by the method below:

• hive-browser interact/insert_text strips newlines (breaks the manifest) and the type path tries to parse the braces. Don't type into it.
• A paste/InputEvent that carries the text gets appended on top of the current value — paste it after setValue and you get the manifest twice.

Set the value once, then fire a value-less input event so React re-reads it. Run this with hive-browser evaluate in the terminal — because the script is multi-line and quote-heavy, pass it via --js - (stdin) or --js @file rather than inline --js '…':

// JSON tab must be active first (click the tab labelled "JSON").
const cm = document.querySelector('.CodeMirror').CodeMirror;
// Embed the manifest as a single-quoted JS string (it contains only double quotes).
const manifest = '{"display_information":{"name":"Hive Sentinel"}, ... }';
cm.setValue(manifest);                 // setValue REPLACES — it never appends
cm.getInputField().dispatchEvent(      // notify React WITHOUT carrying any text
  new InputEvent('input', { bubbles: true, inputType: 'insertFromPaste' })
);
return { valueLen: cm.getValue().length };  // sanity-check: one copy, not double

If valueLen is ~2× the manifest length, it doubled — cm.setValue('') then cm.setValue(manifest) once and re-dispatch. Never solve a "Next is disabled" by pasting again.

Step 2 — Generate the App-Level Token (xapp-…)

Manifests can't mint app-level tokens. In the app: Basic Information → App-Level Tokens → Generate Token and Scopes, name it (e.g. socket), add the scope connections:write, Generate, then read the xapp-… value.

Step 3 — Install to the workspace, get the Bot Token (xoxb-…)

1. Open OAuth & Permissions and click Install to Workspace → Allow.
2. Read the Bot User OAuth Token (xoxb-…). Read it from the input's value, not off a screenshot, so you don't truncate it:

Array.from(document.querySelectorAll('input')).map(i => i.value).filter(v => v.startsWith('xoxb-'));

Step 4 — Store both tokens with sentinel_setup

sentinel_setup({"action": "store_token", "provider": "slack",     "token": "xoxb-…"})
sentinel_setup({"action": "store_token", "provider": "slack_app", "token": "xapp-…"})

The bot token is validated (auth.test) before storing and reports the workspace. If it's rejected, re-read the token from the input value (you likely grabbed a partial string) and retry — never invent or pad a token.

Prefer not to handle the tokens yourself? Fall back to credentials(action="collect", …) with credential_id slack / slack_app (field access_token) to pop a secure form. The default flow above is hands-off and is what the in-app "Set this up with the agent" button expects.

Step 5 — Channel: do it with the Slack API, not the UI

Don't click around the Slack web UI to make or find a channel — use the Slack API with the bot token via terminal_exec + curl. (A browser fetch() to slack.com is CORS-blocked from the app tab; curl from the terminal is the reliable path.)

Reuse an existing channel the user named — list and grab its id:

curl -s -X POST 'https://slack.com/api/conversations.list' \
  -H 'Authorization: Bearer xoxb-…' \
  -H 'Content-Type: application/x-www-form-urlencoded' \
  -d 'types=public_channel,private_channel&limit=1000'

Create a dedicated channel (recommended default, e.g. hive-sentinel). The bot joins automatically as the creator, and the response carries the new channel id:

curl -s -X POST 'https://slack.com/api/conversations.create' \
  -H 'Authorization: Bearer xoxb-…' \
  -H 'Content-Type: application/json' \
  -d '{"name":"hive-sentinel","is_private":false}'

If you reuse a channel the bot didn't create, add the bot so it can post and read replies (look up the bot user id from auth.test if needed):

curl -s -X POST 'https://slack.com/api/conversations.invite' \
  -H 'Authorization: Bearer xoxb-…' -H 'Content-Type: application/json' \
  -d '{"channel":"C0123ABC456","users":"<BOT_USER_ID>"}'

Take the channel id (C…) from the JSON response. channels:manage is already in the manifest, so none of this needs a reinstall.

Step 6 — Turn Sentinel on and verify

configure/test target the colony bound to your session (pass colony_id only to set up a different one):

sentinel_setup({"action": "configure", "channel": "slack",
                "target": {"channel": "C0123ABC456"}, "enabled": true})
sentinel_setup({"action": "test"})

test posts to the channel — ask the user to confirm it arrived. If it didn't, the usual causes are the bot isn't in the channel (re-run conversations.invite) or the channel id is wrong. sentinel_setup({"action": "status"}) shows stored tokens + the colony's current config any time.

When test succeeds you're done: tell the user Sentinel is live on #<channel> and they can reply to its messages from Slack to keep the colony moving. (They can review or change any of this under Automations → Set up Sentinel.)

Guardrails

• Activate hive.browser-automation and obey its lifecycle rules. Never run google-chrome/chromium, --remote-debugging-port, or kill any browser/bridge process — you drive the user's existing Chrome.
• Reuse, don't duplicate. If a Hive Sentinel app already exists (or sentinel_setup status shows both tokens configured), use it — never create a second app.
• Never invent, guess, or paraphrase a token. If you can't read one cleanly, re-read the input value rather than fabricating.
• The two store_token providers are exact: slack (bot, xoxb-) and slack_app (app-level, xapp-). Swapping them breaks send vs. receive.
• Manifest editor: JSON tab + cm.setValue once + a value-less input event. Never type the manifest, never use the YAML tab, never re-paste to "fix" a disabled Next button (that doubles the content).
• If Slack shows an org/admin approval wall on install, the workspace requires admin approval for apps — tell the user to request it (or use a workspace where they're admin); you can't bypass it.