Grok BotGuides

Needs re-check

· Power user · Safety and approvals · Part 7 of Run a fleet of Bots

Set approvals and permissions for a fleet of Grok Bots

Not tested yet: Local execution label: Settings > Computer > Execution on this computer, or Settings > General > Bot > Execution on Local Computer. Rules label: docs say Settings > General > Auto-review; Oct 5 changelog says Custom Rules.

Give each Bot its own stop points, keep the same hard lines across all of them, and know which Bots can't stop to ask.

With one Bot, you see its approval cards because you're in the chat. With a fleet, most work happens while you're elsewhere: routines, handoffs between Bots, Team Bots answering teammates. This guide covers how to set stop points across many Bots so the right actions wait for you and the rest keep moving. It's part 7 of our series on running a fleet of Bots.

How approval cards, Auto-review, and personal rules work for a single Bot is covered in our guide to approvals and Auto-review. We don't repeat it here.

We haven't set up approvals across a fleet for this guide. The Daily Bot works on a Bot computer without the desktop app. The steps come from Cursor's docs and help center, the changelog, and SpaceXAI staff guides, each linked.

Why it matters#

Background approvals don't wait long. "Approvals from work that started without you, like a routine or a message from another Bot, expire after about 10 minutes," Cursor's FAQ says. "Approvals in a chat you are in wait for you."

Some Bots can't ask at all. A Team Bot "asks for approval only in its owner's own chat with it," per the help center. Elsewhere, "nobody who can answer an approval card is there. So the Bot works within the permissions its owner set up, without stopping to ask."

How to use it#

1. Give each Bot its own stop point#

An approval boundary is one of the five reasons Cursor's docs give for making a separate Bot. Write each Bot's limit into its description, next to its job.

The published setups do this per Bot. Josh Kim's marketing analyst Bot, in his marketing guide, "does not change a live ad, bid, or budget." Ryan Perry's analytics Bot, in his mobile app development guide, says "Check with me before anything that spends money."

Start narrow and widen. Cursor's use-case docs say to start each role with read-and-prepare work, "review the result, and only then add approved actions or a routine."

2. Keep the same hard lines everywhere#

Some actions should stop for you no matter which Bot takes them. Cursor's docs list them for routines: "keep sending, purchasing, deleting, publishing, and production changes behind approval."

SpaceXAI staff draw similar lines. Kevin Niparko, a PM, writes in his Grok Bot for PMs guide: "I still keep final review for any external email sends, purchasing anything, or destructive actions like deletes." Stefan Markarian's first team rule in his SDR guide is "Never auto-send. The bots can research, draft, create records, and stage work. I review and send."

A prompt written by us, for any Bot's description:

Prompt
Before you do anything that sends a message, spends money, deletes something, publishes, or changes production, stop and show me exactly what you'll do and to what. Everything else in your job you can do without asking.

3. Make background approvals reach you#

If a Bot works in the background, its approvals need a way to find you within 10 minutes. The FAQ suggests two fixes: "Turn on the Bot's Notifications, or choose Always allow for actions you trust."

  • Notifications. Turn them on per Bot in Bot settings, per Cursor's docs. Since Oct 5, notifications say the Bot "is waiting for you."
  • Another device. Since Aug 28, "you can approve a task on one computer from another computer where you're signed in."
  • Slack. On Cursor teams, since Sep 29, a Bot that needs your approval in a Slack thread "posts a Review an action card only you can answer, instead of dropping the action."

4. Audit what Always allow saved#

Always allow is fast, and it lasts. It "can save a matching rule," per Cursor's security docs. Since Oct 2, approval cards say when Always allow added a rule, "with Edit rule to change it in Settings." Since Oct 5, Settings calls saved rules Custom Rules.

Look through those rules from time to time. The security docs say to keep rules "narrow and tied to a known action," and that "Ask first wins when rules conflict."

5. Set Team Bot limits for when nobody's watching#

Since a Team Bot won't stop to ask in teammates' chats, Slack, or group chats (see above), its owner sets its limits up front. Write them into its description, and only give it plugins and secrets you'd let it use unsupervised. "Only bring keys you're happy for the whole team to use," the help center says.

Others may be able to change it. Since Sep 30, a Team Bot's creator "can add teammates under Managers, who can then change its profile, skills, plugins, secrets, and Slack app."

6. Decide about your own computer once#

Bots can run commands on your desktop, separately from their cloud computer. The policy lives under Settings > Computer > Execution on this computer, per the security docs: "ask every time, always allow, or never." The docs "recommend Never unless a Bot has a specific reason to work on local files." The setting is per desktop, not per Bot. Until a desktop is listed under Computer, the docs put it at Settings > General > Bot > Execution on Local Computer. Needs re-check: we haven't seen either label in the app.

7. On a team, keep admin rules few#

Cursor team admins can add Team Rules that "every member's Bots follow," per the teams docs. The docs add: "Rules applied to Grok Bot are always required, so members can't turn them off. Keep them short and few, like 'never move company data to personal accounts.'" On Enterprise, admins can also enforce Auto-review with team-wide rules, and "Ask first" still wins on conflict.

Good to know#

  • Approvals cover what's next, not what's done. "An approval controls the proposed action, not work already completed, and nobody should approve an action whose target or effect they can't identify," the security docs say.
  • Auto-review "doesn't review every side effect. Memory writes and most settings changes are examples," per the same docs. That's one more reason to put limits in descriptions.
  • Cursor's security docs still name the rules page Settings > General > Auto-review. The newer Oct 5 changelog says Settings calls saved rules Custom Rules. Needs re-check: look for either name.

Sources#

Steps

  1. Write each Bot's stop point into its description.
  2. Keep the same hard lines for every Bot: sends, purchases, deletes, publishing, and production changes.
  3. Make sure approvals from background work reach you before they expire.
  4. Review Always allow rules you've saved.
  5. Set a Team Bot's limits as if nobody will be there to approve.
  6. On a team, keep admin rules short and few.

Source: Cursor Docs: Grok Bot security

Sources

  1. cursor.com
  2. cursor.com
  3. cursor.com
  4. cursor.com
  5. cursor.com
  6. cursor.com
  7. x.ai
  8. x.ai
  9. x.ai
  10. x.ai
  11. cursor.com
  12. x.ai

Something wrong? Tell the bot