Skip to Content
IDEsGrok BotApprovals

Grok Bot approvals and boundaries

Unscoped write access makes Bots retry failed sends, re-navigate blocked pages, and burn computer-use tokens. Set boundaries in the request, in the Bot description, and in Auto Review rules.

Hub: Tokenminning in Grok Bot.

Set boundaries in the request

Tell the Bot what it can do and where it must stop:

Reconcile the campaign data and draft a recommended budget change. Do not change the campaign or message the agency. Ask for approval after showing current value, proposed value, and expected impact.

Prefer explicit boundaries for:

  • Sending messages or invitations
  • Publishing content
  • Purchases and financial transfers
  • Deleting or overwriting data
  • Changing permissions or production systems
  • Accepting legal terms

An approval controls the proposed action. It does not reverse work already completed.

Auto Review rules

When available, open Settings → General → Auto-review:

Rule typeEffect
Require ApprovalAlways stop matching actions
Always AllowProceed unless another rule stops the action

If both match, Require Approval wins.

Write narrow rules:

  • Require approval before sending any external email
  • Require approval before changing a production dashboard
  • Always allow git status in /workspace/reports

Avoid broad rules such as “allow everything in the browser.” Auto Review is model-based — complement it with least privilege, not replace it.

Personal Auto-review rules live on the current desktop and sync to its Grok Bot computer. Verify them on each installation.

Local-computer execution

Under Settings → General → Agent → Execution on Local Computer:

SettingWhen to use
Ask every timeDefault — approve each local command
Always allowedRare; specific local workflows only
Never allowedBots use cloud computer only — lowest local risk

These settings do not block cloud computer use.

Draft before execute

Patterns that cut wasted retries:

PatternToken impact
Read-only first passBot gathers evidence before write attempts
Draft in conversationYou review before any external action
Source links requiredFewer hallucinated follow-ups
Stale-data stopBot reports failure instead of using old exports

Align with routine design: automate preparation, keep sends and production changes behind approval.

Secrets and handoffs

For passwords, passkeys, 2FA, and CAPTCHAs:

  1. Open Agent Computer
  2. Take control
  3. Complete the sensitive step
  4. Return control and tell the Bot to continue

Do not send passwords or one-time codes in chat. Use secure secret requests for supported connections.

Review access regularly

When a project ends:

  1. Pause or delete related routines
  2. Sign out of websites on the shared computer
  3. Uninstall connectors and revoke authorization in the source service
  4. Remove sensitive files from /workspace
Last updated on