Grok Bot Skills and Routines: Schedules, Triggers and Fixes

A skill holds the method, and a routine holds the timing. Cursor's docs give the order: start with a one-time task, make it reliable, save the method as a skill, and only then automate it.

Follow that order, because a routine runs without you. It runs in the cloud while your laptop is closed, so any gap left in the skill repeats on every run, with nobody watching to catch it.

Ask the Bot directly once a task has gone well. Name the skill and say what it should capture, such as the source systems, risk definitions, output format, and the rule that customer contact always requires approval. A useful skill states when to use it, the required inputs and access, the sequence of work, how to validate the result, what to return, and what requires approval.

To use a skill, ask the Bot for it by name. If the Bot can't find a private skill, open Connect apps in the sidebar, select the connected count next to the title, and enable the skill for the current Bot under Private skills.

Teach a task is the shortcut for browser work. From a one-to-one conversation with the computer view open, choose Teach a task, describe the result, and perform the workflow once. Teaching records visible computer interaction for up to ten minutes and doesn't record microphone audio. Keep secrets out of the demonstration, and use the secure secret request for credentials.

Ask the Bot that should own the job, in chat, and say what it should do and when. Cursor's example is "Every weekday at 9:00 AM, summarize new support tickets in this chat." The Bot creates the routine, and it appears in the Bot's routines. A fuller request also confirms the input source, the expected result, the approval boundary, and what happens when a source is missing, such as reporting the failure instead of using old data.

Schedules use the time zone in Settings > General, under Bot, labeled Timezone. Auto-detect follows the computer you're on, so pick a zone when the routine should follow a different one, such as the office where the work belongs. Plain phrasing works, like "every weekday at 8:00 AM" or "every 2 hours".

Then open the routine and read When to run, which shows the saved schedule in words. If it doesn't match what you meant, choose Edit and tell the Bot what to fix.

A new routine doesn't run when you create it. It waits for its next scheduled time, so a daily 8:00 AM routine set up at 8:05 first runs the next day. Use Test to try it right away.

Editing a routine and adding one both happen in chat, and the apps start the message for you. On desktop, open a routine and choose Edit: Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. starts a message that begins "Edit your routine:" with the routine's name, and you add what should change. On the phone, Add routine under Routines in the Bot's profile starts a message that begins "Set up a routine to " for you to finish.

Open the routine on desktop and choose Test. The button reads Running… until the run finishes, and the result shows up in the Bot's chat. If the test can't start, you see Couldn't start a test run or a message that explains why. Some routines don't show Test at all, and they still run on their schedule or trigger.

On desktop, a Bot's routines are in its details: click the Bot's name, then open the Details tab. On the phone, they're in the Bot's profile.

Deleting a routine stops future runs and can't be undone. If you delete the Bot itself, its routines go with its profile and conversation. After a long period away, Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. may ask whether to keep routines running, and it pauses them if you don't respond.

A Slack trigger is part of a routine, separate from the Slack app. The app lets a Bot post and read as the Slack user you connected. The trigger starts the routine when something happens in Slack. Connect Slack on the Cursor account that owns the Bot before you rely on a trigger, then ask the Bot for the trigger you want.

A routine can start when the Bot is mentioned, when a word or phrase is mentioned, when you react to a message, or on any message. It can watch one channel or all of Slack. Only new activity counts, so messages from before the trigger existed don't start a run.

Keep the rule narrow. Cursor's example is a message in #customer-escalations that contains a ticket link and the phrase "needs repro". Broad listeners such as every new message create noise, spend usage and act on irrelevant input.

For a webhook, ask the Bot to add a webhook trigger, then open the routine on desktop. The Webhook section has three fields, and clicking one copies it:

You can send a JSON body, and the Bot receives it with the routine instruction. A 200 response means Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. accepted the call and started a run. It doesn't mean the Bot has finished, so check the Bot's chat for the result. Any other response means no run started. Confirm the routine isn't paused and that the request uses the current key.

Start with how often they run. Each run spends usage, and Cursor says an hourly schedule, a short interval or a Slack trigger on a busy channel can use a week of usage in a day.

If your weekly usage is running out faster than you expect, first pause the routines you don't need. Then ask the Bot to widen schedules and narrow Slack triggers. Only after that, consider a bigger plan or a higher limit on extra usage. A routine that never runs spends nothing, but Test does spend usage, because it performs the work. Plan and limit details are on Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. pricing.

Check these before you assume anything is broken:

• The routine isn't Paused. On the phone, Active is on.
• When to run shows the schedule or trigger you expect, in the right time zone.
• A new schedule waits for its next time. Use Test to run it now.
• For Slack, the message is new, in the channel the trigger watches, and matches its phrase.
• For a webhook, the call returned 200 and used the current key.

On the phone, Run history lists recent runs, each marked Running, Succeeded or Failed, and a failed run shows its reason. No runs yet means nothing has started the routine.

Rate-limit failures read "Automation run rate limit exceeded". Your routines can run up to 300 times an hour in total, and each routine up to 200 times an hour. If one routine keeps hitting its own limit, ask the Bot to narrow its trigger or widen its schedule.

Cursor also lists a known issue: some runs fail with this error because of a limit all Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. routines share, even when your own routines run less than that. Its engineers are working on a fix. Either way, a run that fails with this error isn't retried automatically, so rerun it with Test or send the webhook call again.

Cursor's docs say routines earn trust the same way people do, and they list four habits:

• Automate preparation before execution.
• Keep sending, purchasing, deleting, publishing and production changes behind approval.
• Report missing or stale sources instead of working around them.
• Re-test after a website, app or source format changes.

Write the third habit into the routine request itself, because a routine that quietly falls back to last week's data looks like a success. Cursor's own example request ends with that rule: "If the source data is unavailable, report the failure instead of using old data."

Approval rules, including what Auto-review stops, are covered in Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. approvals. Routines that start from email to your Bot's own address are covered in Grok Bot agent email.