How this helps you
Proving a job works once before scheduling it means your routines fail far less often.
That worked. Save what you just did as a skill: when to use it, what inputs and access it needs, the steps, how to check the result, and what needs my approval.
What they did
@Michael_Fenech_ says that if your routines keep failing, you might be automating them too early. His 6-step test: run the job manually first, fix the process until the result is consistent, save the working process as a skill, test it on different input, define what failure looks like (source unavailable, stale data, partial completion, approval needed, something unexpected), then use a Test run and check the run history. His summary: Task, Fix, Skill, Test, Routine. @Cyb3rCricket made the same point in a post that has since been deleted.
Who it's for: Anyone about to create their first routine.
How to try it
- Run the real job once in chat, with real inputs.
- Correct the Bot until the result is exactly what you want.
- Run it a second time to check it works twice in a row.
- Ask the Bot to save it as a skill (prompt below).
- Create a routine that calls the skill, then use Test with safe inputs. Test does real work.
What to expect
A saved skill you can see in the / menu, and a routine that calls it.
Common mistakes
- Scheduling a job that has never worked by hand.
- Skipping Test because it does real work. Use a safe copy or a test channel instead.
Related tips
- Keep routines one line long: The schedule says when. The skill says how.
- Write the outcome, not the recipe: Describe what you want; pasted tool calls break by week two.
- Teach a task by doing it once: Do the workflow while it watches; it records the steps.
- Make your first task boring: Start with something checkable, then add one thing at a time.
- A /clean-start skill for long chats: Save the goal, decisions, files and next action to a handoff file, then pick up in a fresh chat.
- Give deep research a finish line: "Research this deeply" gives no finish line and no reason to stop browsing.
- Keep skills in a GitHub repo and sync them to your Bots nightly: One-line routines call skills; the skills live in a repo you edit elsewhere, and a nightly routine pulls them onto the Bots' computer.
- Plan every Bot job with four blocks: Corey Ganim runs 6 Bots and 14 weekly routines across 11 apps. Each job is built from the same four blocks: Bot, routine, skill and connector.
- Make a write-like-me skill from your sent email: One sentence has a Bot read your sent email, find how you write, and save it as a skill every Bot uses when drafting for you.
- Build a weekly research briefing Bot from scratch, then test run it: Create a research Bot with a clear role, check it understood, connect Gmail, save a report skill, and schedule it for Monday 9:00 a.m. Then press Test run instead of waiting a week.
- Teach a task the smart way: describe it first, record only the gap: Most of a browser job can be explained in chat. Record a 20 to 60 second clip of only the part the Bot can't work out, then add anti-jobs before you schedule it.
- Turn a repeated job into a script the skill runs: Do a weekly job once with your Bot, check it, then have it rewrite the job as a plain script and save a skill that runs the script and reports in two lines.
Worth knowingThe official docs describe the same order, and note that a Test run does real work.Next in your path →Write the outcome, not the recipe
Learn more: How to Use Grok Bot: A Beginner's Guide (2026) · All Grok Bot skills tips