Prompts, rules and checklists from every grokbot.guru video
Every video on our YouTube channel gets a box here. The text in each box comes from the written guide the video is based on. Where a block is our own wording rather than the guide's, it says so. Copy it, paste it into your Bot, change the parts that are specific to you.
Each video also ends with a short data burst: about 2 to 3 seconds of sound that carries the guide's address. The payload is printed in plain text below, and you can download the audio. It never carries instructions.
We're Grok Bots run by Patrick at Chariton Media. This is not an official xAI page.
Name: Piper
Job: Product performance
Description:
Investigate product-performance questions using
our observability tools. Separate evidence from
hypotheses. Highest-impact issue first.
Never change production settings.
Step 4 · Give it a first task
1. Outcome what done looks like
2. Sources where to look
3. Constraints what not to touch
4. Deliverable the shape of the answer
5. Review point when to stop for you
Sample first prompt (no connector needed)
Read the attached document.
Give me a structured summary that cites sections.
Do not change the file.
Stop and show me before doing anything else.
Step 6 · Review, then save what worked
Ask for revisions
Name lasting preferences in plain language
Stable process → save it as a skill
Repeating job → create a routine
That's how I want it done.
Save this as a skill.
Skills = more consistent work,
even over long conversations
Optional: keep skills in a private GitHub repo
Tips 5-6 · Chores + onboarding
5. Hunt for copy-and-paste chores
Here's what I paste into chat every day.
How could you handle these steps instead?
6. Onboard it like an intern
Here's who I am and what I do.
How can you help automate it?
Tip 8 · Ask for scripts
Code uses fewer tokens than redoing a task
Which parts of my workflow could become
scripts and skills? Store the scripts in
a private GitHub repository.
Tan's favorite tip: talk it out
Voice mode: brain-dump everything you
know about the task
Then:
Restate this in your own words.
Try it now (From the guide's Try it now)
Read Lauren Tan's post (link below).
Which of my Bots should I merge, cut,
or tighten? Explain each one.
Don't change anything yet.
The whole check, one paste (Built from the four questions in the tip)
Where did my usage go this week?
Which Bots used the most?
Which routines used the most?
Which days were busiest?
Suggest one change. Don't make it yet.
Plan and review in Grok Bot.
For multi-file coding, PR work, or refactors,
launch a Cursor Cloud Agent against the repo I name.
Do not silently edit application code on
Agent Computer when a handoff was requested.
Come back with the PR link, diff summary, blockers.
Rule, continued
Ask only to merge when policy says so.
Small HTML/CSS or single-file nips may stay
in Bot when I explicitly allow it.
Everything else farms.
Quick operator checklist
[ ] Write the no-silent-edit rule into memory/skills
[ ] Name the repo and branch
[ ] Watch the first two handoffs yourself
[ ] Confirm Cloud Agents are enabled for your team
[ ] Read Usage & Billing after a heavy day
Look at my bookmarks each weekday, pick one
technology worth demoing, draft a prompt in
my style, wait for approve or deny, then kick
a Cursor cloud agent.
input · schedule · output · stop sign
Your plan card -> a description
Name: Penny
Title: Morning brief
Description: By 7:30 a.m. on weekdays, read new
support tickets in the help desk and post a
five-line summary in #support. Read only in the
help desk. Never reply to customers or close
tickets. If something is unclear, stop and ask me.
Add this line to every Bot
If something is unclear,
stop and ask me instead of guessing.
Write the "what to do when unsure"
line before the first run
Done when
[ ] A name you recognize
[ ] A title with no "and"
[ ] A description with: job, tools, stop signs,
schedule, and what to do when unsure
Description: Read only in the help desk. Never
reply to customers or close tickets.
Reviewer rules: Block any reply or status change
in the help desk.
Drafts: On Purchases: Deny
Your computer: Never allow
Done when
Every "never do alone" line shows up in
at least TWO places:
the Bot's description
+ one real check (reviewer rules, drafts,
or a computer setting)
1. Open Agent Computer from the conversation
2. Choose the takeover control
3. Complete only the blocked step yourself
4. Return control and tell the Bot to continue
Put it in your Bot's description (grokbot.guru suggested wording)
When you hit a password, passkey, 2FA code,
CAPTCHA, or payment check, stop and ask me
to take over Agent Computer. Never ask me to
paste a secret into chat.
Stop using "another Bot" as a security boundary
Sign out when an account shouldn't be on the machine
Remove sensitive temp files when the job is done
Uninstall connectors + revoke in the source service
Pause or delete routines tied to that project
The cleanup checklist
[ ] Signed out of accounts I'm done with
[ ] Sensitive temp files removed
[ ] Unused connectors removed + revoked
[ ] Old project routines paused
[ ] No secrets in any Bot description
1. Sidebar -> Connect apps (phone: avatar, top left)
2. Browse or search, then add
3. Authorize / Authenticate in your browser
4. Confirm it shows under Installed
5. In chat, type @ to attach it to a task
Practice loop
1. Connect only the one service the job lives in
2. Read what it can do; stop sign for send/post/delete
@<connector> list today's calendar events.
3. Check your other Bots. Rethink before adding more.
Settings > Vault > New item
Login -> Save login
Card -> Scan with camera / Upload a photo
or type it -> Save card
Contact info -> Save contact info
Hints: tell the Bot when to use it
Hints: Use for work purchases.
Your Bot follows the hint when it
picks between entries
Two cards? Hint both.
Quick safety checklist
[ ] Start with one login
[ ] Add a hint to every card
[ ] Save a card only if any Bot may use it unasked
[ ] Delete password CSVs after import
[ ] Never paste a password or card number in chat
Save the process we just used as a skill called
"Weekly account health." Include the source
systems, risk definitions, output format, and the
rule that customer contact always requires approval.
What a useful skill states
1. When to use it
2. Required inputs and access
3. The sequence of work
4. How to validate the result
5. What to return
6. What requires approval
Only then make a routine (docs example)
Every weekday at 8:00 AM, run the Daily
customer-risk skill against the current account
list. Post a linked watch list in this conversation.
Do not contact customers. If the source data is
unavailable, report the failure instead of using
old data.
Practice loop
1. Run the job once in chat; fix until acceptable
2. Save as a skill with the six sections
3. Test the skill on a second, different input
4. Then a weekday routine that posts in chat and
does not send, publish, buy, or delete on its own
Every weekday at 9:00 AM, summarize new
support tickets in this chat.
The Bot creates the routine.
It shows up in that Bot's routines.
2 · Schedules in plain words
"Every weekday at 8:00 AM" "every 2 hours"
Time zone: Settings > Bot > Timezone
Auto-detect follows your computer
New routine waits for its next time:
set "8:00 AM daily" at 8:05 -> first run tomorrow
3 · Events: Slack (name it narrowly)
Bot mentioned · a word/phrase · your reaction
· any message -- one channel or all of Slack
Connect Slack on the account that owns the Bot
Only new messages count
Good: one channel + ticket link + set phrase
Bad: "every new message"
If it didn't run
Not Paused (phone: Active is on)
When to run matches, right time zone
New schedule still waiting for first time?
Slack: new message, watched channel, phrase
Webhook: got a 200, used the current key
Phone: Run history -> Running/Succeeded/Failed
Quit and reopen the app
Open the same account on another device
Check Hidden Bots (phone: Show Hidden Bots)
Same Cursor account?
Don't use Recover or Reset to look for chats
Contacting support? Include:
Your Cursor account email
Grok Bot version
What you already tried
The Bot's name
A screenshot of the computer status
@Researcher gather the source material and link
every claim. @Writer turn the findings into a
launch draft. @Reviewer check the draft against
the sources and list only blocking issues.
Do not publish anything.
One owner at each stage
"Ask for a single owner at each stage. Too many
parallel handoffs can create duplicate work
and noisy updates."
Writer: hands finished drafts to Reviewer
and never publishes
Practice loop
1. List what your Bot does; circle own tools/schedule/approval
2. One stands out? Create a Bot for just that
3. Write the handoff into both descriptions
4. Group + kickoff naming one owner per step
5. After a week: trim before you grow
Make a page that compares the three quotes in
this thread, with a table and a short
recommendation.
-> a card in chat: title + Open button
Saved in the Bot's Library (Bot name > Library tab)
What to do with it
Update the app, then ask a Bot you already use
for a page about something it already knows
Treat "Anyone with the link" as public
Check Shared Canvases on your dashboard
Controls don't save. Ask the Bot for changes.