Skip to content

Unofficial community registry. Not affiliated with, endorsed by, or sponsored by xAI.

中文

Guides

Test once before you trust it

test-before-use

Run an example whose answer you already know. If it does not match, do not put it on a schedule.

for Grok Bot · unofficial, not affiliated with xAI

CNPS0.1.02026-11-18New

Paste this note into Grok Bot and say: follow this note.

Note: Test once before you trust it
Run an example whose answer you already know. If it does not match, do not put it on a schedule.

For: People about to hand a new skill to a daily schedule.

What to check:
1. Asks for a small input and an answer you wrote down first.
2. Checks the output file, the finish line, and whether it did something the entry does not name.
3. Keeps a failed test on manual runs. It does not go on the calendar.

Steps:
1. Prepare a small input with no customer data. Write two or three points you expect to see.
2. Run it by hand. Do not turn on the schedule yet.
3. Compare the expectation, the file, and the finish line. A missing finish line means add a run receipt before any schedule.
4. Look for a send, a delete, or a settings change. If one appeared, do not automate it, even if the prose looks right.

Trial input (no customer data):
Three things I expect:
1.
2.
3.
Actual file path:
Finish line: yes / no
Action the entry does not name: no / yes (what)
Ready for a schedule: no, until the lines above match

Do not:
- Do not use a real customer, a password, or payment details in a trial.
- Do not lower the risk level because a trial passed.
- Do not grant send, delete, or pay access for a trial.

Paste this note into Grok Bot and say: follow this note.

What it does

  • Asks for a small input and an answer you wrote down first.
  • Checks the output file, the finish line, and whether it did something the entry does not name.
  • Keeps a failed test on manual runs. It does not go on the calendar.

Who it is for

People about to hand a new skill to a daily schedule.

How to put it in Grok Bot

  1. Prepare a small input with no customer data. Write two or three points you expect to see.
  2. Run it by hand. Do not turn on the schedule yet.
  3. Compare the expectation, the file, and the finish line. A missing finish line means add a run receipt before any schedule.
  4. Look for a send, a delete, or a settings change. If one appeared, do not automate it, even if the prose looks right.

Example

Test note

Trial input (no customer data):
Three things I expect:
1.
2.
3.
Actual file path:
Finish line: yes / no
Action the entry does not name: no / yes (what)
Ready for a schedule: no, until the lines above match

Connectors it needs

No connector.

Safety and permissions

  • Do not use a real customer, a password, or payment details in a trial.
  • Do not lower the risk level because a trial passed.
  • Do not grant send, delete, or pay access for a trial.

Risk: Low

Questions

Is one trial enough?

It is enough to catch an obvious gap. Retest after you change a step or a permission. One pass is not a promise about the next run.

Can I trial a routine on production data?

Not the first time. Prove the path and the stop on a small sample. Then consider a read-only look at live data.