Start with ready-made AI agents with instructions on how to manage them on the marketplace. Browse the library

Ignoring Safety Rules

The security rules most often skipped during setup: access scopes, reviewing agent actions, and limits that protect your data and accounts.

ASCN Team
11 June 2026

You created a task with the schedule "every day at 8:00," saved it, and closed your laptop. In the morning you check — nothing's there. Or there is, but it's nothing like what you expected. Or the agent sent a message with a data error.

This could have been prevented with 5 minutes of testing before launch.


Why you can't trust the schedule right away

A scheduled task runs without you around. If there's an error in the instructions, a wrong table name, or an incorrect format — you'll only find out when the schedule fires. By then, the important moment has already been missed.

Testing before launch takes 5–10 minutes. It saves an entire day of waiting and rework.


How to test a task manually

The simplest way is to write to the agent in chat:

"Run the task [name] now"

or

"Execute the following instruction right now: [paste the task prompt]"

The agent will execute the task immediately, without waiting for the schedule. You see the result right now.


What to check in the test result

1. Did the agent actually start and complete the task?

If it said it can't access the data — there's an integration issue. Reconnect it.

2. Is the data correct?

Compare what the agent returned against what you see in the source manually. Do the numbers match? Is the list complete?

3. Does the message format look right?

Is it easy to read? Is there anything unnecessary, or are there too few details?

4. Is the sending working?

If the task is supposed to send to Telegram — did the message arrive? If to Slack — did it appear in the channel?

5. Are there any false positives?

If the task should only notify under a condition (for example, "only if there are new submissions") — verify that it doesn't send an empty notification when the condition is not met.


Pre-launch testing checklist

  • ☐ Ran the task manually with the "run now" command
  • ☐ Task completed without errors
  • ☐ Data in the result is correct (verified manually in the source)
  • ☐ Message format is easy to read
  • ☐ Notification arrived in the right place (Telegram / Slack / email)
  • ☐ If the task has a condition — tested both scenarios: when the condition is met and when it isn't
  • ☐ Ran the test a second time after making corrections (if any were needed)

Only after passing all items — activate the schedule.


How to read errors and what to do with them

If the task returned an error or an unclear result, ask the agent to explain:

"You just ran the task and here's what happened: [result]. Explain what went wrong and how to fix it."

Or point out a specific problem:

"You said you couldn't find the sheet 'Продажи' in the spreadsheet. The sheet is called 'Sales'. Fix the instruction."

Then run the test again.


Common errors on the first test and how to fix them

Error: "Can't find the table [name]"

Fix: Check the exact table name in Google Sheets. Make sure the agent has access through the Google Drive integration.

Error: The agent found data but for the wrong period

Fix: Clarify the period in the instructions. "Yesterday" works, but "in the last 24 hours" is sometimes more precise — it depends on your task.

Error: The message arrived in Telegram but in the wrong chat

Fix: Specify in the instructions: "send to me personally" or provide a specific chat ID / username.

Error: The agent found something but the notification didn't arrive (the condition didn't trigger)

Fix: Add a test variant to the instructions: "run and show me the result regardless of the condition."

Error: The task runs but the result is slightly different each time

Fix: Add an exact format to the instructions: a message template with placeholders.


How to confirm the schedule is actually activated

After activating the schedule — check the task status in Tasks. The task should show:

  • Status: active
  • Next run: date and time

If there's no next run — the schedule wasn't saved. You need to reactivate it.

You can write to the agent:

"Which scheduled tasks do I have active right now?"

The agent will list the active scheduled tasks.


Debug checklist: task isn't working as expected

If after a day or two the task is not working correctly:

  • ☐ Check the logs — did the task run at all (a task can be active but fail to launch if there's an error)
  • ☐ Run it manually again and look at the result
  • ☐ Check that integrations haven't dropped (Google token expired and requires re-authorization)
  • ☐ Check that the data source hasn't changed (was the sheet in the spreadsheet renamed?)
  • ☐ Simplify the task to the minimum and test it piece by piece
Build your first AI agent for free — it only takes a few minutes.
Get started for free
By continuing to use our site, you agree to the use of cookies.