Skip to main content
Use scheduled tasks to check conditions and run a sequence of steps. The task wakes on its schedule, then evaluates available evidence; a condition in the prompt is not a real-time market-event subscription. The concept behind this is Scheduled Tasks. Conditions in a prompt guide analysis; they are not notification switches. Scheduled conversation results may still be sent to a configured channel, including a no-events summary. Test runs and delivery; change task delivery settings if you do not want Telegram pushes.

Conditional tasks

Every scheduled run checks the condition first. Ask for deeper analysis or notification only when it is met, then inspect run logs to confirm the behavior:
The pattern is: check a condition, and act only if it is true. This is more useful than a task that runs the full analysis every day regardless.

Multi-step tasks

A multi-step task does several things in sequence, producing a more complete output:
Each step builds on the last, and you get one coherent report instead of three separate alerts.

Combining conditions and steps

The most powerful tasks combine both:

Tips for advanced tasks

  • State the condition precisely. “If a name moves a lot” is vague; “if a name moves more than 5%” is actionable.
  • Order the steps logically. Put the gating check first so the task does not do unnecessary work.
  • Keep output consolidated. A single combined briefing is easier to act on than several fragmented alerts.
  • Test before relying. Run the task a few times and refine the logic before trusting it for anything important.

When to split into multiple tasks instead

Not everything belongs in one complex task. Split when:
  • The pieces run on different schedules (hourly monitor vs weekly review)
  • The pieces should go to different channels (urgent alerts to Telegram, deep reviews to web)
  • One task is becoming hard to reason about

Common mistakes

  • Over-engineering. A task with too many branches is hard to debug. Start simple and add complexity only as needed.
  • Vague conditions. Conditional logic needs concrete thresholds to work reliably.
  • Silent failures going unnoticed. Review logs periodically so you know your tasks are running as intended.