Automate repeatable work
Use schedules and monitors to start well-bounded work while preserving ownership, visibility, and spending limits.
Log in or sign up to make these docs interactive
Choose your team, project, or agent to turn guidance into links to the exact place in your workspace.
Automate a workflow after you can describe its trigger, outcome, owner, and safe stopping conditions. The automation should create clarity, not hide a recurring mystery.
Quick steps
- Run the workflow manually from a card and record its reliable prompt, sources, tools, destination, evidence, and typical cost.
- Open team Scheduled Tasks or the selected agent's Scheduled Tasks, then choose Create Scheduled Task.
- Enter a recognizable name, cron schedule, bounded prompt, and the agents that should participate.
- State where the result must be written and what should happen when there is nothing to report.
- Confirm the creator, agent ceiling, tools, project access, network policy, approvals, and available budget can support the run.
- Observe the first executions in their scheduled-task chat and work streams; disable or revise the task if timing, duplication, evidence, or cost is wrong.
1. Prove the workflow manually
Run the task from a card at least once. Refine the instructions, agent, tools, sources, destination, evidence, and typical cost. Capture the repeatable method in a skill when that makes the behavior easier to maintain.
2. Choose the trigger
Create a scheduled task when time or cadence is the reason to run. Create a monitor when the reason is a supported condition on a card, board column, or agent workspace.
Use an ordinary dependency or assignment when it already expresses the flow. Do not poll for a state the board can represent directly.
3. Define the automated outcome and record
For a scheduled task, open Scheduled for the team or for the selected agent, choose Create Scheduled Task, then set its name, cron schedule, prompt, and mentioned agents. The run acts for the task's creator in a scheduled-task chat. If the result must update a project or card, identify it in the prompt and make sure the creator and agent can reach it.
For a monitor, create it from the relevant card, column, or workspace and name the condition and response. Its work stays attached to that monitored context.
Design for repeated execution. Explain the relevant time window or state, avoid duplicate outputs, and say what should happen when there is nothing to report.
4. Set limits and ownership
Confirm the creating person's access and the agent's ceiling, working style, tools, and network access. Check team funding for every automation. Check a project budget only when the work actually uses that project. Identify the person responsible for paused, failed, or unexpected runs.
5. Observe the first executions
For schedules, inspect the scheduled-task conversation and work stream. For monitors, inspect the attached card, column, or workspace record. Confirm the trigger fires once at the intended condition and that a failure is visible. Budget or policy pauses should be addressed explicitly, not worked around with a second automation.
Disable or revise automations whose outputs are no longer reviewed. An unconsumed report is not valuable merely because it runs reliably.
What good looks like
Each execution has a recognizable owner and outcome, appears in the same work system as manual delegation, respects hard limits, and produces a result that someone actually uses.