When to use a worker
Use a worker when all of these are true:- Work arrives from a repeatable trigger.
- The job can be split into clear stages with explicit outputs.
- The channel’s context and tool policy provide a safe operating boundary.
- A human gate can catch steps that need judgment.
- The organization accepts cloud execution for the selected tools and data.
Worker structure
New workers begin as drafts. Configure and test them before activation.
Create a worker
1
Prepare a shared channel
Set the channel’s members, grounding, and tool policy first. A worker cannot safely compensate for an ambiguous channel boundary.
2
Open Administration → Workers
Select Create worker. If the action is unavailable, confirm the organization owner has enabled the feature.
3
Define identity and input
Name the worker for the outcome and define structured input fields. Avoid one free-form field when the trigger can provide tenant, device, priority, or ticket ID separately.
4
Build stages
Give each stage one job, a clear completion condition, and only the tool domains or individual tools it needs.
5
Add gates and verification
Use a human gate before consequential external actions. Add exit criteria that prove the stage produced usable evidence.
6
Test as a draft
Run representative inputs, failure cases, missing credentials, and policy-denied tools. Review the resulting work item and channel activity.
7
Activate and monitor
Enable the trigger only after the draft is ready. Pause the worker if behavior, source data, or policy becomes uncertain.
Tool policy and credentials
A worker stage can narrow the tools available to it, but cannot exceed the bound channel or organization policy. If a selected domain is disabled at the channel layer, the runtime drops it even if the worker editor still contains the old grant. Workers are organization actors rather than member sessions. Plan credential access deliberately:- Prefer read-only organization service credentials for unattended collection.
- Keep tenant and target explicit in stage instructions and tool inputs.
- Do not assume a member’s personal Key Chain entry is available to a cloud worker.
- Use human gates for writes, purchases, user communication, or other irreversible external effects.
Human gates
A stage with Human gate pauses before leaving the stage. The reviewer should inspect input, evidence, proposed action, and channel scope before approving or answering the task. A good gate asks for a decision a person can actually make. “Approve remediation onedge-03 using change CHG-123” is better than “Continue?”
Open worker items can surface in the channel’s work flow and chat board. Assign ownership for reviewing gates; an unattended gate is a stalled automation.
Operate the lifecycle
- Draft — editable and not trigger-active.
- Active — accepts enabled triggers and runs work.
- Paused — stops new triggered work while preserving the definition and history.
- Archived — retained for history and no longer operated.
Troubleshooting
Related
Channels and the chat board
Define the grounding, membership, and tool ceiling a worker inherits.
Organization administration
Manage organization context, integrations, members, and effective access.