Skip to content

Scheduled work

A schedule can start work at a configured time. A wait pauses an existing run until a condition is met. Both let a team continue work after the original interaction ends.

A scheduled occurrence can start a new goal or playbook run once the destination workspace enables its policy. A signal is an observation evaluated against a trigger. A wait belongs to an existing run and preserves its identity when work resumes. These paths all use the current team’s admission policy, limits, and connector grants.

Use an input or prior output that resolves to the target time. This step is a fragment of a structured playbook’s steps array.

{
"id": "wait-for-window",
"title": "Wait for the reporting window",
"instruction": "Resume when the declared reporting window opens.",
"type": "wait_until",
"waitUntil": "/input/resumeAt"
}

A wait_until step reads the target timestamp through its JSON Pointer. Supply a time with an explicit offset so the intended moment is unambiguous. When the wait is due, continue from the saved step in that attempt and apply the authorization rules for the next operation.

Store the trigger occurrence or wait identity with the execution. When a worker restarts, resume that run from its recorded state. Deduplicate repeated trigger delivery. Recheck access, activation and limits before dispatching a due occurrence. Inspect held or uncertain work before release. Scheduled work imported in a package remains inactive until enabled in the destination.

The Standard Agent Specification models deferred side effects as independently scheduled handlers. Gabriel’s examples use schedules, team triggers, and playbook waits. Compare Standard Agent Specification’s Effects.