Triggers

Start agents on a schedule, by webhook or on system events.

Triggers start agents without a human: on a schedule, by an inbound webhook or on a system event. The Triggers page (/triggers); the mechanics — docs/triggers-and-escalations.md.

#Cron trigger

A crontab-format schedule + project, agent and prompt. The classics: "every morning at 09:00 — sweep monitoring and summarize incidents", "every Friday — a sprint report draft". Every fire is a normal run: with a trace, a budget and approvals.

#Webhook trigger

An external system POSTs a task in: GitLab/GitHub CI failed, a request arrived, a ticket moved. Each trigger gets its own token (the generate button in the form) — no shared keys; input over HTTPS only, signatures are verified.

#Event triggers

A Journal event listener: "when knowledge goes challenged", "when a run fails", "when an artifact draft is created" — start a triage agent. The event constructor: type, source, data filters.

#Execution identity

On whose behalf the trigger starts tasks: a trigger is bound to an execution identity — run quotas and project visibility are counted against it. Channel tasks keep counting against the sender.

#Observing

Every fire is visible in "Runs" as a normal run (with a source marker), and the firing history is in the trigger's card. A failed trigger run is still a run: open it, understand the cause, restart it.