@objectstack/plugin-trigger-record-change

Auto-launch ObjectStack flows on record changes.

The automation engine ships the FlowTrigger extension point and the wiring that turns a flow's start node into a normalized trigger binding — but the concrete record-change trigger lives here, as a plugin. This mirrors the connector split (engine baseline + connector-rest plugin) and reuses the same kernel:ready → getService('objectql') pattern plugin-audit uses to reach the data engine's lifecycle-hook surface.

What it does

A flow whose start node declares a record-change trigger:

{
  type: 'start',
  config: {
    objectName: 'showcase_task',
    triggerType: 'record-after-update',
    condition: "status == 'done' && previous.status != 'done'", // optional
  },
}

auto-launches whenever a matching mutation happens — no manual engine.execute(). The engine evaluates the optional condition (the start node gate) before running the flow, with record (the new row) and previous (the old row) in scope.

Trigger event → hook mapping

| triggerType | ObjectQL hook | | ------------------------ | -------------- | | record-after-create | afterInsert | | record-after-update | afterUpdate | | record-after-delete | afterDelete | | record-before-create | beforeInsert | | record-before-update | beforeUpdate | | record-before-delete | beforeDelete |

Usage

import { AutomationServicePlugin } from '@objectstack/service-automation';
import { MessagingServicePlugin } from '@objectstack/service-messaging';
import { RecordChangeTriggerPlugin } from '@objectstack/plugin-trigger-record-change';

kernel
  .use(new AutomationServicePlugin())   // engine + flows
  .use(new MessagingServicePlugin())    // notify channels (optional)
  .use(new RecordChangeTriggerPlugin()); // ← makes record-change flows live

Requires the ObjectQL engine (com.objectstack.engine.objectql). If either the automation service or the data engine is unavailable at kernel:ready, the plugin logs a warning and no-ops rather than failing startup.

Error isolation

A flow that throws during a triggered run is logged and swallowed — it never breaks the CRUD write that triggered it.