Conditions

A condition is a piece of “world state” the handler may want to look at when it runs. A provider component registers a getter under a condition name; when the trigger fires, the runtime calls the getter once and freezes the value for the rest of that run.

That’s a small idea with a big consequence: the trigger pulls state, the provider doesn’t push it. No subscriptions, no re-renders, no diff tracking. Triggery doesn’t have a reactive graph — it has a Map of () => T functions that get called exactly when an event needs them.

Declaring a condition

Conditions live in the schema’s conditions map. Each entry maps a name to the value type the getter must return:

src/triggers/message.trigger.ts

import { createTrigger } from '@triggery/core';

type Settings = { sound: boolean; notifications: boolean; dnd: boolean };

export const messageTrigger = createTrigger<{
  events:     { 'new-message': { channelId: string; text: string } };
  conditions: {
    settings:        Settings;
    activeChannelId: string | null;
    currentUserId:   string;
  };
  actions: { showToast: { title: string; body: string } };
}>({
  id: 'message-received',
  events: ['new-message'],
  required: ['settings', 'currentUserId'],
  handler({ event, conditions, actions, check }) {
    if (!conditions.settings) return;
    if (event.payload.channelId === conditions.activeChannelId) return;

    if (check.is('settings', (s) => s.notifications)) {
      actions.showToast?.({ title: 'New message', body: event.payload.text });
    }
  },
});

Three names, three value types. Whether they’re actually registered at fire time is a separate question — see required gate below.

Registering a getter

Providers register getters through a binding hook. useCondition takes the trigger, the condition name, a getter function, and (in React) an optional deps array with the same semantics as useMemo:

React

src/features/SettingsPanel.tsx

import { useCondition } from '@triggery/react';
import { useState } from 'react';
import { messageTrigger } from '../triggers/message.trigger';

export function SettingsPanel() {
  const [settings, setSettings] = useState({
    sound: true, notifications: true, dnd: false,
  });
  // The runtime calls `() => settings` only when 'new-message' fires.
  useCondition(messageTrigger, 'settings', () => settings, [settings]);
  // …UI to edit settings
}

Solid

src/features/SettingsPanel.tsx

import { useCondition } from '@triggery/solid';
import { createSignal } from 'solid-js';
import { messageTrigger } from '../triggers/message.trigger';

export function SettingsPanel() {
  const [settings] = createSignal({ sound: true, notifications: true, dnd: false });
  // Solid signals are already getters — pass them directly.
  useCondition(messageTrigger, 'settings', settings);
}

Vue

src/features/SettingsPanel.vue

<script setup lang="ts">
import { useCondition } from '@triggery/vue';
import { ref } from 'vue';
import { messageTrigger } from '../triggers/message.trigger';

const settings = ref({ sound: true, notifications: true, dnd: false });
useCondition(messageTrigger, 'settings', () => settings.value);
</script>

The deps array works exactly like useCallback / useMemo: when any dep changes, the runtime re-reads the latest closure. The getter itself is wrapped in a stable ref so re-renders don’t re-register the condition with the runtime.

Pull-only — what that means in practice

The getter runs only at fire time. Concretely:

• No subscriptions are set up between provider state and the runtime.
• Provider re-renders do not call the getter, do not invalidate anything, do not notify any other component.
• A trigger that never fires never asks for the value. Cheap.
• The same handler reading conditions.settings twice in one run sees the same value. The proxy caches per-run.

This is the property that decouples the trigger from the React render cycle. A toast layer, a settings panel and a chat list can all live next to a trigger without any of them rendering when a message arrives.

provider state changes  →  no runtime work
event fires             →  for each subscribed trigger:
                              for each condition the handler reads:
                                  call getter() exactly once
                              run handler with the frozen snapshot
                              dispatch actions

Reading conditions inside the handler

Every condition the handler can mention is typed as T | undefined. The reason is honest: until V1.1’s builder API lands, TypeScript can’t narrow the type based on required: [...]. So you have three safe ways to read:

handler({ conditions, check }) {
  // 1. Manual guard — fine for one or two.
  if (!conditions.settings) return;
  if (!conditions.settings.notifications) return;
  // conditions.settings is now Settings (narrowed by TS).

  // 2. `check.is` — typed predicate. Returns false if the condition is absent.
  if (!check.is('settings', (s) => s.notifications)) return;

  // 3. `check.all` / `check.any` — multi-condition predicates (see Handlers page).
  if (!check.all({ settings: (s) => s.notifications, currentUserId: () => true })) return;
}

Use the required field for “the handler is meaningless without this”. Use check.is for “the handler should skip when X isn’t true”. They compose.

The required gate

A condition listed in required: [...] must have at least one registered getter at fire time, or the handler is skipped before it runs. The inspector records a 'skipped' entry with reason 'missing-required-condition:<name>'.

createTrigger<Schema>({
  id: 'message-received',
  events: ['new-message'],
  required: ['settings', 'currentUserId'],   // both must be registered
  handler({ conditions }) {
    // We know they exist at runtime — but TS still sees `T | undefined` in V1.
  },
});

This is the gate that lets you wire scenarios safely. If <UserProvider> hasn’t mounted yet, 'new-message' events arrive, get a recorded skip, and the user sees nothing. Once the provider mounts, the next event runs the handler normally. The provider can be in a different feature folder than the trigger; the trigger doesn’t import it.

In the inspector this shows up as:

triggery/message-received/fire   status: 'skipped'   reason: 'missing-required-condition:currentUserId'

— a precise signal that something is mounted late or not at all.

See Trigger anatomy → required for the full mechanics.

Adapter pattern — wrapping external stores

A useState value is a trivial getter. The same pattern works for any external store — Zustand, Redux, Jotai, MobX, Signals, TanStack Query. The adapter packages do this in ~30 lines: they read from the store synchronously and hand the value back through useCondition.

With @triggery/zustand

import { useStoreCondition } from '@triggery/zustand';
import { useCurrentUserStore } from '../stores/user';
import { messageTrigger } from '../triggers/message.trigger';

function CurrentUserProvider() {
  useStoreCondition(messageTrigger, 'currentUserId', useCurrentUserStore, (s) => s.id);
  return null;
}

The Zustand store re-renders the provider when its slice changes — but that’s the provider’s business, and Triggery doesn’t care. The runtime still only calls () => s.id at fire time, and the cached useStore selector hands back the latest value.

The same shape works for @triggery/redux, @triggery/jotai, @triggery/query, and any store you author yourself — see Adapters.

Last-mount-wins ownership

What if two providers register the same condition name on the same trigger? The runtime keeps both registrations on an internal stack, with the most recently mounted at the top — the value the handler sees.

function ProviderA() {
  useCondition(messageTrigger, 'currentUserId', () => 'alice');   // mounted first
  return null;
}
function ProviderB() {
  useCondition(messageTrigger, 'currentUserId', () => 'bob');     // mounted second — wins
  return null;
}

// Mount order: <ProviderA /> then <ProviderB />.
// Handler reads conditions.currentUserId → 'bob'.
// When <ProviderB /> unmounts, 'alice' takes over (stack pop).

Two intentional consequences:

1. Tests and overrides are simple. A test mounts a provider with the test value second; it wins. A feature flag overrides the global provider by mounting deeper in the tree.
2. StrictMode is safe. React 18 StrictMode mounts → unmounts → mounts in development. The stack model handles this without spurious collisions because each registration is stable.

In DEV, the runtime emits a one-time console.warn per (triggerId, conditionName) when a second live registration arrives:

[triggery] multiple condition registrations for "currentUserId" on trigger "message-received" —
last-mount-wins. To compose values from several sources, register through a single hook.

This is a heads-up that probably something is wrong. If you genuinely want last-mount-wins (overrides, tests), it’s fine to ignore. See Ownership for the full discussion and patterns for composing values when you have multiple sources of truth.

Scoped conditions

Conditions register globally by default. Wrap a subtree with <TriggerScope id="..."> and conditions registered inside become visible only to triggers whose scope matches:

<TriggerScope id="chat">
  <SettingsPanel />          {/* settings condition — only visible to scope='chat' triggers */}
  <ChatRoom />
</TriggerScope>

<TriggerScope id="settings-screen">
  <SettingsPanel />          {/* same component, separate registration, isolated */}
</TriggerScope>

Two <ChatRoom> instances in two scopes operate on two independent settings conditions. The scope key is a string — you can match by tenant, by route, by feature flag. See Scopes.

Common patterns

Conditions are nouns, not actions. settings, currentUserId, cartTotal — not getSettings or loadCart. The verb is implicit in “read this at fire time”.

One condition per concept. Don’t bundle unrelated values into one appState condition just because they happen to be in the same Zustand store. The runtime cares about whether the value is registered; mixing concerns makes the required list lie.

Keep getters cheap. They run at fire time, and a misbehaving getter throws inside the dispatch loop. Don’t fetch from a getter. Don’t synchronously decompress a 3 MB blob. If the value is expensive, memoize at the provider level and have the getter return the cached result.

Use check.is instead of nested if. It’s shorter, it handles the undefined case for you, and the inspector records a snapshotKeys entry only when the condition was actually consulted. Cleaner traces.

Actions The opposite port — side-effect channels registered by reactors.

Ownership Last-mount-wins, multiple providers, DEV warn-once, composing sources.

Trigger anatomy The full createTrigger reference — including the required gate.

Handlers ctx walkthrough — including check.is / check.all / check.any patterns.