Writing Script Components for Creator Hub

This document explains how to write the .ts (or .tsx) file a Script component points at on an entity in a Creator Hub scene. These scripts run as self-contained classes attached to individual entities.

Where script files must live

Script files referenced by a Script component MUST live inside the assets/ folder — use assets/scripts/ (e.g. assets/scripts/MyScript.ts). Do NOT place them in a root-level scripts/ folder.

Why: the scene's tsconfig.json only includes src/**/* and assets/**/*:

"include": ["src/**/*.ts", "src/**/*.tsx", "assets/**/*.ts", "assets/**/*.tsx"]

A file in a root-level scripts/ folder falls outside these globs, so the TypeScript type checker (and the build's type-check step) won't cover it. The Script component's path field is resolved relative to the project root, so it must match the real file location — e.g. "path": "assets/scripts/SitChair.ts".

Script structure

Every script is a single exported class with:

• A constructor that receives configurable parameters (exposed in the Creator Hub UI).
• An optional start() method, called once when the scene loads.
• An optional update(dt: number) method, called every frame (~30 FPS; see "When scripts run" below for ordering — by default scripts update after all regular systems).

The first two constructor parameters must always be public src: string and public entity: Entity — do not remove or reorder them.

import { engine, Entity, Transform } from '@dcl/sdk/ecs'

export class MyScript {
  constructor(
    public src: string,
    public entity: Entity,
    public speed: number = 1
  ) {}

  start() {
    console.log('Script started on entity:', this.entity)
  }

  update(dt: number) {
    const transform = Transform.getMutable(this.entity)
    transform.rotation.y += this.speed * dt
  }
}

When scripts run — the priority field (IMPORTANT)

The Creator Hub Script component has a priority field (separate from constructor params — it's set in the component UI, not in your class). It defaults to 0, and this has non-obvious consequences:

• Default priority 0 = scripts run LAST each frame. At build time @dcl/sdk-commands groups all scripts by their priority value and registers ONE engine system per group via engine.addSystem(updateLoop, Number(priority)). Because engine systems run highest-priority-first (regular systems use 100000, UI uses 100000), priority 0 runs after everything else.
• All scripts sharing the same priority share ONE system callback and run sequentially inside it. A single heavy script's update() therefore delays every other script in the same priority group that frame.
• To run a script's update() before regular systems, raise its priority (e.g. 1000000) in the Script component. This also gives it its own dedicated system. This is how you'd step a physics library (cannon.js) inside update(dt) ahead of systems that read the result — there is no separate physics loop.

Higher priority number = earlier execution — the OPPOSITE of the "priority 1 = first" assumption. See the scene-runtime skill's "System Execution Order & Priority" section for the underlying engine rule.

Constructor parameters

Parameters declared in the constructor are exposed in the Creator Hub UI and can be configured per-entity. Allowed types:

Both public and private parameters are exposed to Creator Hub. Use this.<paramName> to access values in your code.

Slider<Min, Max, Step> — a number edited with a slider

public speed: Slider<0, 10, 0.5> = 1

• The runtime value is a plain number — the type is a pure alias (type Slider<Min, Max, Step = 1> = number), so nothing changes in your code or in serialization. It only tells the editor to render a slider.
• Step defaults to 1 — Slider<0, 100> steps by 1.
• An invalid declaration degrades to a plain number field (no error): missing bounds, min >= max, or step <= 0. Negative bounds and negative defaults are fine.
• A default outside the range is clamped; if you later widen or narrow the bounds, the stored value is re-clamped into the new range on script reload.

• Exported from @dcl/asset-packs, next to ActionCallback. Scenes cannot import it yet, so the Creator Hub's script template declares a local alias — copy this line into your script file:

  // A number edited with a slider in the Creator Hub UI: Slider<min, max, step>
  type Slider<Min extends number, Max extends number, Step extends number = 1> = number

  (~sdk/script-utils is its eventual home; until then the local alias is the supported form.)

Default values

Provide default values so the script works out of the box:

constructor(
  public src: string,
  public entity: Entity,
  public radius: number = 5,
  public label: string = 'Hello',
  public enabled: boolean = true,
) {}

Optional parameters

Use ? for parameters that may be left empty by the user:

constructor(
  public src: string,
  public entity: Entity,
  public targetEntity?: Entity,
  public message?: string,
) {}

Parameter tooltips

Add @param annotations in a JSDoc comment block directly before the constructor to show tooltips in the Creator Hub UI:

/**
 * @param startDate - The start date of the campaign in YYYY-MM-DD format
 * @param endDate - The end date of the campaign in YYYY-MM-DD format
 * @param wearableYOffset - How many meters above the ground the wearable should be displayed
 */
constructor(
  public src: string,
  public entity: Entity,
  public startDate?: string,
  public endDate?: string,
  public wearableYOffset: number = 0.5,
) {}

The layout field (setting asset-packs::Script by hand)

When the scene is open in the Creator Hub, write the script file and call the MCP attach_script tool with its path; it builds the component (creator-hub-mcp). When you set asset-packs::Script by hand instead, its layout JSON string holds the params, and they reach the constructor positionally: the key order in params must match the constructor order after src/entity, every param up to the last one set must be present, and the actions / events keys must be carried through on every edit. Shape, resolution and a full example: {baseDir}/references/script-layout.md.

Referencing assets with this.src

If your script uses additional assets that are only loaded via code (sound files, textures, models, etc.), they won't be automatically included in the custom item folder. You must add those files manually.

Always use this.src to build the path to bundled asset files, because the actual file location may differ when the item is used in another scene:

import { AudioSource } from '@dcl/sdk/ecs'

start() {
  AudioSource.create(this.entity, {
    audioClipUrl: this.src + '/sounds/click.mp3',
    playing: false
  })
}

Referencing child entities

Do not pass entities that belong to the same custom item as Entity input parameters. Entity IDs are not stable across scenes — an entity ID that is valid in your development scene may not exist in a user's scene.

Instead, find child entities at runtime by iterating over the entity hierarchy and matching their Name component value with a substring match (e.g. .startsWith(...) or .includes(...)).

Best practice: match a substring of the child's name, not an exact name, and not a per-instance constructor parameter. When a user duplicates a smart item or drops multiple copies into a scene, the Creator Hub editor auto-numbers the duplicated child entities — e.g. "Sit Spot", "Sit Spot 2", "Sit Spot 3", … A substring match (name.includes('Sit Spot')) finds the matching child in every copy automatically, with zero per-instance configuration.

• Exact-name match fails on every duplicate after the first, because their names are auto-suffixed.
• A constructor string/Entity parameter for the child name forces the user to manually rename or rewire each copy, which defeats the purpose of a reusable scripted item.

The example below uses .startsWith('Needle') — a substring-style match — for exactly this reason.

import { engine, Entity, Transform, Name } from '@dcl/sdk/ecs'

export class ClapMeter {
  private needleEntities: Entity[] = []

  constructor(
    public src: string,
    public entity: Entity,
  ) {}

  start() {
    for (const [childEntity, transform] of engine.getEntitiesWith(Transform)) {
      if (transform.parent === this.entity) {
        const nameComponent = Name.getOrNull(childEntity)
        if (nameComponent && nameComponent.value.startsWith('Needle')) {
          this.needleEntities.push(childEntity)
        }
      }
    }
  }
}

This pattern keeps the script portable: as long as the child entities have names containing the expected substring, it works in any scene and across any number of duplicated copies.

Defining actions (@action)

If your script has functions that could be useful to call from other items in the scene, mark them by adding a JSDoc comment block (/** ... */) with an @action tag directly before the method. Then add an Action component to the entity and define a corresponding action. This lets other smart items (e.g. a button) pick and trigger this action.

CRITICAL: Use ONLY the @action JSDoc tag — NEVER decorator syntax (@action() above the method). The Creator Hub parser has no decorators plugin, so a decorator makes parsing fail and ALL params and actions silently disappear from the UI.

An optional description line before the @action tag becomes the action's description shown in the Creator Hub UI.

import { engine, Entity } from '@dcl/sdk/ecs'

export class TreasureChest {
  private isOpen = false

  constructor(
    public src: string,
    public entity: Entity,
  ) {}

  /**
   * Opens the chest
   * @action
   */
  open() {
    if (this.isOpen) return
    this.isOpen = true
    console.log('Chest opened!')
  }

  /**
   * Closes the chest
   * @action
   */
  close() {
    if (!this.isOpen) return
    this.isOpen = false
    console.log('Chest closed!')
  }
}

With the @action JSDoc tag, open and close become available in the Actions component dropdown and can be triggered by other smart items or scripts.

ActionCallback parameters

Use the ActionCallback type from ~sdk/script-utils to let users wire up editor-configured actions as callbacks on your script. The user can then assign any action (from any item) to that callback in the Creator Hub UI.

import { Entity } from '@dcl/sdk/ecs'
import type { ActionCallback } from '~sdk/script-utils'

export class Padlock {
  constructor(
    public src: string,
    public entity: Entity,
    public onUnlock: ActionCallback,
  ) {}

  /**
   * @action
   */
  solve() {
    this.onUnlock()
  }
}

Unwired ActionCallback params are undefined

An optional ActionCallback the user never assigns in the editor resolves to undefined, so the obvious guard works:

constructor(
  public src: string,
  public entity: Entity,
  public onUnlock?: ActionCallback,   // ActionCallback | undefined
) {}

solve() {
  if (this.onUnlock) this.onUnlock()   // safe: no-op when unwired
}

Declare these params optional (?) when the behavior should work with nothing wired; a non-optional ActionCallback says the item is broken without one. The truthiness guard needs @dcl/sdk 7.28.0+; on an older pin an unwired param is a truthy placeholder whose call logs a console.error.

Calling other scripts from code

Use the runtime utilities in ~sdk/script-utils to call methods on other Script component instances:

import {
  callScriptMethod,
  getScriptInstance,
  getAllScriptInstances,
  getScriptInstancesByPath
} from '~sdk/script-utils'

callScriptMethod(entity, 'assets/scripts/Padlock.ts', 'solve', 123)

const instance = getScriptInstance(entity, 'assets/scripts/Padlock.ts')
const allOnEntity = getAllScriptInstances(entity)
const allByPath = getScriptInstancesByPath('assets/scripts/Padlock.ts')

Script-based Smart Items

Most catalog Smart Items are asset-packs::Script code with @action methods, not an Actions/Triggers graph: their "When X" hooks are optional ActionCallback params, and their state is a synced asset-packs::States component. Before inspecting, extending or copying one (which items are Script-based, hook names, reference items, seat and teleport behavior, the first-item preview error): {baseDir}/references/smart-items.md.

@event — declaring hooks for reaction scripts

A class (or start() function) JSDoc may carry @event <name> tags alongside @param and @action:

/**
 * @event enter
 * @event exit
 */
export class TriggerAreaDetector { /* … */ }

• The inspector stores them as an events: string[] key in the Script's layout JSON, next to params and actions. A non-empty events is what makes the inspector show a Reactions section, with one prompt button per event; clicking one opens the AI assistant with a natural-language sentence already seeded. Use one of the nine recognised event names so the button and the seeded sentence read naturally:

An unrecognised name still works but falls back to the generic When <event>… / when <name> fires "<event>", which reads badly to the creator. Declare @event and the no-code reaction UI appears for free.

• Tags are accepted on the export statement, the class declaration, or the constructor JSDoc — any of the three.
• Names match [A-Za-z0-9_-]+; duplicates are collapsed.

The detector/reaction split this enables. The SDK keeps only one callback per (entity, event) pair, so a second script calling triggerAreaEventsSystem.onTriggerEnter on the same entity silently replaces the first. The shipped pattern avoids that: one script owns the SDK callbacks and exposes onEvent(name, fn); reaction scripts on the same entity find it with getAllScriptInstances(entity) from ~sdk/script-utils and subscribe. A reaction must never touch triggerAreaEventsSystem itself. In the Trigger Area item's TriggerArea.tsx ({baseDir}/references/smart-items.md) the detector also replays enter to a late subscriber for anything already inside, and exposes isInside() for "while inside" logic.

⚠️ When writing the layout JSON by hand, carry events through on every edit. Rebuilding the layout from params alone drops it and the Reactions section is wiped for good. Spread the existing layout and overwrite only the one param you are changing.