AI-assisted development

AI coding agents work best when the project gives them concrete local instructions and a predictable way to add Vendure code. New Vendure projects created with @vendure/create include an AGENTS.md file for this purpose.

Project instructions

The generated AGENTS.md file summarizes the project layout, common commands, and Vendure-specific guidance. Agents should read this file before making changes. It explains where plugin code belongs, where runtime configuration lives, and which commands should be run after code changes.

Non-interactive scaffolding

The Vendure CLI supports non-interactive commands, which are useful for AI agents and automation. For example:

# Create a pluginnpx vendure add -p reviews# Add a service to a pluginnpx vendure add -s ReviewService --selected-plugin ReviewsPlugin# Add job queue support to a servicenpx vendure add -j ReviewsPlugin --name sync-reviews --selected-service ReviewService

Prefer these commands over manually creating boilerplate, because the CLI can update the surrounding Vendure configuration for you.

Plugin configuration

Plugin code should not read environment variables directly. Read deployment-specific values in src/vendure-config.ts and pass them through plugin init options:

ReviewsPlugin.init({    apiKey: process.env.REVIEWS_API_KEY,});

Then inject the plugin options token into services that need those values. This keeps plugins reusable and makes tests easier to configure.

Transactions and request context

When a RequestContext is available, pass it to Vendure services and to TransactionalConnection methods:

async update(ctx: RequestContext, id: ID) {    return this.connection.getRepository(ctx, Review).save({ id });}

This allows Vendure to apply the correct transaction and request-scoped behavior.

Job queues

Create custom job queues in a Nest lifecycle hook, then reuse the queue when adding jobs:

async onModuleInit() {    this.queue = await this.jobQueueService.createQueue({        name: 'sync-reviews',        process: async job => {            // ...        },    });}

Do not pass a raw RequestContext as job data. Use ctx.serialize() or pass only the fields needed to recreate the context in the worker.