Integrate Serverless Background Jobs with Inngest

Inngest Integration is an expert skill for serverless-first background jobs, event-driven workflows and durable execution without managing queues or workers. Core principles: events, not queues, are the primitive that triggers everything; each step's result is durably stored as a checkpoint; Inngest sleeps are real waits, not blocking threads; retries are automatic but policy-controlled; functions are just HTTP handlers deployable anywhere that serves HTTP; concurrency limits are a first-class way to protect downstream services; idempotency keys prevent duplicate processing on critical operations; and fan-out is built in, so one event can trigger many functions.

Its capabilities cover Inngest functions, event-driven workflows, step functions, serverless background jobs, durable sleep, fan-out patterns, concurrency control and scheduled functions. Named framework support includes Next.js, Express, Hono, Remix and SvelteKit, with deployment targets Vercel, Cloudflare Workers, Netlify, Railway and Fly.io.

Example patterns include: a basic Next.js setup defining typed events with EventSchemas().fromRecord<Events>() and a function that runs steps in sequence - fetch a user, send a welcome email, then step.sleep('wait-for-tips', '24h') before sending a tips email - served through app/api/inngest/route.ts's serve() export; a multi-step order-processing workflow that runs inventory and payment checks in parallel with Promise.all, sends a step.sendEvent fan-out event on backorder instead of a direct call, and sets retries: 3 and concurrency: { limit: 10 }; a scheduled/cron function ({ cron: '0 9 * * *' }) that fans out a daily digest to per-user child events handled by a separate concurrency-limited function; a Stripe webhook handler deduplicated with idempotency: 'event.data.stripeEventId'; and a long-running AI document pipeline that extracts text, chunks it, generates embeddings and stores vectors as four durable steps with concurrency: { limit: 5 } to bound API usage.

Ten validation checks are defined with severity: CRITICAL for a missing serve handler or a function without a unique ID; ERROR for functions not registered in the serve() call, a waitForEvent without a timeout, or a payment-related function missing an idempotency key; and WARNING for non-descriptive kebab-case step names, missing concurrency limits, undefined event schemas, a step.sleep using milliseconds instead of a duration string like '1h', and an unconfigured retry policy.

The skill's scope explicitly hands off adjacent needs to other specialists: Redis-based queues to a bullmq specialist, serverless HTTP/cron queues to Upstash QStash, complex saga/compensation workflow orchestration to a Temporal specialist, event sourcing/CQRS to an event architect, and general infrastructure to an infra architect. Five collaboration workflows are covered - Vercel background jobs, an AI pipeline, webhook processing, email automation and scheduled tasks - each naming which skill owns which step.

Use it when the user mentions or implies Inngest, a serverless background job, an event-driven workflow, a step function, durable execution, a Vercel background job, a scheduled function, or fan-out. Route Redis-queue, plain-HTTP-queue, saga/compensation, event-sourcing, deployment-configuration, database-schema or API-endpoint needs to their respective specialist skills instead of handling them here.

Input is the event schema and the business logic each step should run. Output is Inngest function definitions with typed events, durable steps, sleeps, concurrency limits, retry policy and idempotency keys, served through a framework's serve() handler.

It integrates with Next.js, Express, Hono, Remix and SvelteKit as the serving framework, and deploys to Vercel, Cloudflare Workers, Netlify, Railway or Fly.io; named companion skills are nextjs-app-router, vercel-deployment, supabase-backend, email-systems, ai-agents-architect and stripe-integration.

It suits developers building background jobs, scheduled tasks, webhook processing or multi-step AI pipelines who want durable, retryable, fan-out-capable execution without standing up and operating their own queue infrastructure.