Nodejs Backend Expert

Node.js Backend Expert

> The boundary is hostile: validate input, handle every rejection, and never trust the client. Keep handlers thin, push logic into services, and make the process production-safe (timeouts, health, graceful shutdown).

When to Use

• Building/structuring an HTTP API or service in Node (Express, Fastify, Hono…).
• Routing, middleware, input validation, auth, sessions, rate limiting.
• Unhandled-rejection, error-propagation, streaming, or backpressure issues.
• Project layout: transport → service → data layer.

When NOT to Use

• Next.js server features (RSC, server actions) → nextjs-expert.
• SQL/schema/indexing → sql-expert. API contract design → api-design-expert.
• Deep auth threat modeling/OWASP → security-expert.

Core Principles

1. Layering

• Handlers/controllers stay thin (parse → call service → format response). Business logic lives in services; data access is isolated and mockable.
• Load config from env, validated at startup (zod/envalid); fail fast on missing/invalid config. No process.env.X scattered through code.

2. Async & error discipline

• async/await throughout. Every route's thrown error must reach a central error handler — wrap async handlers (or use Fastify/Express 5 which await handlers natively).
• Attach process.on("unhandledRejection")/"uncaughtException" to log and exit; don't swallow.
• Validate request body/query/params at the edge with a schema; reject early with 400 and a structured error.

3. Security baseline

• helmet for headers, explicit CORS allowlist, rate limiting on public/auth routes.
• Hash passwords with argon2/bcrypt; sign tokens/sessions with env secrets; cookies HttpOnly+Secure+SameSite. Never log secrets, tokens, or full request bodies with PII.
• Parameterize all DB access (no string-built SQL); cap body size.

4. Production-ready process

• /health (liveness) + readiness; graceful shutdown on SIGTERM (stop accepting, drain in-flight, close DB pool).
• Set server/socket timeouts. Stream large responses/uploads instead of buffering. Structured logging (pino) with request IDs.

Decision Guide

| Need | Reach for | |------|-----------| | Max performance / schema-first | Fastify | | Minimal/edge/runtime-agnostic | Hono | | Ubiquitous ecosystem / familiarity | Express (use v5 for async error handling) | | Input validation | zod / valibot at the boundary | | Heavy CPU work | worker_threads / separate service (don't block the event loop) |

Common Mistakes

• Blocking the event loop (sync crypto/fs, big JSON, CPU loops) → stalls all requests; offload to workers/streams.
• Unhandled async errors in Express 4 (thrown in a promise) → use a wrapper or Express 5.
• Trusting client input (mass assignment, unvalidated query) → validate + allowlist fields.
• Catch-and-ignore (catch (e) {}) → log with context and respond appropriately.
• No timeouts / no shutdown → hung sockets and dropped requests on deploy.
• Secrets in code or logs → env + secret manager; redact logs.

Examples

Central async error handling + validation (Express 5)

import express from "express";
import { z } from "zod";

const app = express();
app.use(express.json({ limit: "1mb" }));

const Body = z.object({ email: z.string().email(), name: z.string().min(1) });

app.post("/users", async (req, res) => {          // Express 5 awaits handlers
  const body = Body.parse(req.body);               // throws -> error middleware
  const user = await userService.create(body);
  res.status(201).json(user);
});

app.use((err, _req, res, _next) => {
  const status = err?.name === "ZodError" ? 400 : 500;
  if (status === 500) logger.error({ err }, "unhandled");
  res.status(status).json({ error: { message: err.message } });
});

Graceful shutdown

const server = app.listen(3000);
for (const sig of ["SIGTERM", "SIGINT"]) {
  process.on(sig, () => server.close(() => db.end().then(() => process.exit(0))));
}

See Also

• api-design-expert — endpoint contracts, versioning, pagination.
• sql-expert — the data layer behind services.
• security-expert — authn/authz and OWASP hardening.
• docker-expert / kubernetes-expert — packaging and running the service.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Miaoge-Ge
• Source: Miaoge-Ge/coding-agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.