Files
awesome-copilot/instructions/security-and-owasp.instructions.md
Gonzalo Fleming 6ef9d3c4fd feat(instructions): update security, a11y, and performance to 2025-2026 standards (#1270)
* feat(instructions): update security, a11y, and performance to 2025-2026 standards

Security: OWASP 2025 (55 anti-patterns, AI/LLM section, 6 frameworks)
Accessibility: WCAG 2.2 AA (38 anti-patterns, legal context EAA/ADA, 4 frameworks)
Performance: CWV (50 anti-patterns, Next.js 16, Angular 20, modern APIs)

* fix(instructions): use globalThis.scheduler to prevent ReferenceError

Access scheduler via globalThis to safely handle environments where
the Scheduling API is not declared as a global variable.

* fix(instructions): correct regex patterns and harden SSRF example

- AU1: anchor jwt.verify lookahead inside parentheses
- AU2: anchor jwt.sign lookahead, add expiresIn alternative
- AU7: fix greedy .* before negative lookahead in OAuth state check
- I5: resolve all DNS records, add TOCTOU production note
- K2: add closing delimiters and multi-digit support to tabindex regex

* fix(instructions): enhance SSRF IP validation with IPv4-mapped IPv6

Normalize IPv4-mapped IPv6 addresses (::ffff:127.0.0.1) before
checking private ranges, preventing bypass via mapped addresses.

* fix(instructions): add noscript fallback for deferred CSS pattern

Without JS, the media="print" + onload pattern leaves the stylesheet
inactive. The noscript tag loads it normally when JS is disabled.

* fix(instructions): add execFileSync to I3 command injection detection

The BAD example uses execFileSync but the regex only matched exec,
execSync, and execFile — missing the sync variant.

* fix(instructions): cover full IPv6 link-local range in SSRF check

fe80::/10 spans fe80-febf (fe8*, fe9*, fea*, feb*). Previous regex
only matched fe80::. Also use normalized variable for consistency.

* fix(instructions): adjust SSRF wording and downgrade reduced-motion severity

- SSRF: replace "full DNS/IP validation" with accurate wording that
  acknowledges TOCTOU limitation
- V5: downgrade prefers-reduced-motion from IMPORTANT to SUGGESTION,
  remove 2.2.2 (A) reference since it's an AAA enhancement

* fix(instructions): rename AU4 heading to include SHA-256

The heading said "Weak Password Hash (MD5/SHA1)" but the detection
regex and BAD example both use SHA-256. Renamed to "Fast Hash for
Passwords" which better describes the actual anti-pattern.

* fix(instructions): clarify WCAG 2.2 SC 4.1.1 status as obsolete

SC 4.1.1 Parsing is still present in the WCAG 2.2 spec but marked
as obsolete (always satisfied). Changed wording from "removed" to
"obsolete" for accuracy.

* fix(instructions): rename I1 example vars to avoid TS redeclaration

Copy-pasting the I1 SQL injection example as a single block failed with a
TypeScript redeclaration error because both BAD and GOOD snippets used
`const result`. Rename to `unsafeResult`/`safeResult` so the block remains
copy-pasteable into a single scope.

* fix(instructions): migrate I3 example to async execFile with bounds

The I3 command injection example used `execFileSync` in both BAD and GOOD
paths, which (a) redeclared `const output` in the same block and (b) blocks
the Node event loop in server handlers, amplifying DoS impact.

Switch the GOOD/BEST paths to a promisified `execFile` call with explicit
`timeout` and `maxBuffer` bounds, and rename variables to
`unsafeOutput`/`safeOutput` so the snippet stays copy-pasteable. Add a
trailing note recommending async child_process APIs for server code.

* fix(instructions): align AU6 heading with session fixation example

The AU6 heading claimed "Session Not Invalidated on Password Change" but
the mitigation example showed `req.session.regenerate`, which is the
canonical defense against session fixation on login rather than bulk
invalidation after a credential change.

Rename the anti-pattern to "Missing Session Regeneration on Login (Session
Fixation)" so it matches the example, and add a trailing note pointing to
the complementary practice of invalidating other active sessions for the
user on password change (e.g., via a `tokenVersion` counter).

* fix(instructions): make L1 critical CSS pattern CSP-compatible

The L1 "GOOD" snippet relied on an inline `onload="this.media='all'"`
handler on a `<link>` tag. Under a strict CSP that disallows
`'unsafe-inline'` / `script-src-attr 'unsafe-inline'`, inline event
handlers are blocked, so the stylesheet would never activate and users
would hit a styling regression.

Replace the pattern with build-time critical CSS extraction
(Critters/Beasties/Next.js `optimizeCss`) plus a normal
`<link rel="preload" as="style">` and standard `<link rel="stylesheet">`.
Add a trailing note explaining why the older inline-onload trick breaks
under strict CSP and how to defer non-critical CSS with an external
script when deferral is truly needed.
2026-04-10 14:40:42 +10:00

1058 lines
30 KiB
Markdown

---
applyTo: '**'
description: 'Comprehensive secure coding standards based on OWASP Top 10 2025, with 55+ anti-patterns, detection regex, framework-specific fixes for modern web and backend frameworks, and AI/LLM security guidance.'
---
# Security Standards
Comprehensive security rules for web application development. Every anti-pattern includes a severity classification, detection method, OWASP 2025 reference, and corrective code examples.
**Severity levels:**
- **CRITICAL** — Exploitable vulnerability. Must be fixed before merge.
- **IMPORTANT** — Significant risk. Should be fixed in the same sprint.
- **SUGGESTION** — Defense-in-depth improvement. Plan for a future iteration.
---
## OWASP Top 10 — 2025 Quick Reference
| # | Category | Key Mitigation |
|---|----------|----------------|
| A01 | Broken Access Control | Auth middleware on every endpoint, RBAC, ownership checks |
| A02 | Security Misconfiguration | Security headers, no debug in prod, no default credentials |
| A03 | Software Supply Chain Failures *(NEW)* | `npm audit`, lockfile integrity, SBOM, SLSA provenance |
| A04 | Cryptographic Failures | Argon2id/bcrypt for passwords, TLS everywhere, no secrets in code |
| A05 | Injection | Parameterized queries, input validation, no raw HTML with user input |
| A06 | Insecure Design | Threat modeling, secure design patterns, abuse case testing |
| A07 | Authentication Failures | Rate-limit login, secure session management, MFA |
| A08 | Software or Data Integrity Failures | SRI for CDN scripts, signed artifacts, no insecure deserialization |
| A09 | Security Logging and Alerting Failures | Log security events, no PII in logs, correlation IDs, active alerting |
| A10 | Mishandling of Exceptional Conditions *(NEW)* | Handle all errors, no stack traces in prod, fail-secure |
---
## Injection Anti-Patterns (I1-I8)
### I1: SQL Injection via String Concatenation
- **Severity**: CRITICAL
- **Detection**: `\$\{.*\}.*(?:SELECT|INSERT|UPDATE|DELETE|FROM|WHERE)`
- **OWASP**: A05
```typescript
// BAD
const unsafeResult = await db.query(`SELECT * FROM users WHERE id = ${userId}`);
// GOOD — parameterized query
const safeResult = await db.query('SELECT * FROM users WHERE id = $1', [userId]);
```
### I2: NoSQL Injection (MongoDB Operator Injection)
- **Severity**: CRITICAL
- **Detection**: `\{\s*\$(?:gt|gte|lt|lte|ne|in|nin|regex|where|exists)`
- **OWASP**: A05
```typescript
// BAD — attacker sends { "password": { "$gt": "" } }
const user = await User.findOne({ username: req.body.username, password: req.body.password });
// GOOD — validate and cast input types
const username = String(req.body.username);
const password = String(req.body.password);
const user = await User.findOne({ username });
const valid = user && await verifyPassword(user.passwordHash, password);
```
### I3: Command Injection (exec with User Input)
- **Severity**: CRITICAL
- **Detection**: `(?:exec|execSync|execFile|execFileSync)\s*\(.*(?:req\.|params\.|query\.|body\.)`
- **OWASP**: A05
```typescript
// BAD — shell interpolation, sync call blocks the event loop
import { execFileSync } from 'node:child_process';
const unsafeOutput = execFileSync('sh', ['-c', `ls -la ${req.query.dir}`]);
// GOOD — async execFile, arguments array, no shell, bounded time/output
import { execFile } from 'node:child_process';
import { promisify } from 'node:util';
const pExecFile = promisify(execFile);
const dir = String(req.query.dir ?? '');
if (!dir || dir.startsWith('-')) throw new Error('Invalid directory');
const { stdout: safeOutput } = await pExecFile('ls', ['-la', '--', dir], {
timeout: 5_000, // fail fast on hung processes
maxBuffer: 1 << 20, // 1 MiB cap to prevent memory exhaustion
});
// BEST — allowlist validation on top of the async, bounded call above
const allowedDirs = ['/data', '/public'];
if (!allowedDirs.includes(dir)) throw new Error('Invalid directory');
```
Prefer async `execFile`/`spawn` over `execFileSync` in server handlers: the sync variant blocks Node's event loop and can amplify DoS impact. Always pass a `timeout` and `maxBuffer` to bound execution.
### I4: XSS via Unsanitized HTML Rendering
- **Severity**: CRITICAL
- **Detection**: `(?:v-html|\[innerHTML\]|dangerouslySetInner|bypassSecurityTrust)`
- **OWASP**: A05
Applies to all frontend frameworks. Each has an API that bypasses default XSS protection:
- **React**: `dangerouslySetInnerHTML` prop with raw user content
- **Angular**: `[innerHTML]` binding or `bypassSecurityTrustHtml` with unsanitized input
- **Vue**: `v-html` directive with user-controlled content
```typescript
// GOOD — sanitize with DOMPurify before rendering any raw HTML
import DOMPurify from 'dompurify';
const clean = DOMPurify.sanitize(userContent);
// BEST — use text interpolation when HTML is not needed
// React: {userContent}
// Angular: {{ userContent }}
// Vue: {{ userContent }}
```
### I5: SSRF via User-Controlled URLs
- **Severity**: CRITICAL
- **Detection**: `fetch\((?:req\.|params\.|query\.|body\.|url|href)`
- **OWASP**: A01
```typescript
// BAD
const data = await fetch(req.body.url);
// GOOD — scheme allowlist + hostname allowlist + DNS/IP validation (see TOCTOU note)
import { promises as dns } from 'node:dns';
function isPrivateIP(ip: string): boolean {
// Normalize IPv4-mapped IPv6 (e.g., ::ffff:127.0.0.1 → 127.0.0.1)
const normalized = ip.startsWith('::ffff:') ? ip.slice(7) : ip;
// IPv4 private/reserved/loopback ranges
if (/^(10\.|172\.(1[6-9]|2\d|3[01])\.|192\.168\.|127\.|0\.|169\.254\.)/.test(normalized)) return true;
// IPv6 loopback, link-local (fe80::/10), and unique-local
if (/^(::1|fe[89ab]|fc|fd)/i.test(normalized)) return true;
return false;
}
const parsed = new URL(req.body.url);
if (parsed.protocol !== 'https:') throw new Error('Only HTTPS allowed');
const allowedHosts = ['api.example.com', 'cdn.example.com'];
if (!allowedHosts.includes(parsed.hostname)) throw new Error('Host not allowed');
// Resolve all A/AAAA records to prevent DNS rebinding via multiple IPs
const resolved = await dns.lookup(parsed.hostname, { all: true });
if (resolved.length === 0 || resolved.some(({ address }) => isPrivateIP(address))) {
throw new Error('Private or reserved IPs not allowed');
}
// Note: for production, pin the resolved IP in the HTTP client to prevent
// TOCTOU rebinding between this check and fetch(). See undici Agent docs.
const data = await fetch(parsed.toString(), { redirect: 'error' });
```
### I6: Path Traversal in File Operations
- **Severity**: CRITICAL
- **Detection**: `(?:readFile|readFileSync|createReadStream|path\.join)\s*\(.*(?:req\.|params\.|query\.|body\.)`
- **OWASP**: A01
```typescript
// BAD
const file = fs.readFileSync(`/data/${req.params.filename}`);
// GOOD — resolve and validate within allowed directory
import path from 'path';
const basePath = '/data';
const filePath = path.resolve(basePath, req.params.filename);
if (!filePath.startsWith(basePath + path.sep)) throw new Error('Path traversal detected');
const file = fs.readFileSync(filePath);
```
### I7: Template Injection
- **Severity**: CRITICAL
- **Detection**: `(?:render|compile|template)\s*\(.*(?:req\.|params\.|query\.|body\.)`
- **OWASP**: A05
```typescript
// BAD — user input as template source
const html = ejs.render(req.body.template, data);
// GOOD — predefined templates, user input only as data
const html = ejs.renderFile('./templates/page.ejs', { content: req.body.content });
```
### I8: XXE Injection (XML External Entity)
- **Severity**: CRITICAL
- **Detection**: `(?:parseXml|DOMParser|xml2js|libxmljs).*(?:req\.|body\.|file)`
- **OWASP**: A05
```typescript
// GOOD — disable external entities in XML parser
import { XMLParser } from 'fast-xml-parser';
const parser = new XMLParser({
allowBooleanAttributes: true,
processEntities: false,
htmlEntities: false,
});
const result = parser.parse(req.body.xml);
```
---
## Authentication Anti-Patterns (AU1-AU8)
### AU1: JWT Algorithm Confusion (alg:none)
- **Severity**: CRITICAL
- **Detection**: `jwt\.verify\((?![^)]*\balgorithms\b)[^)]*\)`
- **OWASP**: A07
```typescript
// BAD — accepts any algorithm including "none"
const decoded = jwt.verify(token, secret);
// GOOD — enforce specific algorithm
const decoded = jwt.verify(token, publicKey, { algorithms: ['RS256'] });
```
### AU2: JWT Without Expiration Check
- **Severity**: CRITICAL
- **Detection**: `jwt\.sign\((?![^)]*\b(?:expiresIn|exp)\b)[^)]*\)`
- **OWASP**: A07
```typescript
// BAD — token never expires
const token = jwt.sign({ userId: user.id }, secret);
// GOOD — short-lived token
const token = jwt.sign({ userId: user.id }, secret, { expiresIn: '15m' });
```
### AU3: JWT Stored in localStorage
- **Severity**: IMPORTANT
- **Detection**: `localStorage\.setItem\(.*(?:token|jwt|auth|session)`
- **OWASP**: A07
```typescript
// BAD — accessible via XSS
localStorage.setItem('accessToken', token);
// GOOD — httpOnly cookie set by server
res.cookie('token', token, { httpOnly: true, secure: true, sameSite: 'strict' });
```
### AU4: Plaintext / Fast Hash for Passwords (MD5/SHA-1/SHA-256)
- **Severity**: CRITICAL
- **Detection**: `(?:createHash|md5|sha1|sha256)\s*\(.*password`
- **OWASP**: A04
```typescript
// BAD — fast hash, no salt
const sha256Hash = crypto.createHash('sha256').update(password).digest('hex');
// GOOD — Argon2id (OWASP recommended)
import { hash as argon2Hash, argon2id } from 'argon2';
const hashed = await argon2Hash(password, { type: argon2id, memoryCost: 65536, timeCost: 3 });
```
### AU5: Missing Brute-Force Protection on Login
- **Severity**: CRITICAL
- **Detection**: `(?:post|router\.post)\s*\(\s*['"]\/(?:login|signin|auth|register|reset)`
- **OWASP**: A07
```typescript
// BAD — no rate limiting
app.post('/api/auth/login', loginHandler);
// GOOD
import rateLimit from 'express-rate-limit';
const authLimiter = rateLimit({ windowMs: 15 * 60 * 1000, max: 5 });
app.post('/api/auth/login', authLimiter, loginHandler);
```
### AU6: Missing Session Regeneration on Login (Session Fixation)
- **Severity**: IMPORTANT
- **Detection**: `(?:session|req\.session)\s*\.\s*(?:userId|user|authenticated)\s*=`
- **OWASP**: A07
```typescript
// GOOD — regenerate session ID on successful login to prevent fixation
req.session.regenerate((err) => {
if (err) return next(err);
req.session.userId = user.id;
req.session.save(next);
});
```
Related: on password change or elevation, also invalidate all other active sessions for the user (e.g., by bumping a `tokenVersion` column and rejecting sessions with a stale version, or by iterating the session store and destroying entries keyed to that user).
### AU7: OAuth Without State Parameter
- **Severity**: CRITICAL
- **Detection**: `authorize\?(?![^\n#]*\bstate=)[^\n#]*`
- **OWASP**: A07
```typescript
// GOOD — include state parameter for CSRF protection
const state = crypto.randomBytes(32).toString('hex');
session.oauthState = state;
const authUrl = `https://provider.com/authorize?client_id=${clientId}&redirect_uri=${redirectUri}&state=${state}`;
```
### AU8: Missing PKCE for Public OAuth Clients
- **Severity**: IMPORTANT
- **Detection**: `(?:authorization_code|code).*(?!.*code_challenge)`
- **OWASP**: A07
Use PKCE (Proof Key for Code Exchange) with S256 challenge method for all public clients (SPAs, mobile).
---
## Authorization Anti-Patterns (AZ1-AZ6)
### AZ1: Missing Auth Middleware on New Endpoints
- **Severity**: CRITICAL
- **Detection**: `(?:app|router)\.\w+\s*\(\s*['"]\/api\/(?:admin|users|settings)`
- **OWASP**: A01
```typescript
// BAD
router.delete('/api/users/:id', deleteUser);
// GOOD
router.delete('/api/users/:id', authenticate, authorize('admin'), deleteUser);
```
### AZ2: Client-Side Only Authorization
- **Severity**: CRITICAL
- **Detection**: Component guards without server-side checks
- **OWASP**: A01
Frontend guards are UX only. ALWAYS verify on server.
### AZ3: IDOR (Insecure Direct Object Reference)
- **Severity**: CRITICAL
- **Detection**: `params\.(?:id|userId|orderId)` without ownership check
- **OWASP**: A01
```typescript
// GOOD — verify ownership
router.get('/api/orders/:orderId', authenticate, async (req, res) => {
const order = await Order.findById(req.params.orderId);
if (!order || order.userId !== req.user.id) {
return res.status(404).json({ error: 'Not found' });
}
res.json(order);
});
```
### AZ4: Mass Assignment
- **Severity**: CRITICAL
- **Detection**: `(?:create|update|findOneAndUpdate)\s*\(\s*req\.body\s*\)`
- **OWASP**: A01
```typescript
// BAD
await User.findByIdAndUpdate(id, req.body);
// GOOD — explicitly pick allowed fields
const { name, email, avatar } = req.body;
await User.findByIdAndUpdate(id, { name, email, avatar });
```
### AZ5: Privilege Escalation via Role Parameter
- **Severity**: CRITICAL
- **Detection**: `req\.body\.role|req\.body\.isAdmin|req\.body\.permissions`
- **OWASP**: A01
```typescript
// GOOD — ignore role from input
const { name, email, password } = req.body;
const user = await User.create({ name, email, password, role: 'user' });
```
### AZ6: Missing Re-Authentication for Sensitive Operations
- **Severity**: IMPORTANT
- **Detection**: `(?:delete|destroy|remove).*(?:account|user|organization)` without re-auth
- **OWASP**: A01
Require current password before account deletion, email change, or other sensitive operations.
---
## Secrets Anti-Patterns (S1-S6)
### S1: Hardcoded API Keys / Tokens
- **Severity**: CRITICAL
- **Detection**: `(?:password|secret|api_key|token|apiKey)\s*[:=]\s*['"][A-Za-z0-9+/=]{8,}['"]`
- **OWASP**: A04
```typescript
// BAD
const API_KEY = 'sk_live_abc123def456';
// GOOD
const API_KEY = process.env.API_KEY;
```
### S2: .env Committed to Git
- **Severity**: CRITICAL
- **Detection**: `git ls-files .env` (should return empty)
- **OWASP**: A04
```gitignore
# .gitignore
.env
.env.local
.env.*.local
*.pem
*.key
```
### S3: Server Secrets Exposed to Client
- **Severity**: CRITICAL
- **Detection**: `NEXT_PUBLIC_.*(?:SECRET|PRIVATE|PASSWORD|KEY(?!.*PUBLIC))`
- **OWASP**: A02
```bash
# BAD
NEXT_PUBLIC_DATABASE_URL=postgresql://...
# GOOD
DATABASE_URL=postgresql://...
NEXT_PUBLIC_API_URL=https://api.example.com
```
Angular: do not put secrets in `environment.ts` files bundled into the client.
### S4: Default Credentials in Config
- **Severity**: CRITICAL
- **Detection**: `(?:admin|root|default|test).*(?:password|pass|pwd)\s*[:=]\s*['"](?:admin|root|password|1234|test)`
- **OWASP**: A02
Use environment variables with validation (zod schema).
### S5: Secrets in CI/CD Pipeline Logs
- **Severity**: IMPORTANT
- **Detection**: `(?:echo|console\.log|print).*(?:\$SECRET|\$TOKEN|\$PASSWORD|process\.env)`
- **OWASP**: A09
Use masked secrets in CI. Never echo environment variables containing secrets.
### S6: Sensitive Data in Error Responses / Stack Traces
- **Severity**: IMPORTANT
- **Detection**: `(?:stack|trace|query|sql).*(?:res\.json|res\.send|c\.JSON)`
- **OWASP**: A10
```typescript
// GOOD — generic error to client, details only in logs
app.use((err, req, res, _next) => {
logger.error({ err, path: req.path, method: req.method });
const isDev = process.env.NODE_ENV === 'development';
res.status(500).json({
error: 'Internal Server Error',
...(isDev && { message: err.message }),
});
});
```
---
## Headers Anti-Patterns (H1-H8)
### H1: Missing Content-Security-Policy
- **Severity**: IMPORTANT
- **Detection**: Absence of `Content-Security-Policy` header
- **OWASP**: A02
### H2: CSP with unsafe-inline and unsafe-eval
- **Severity**: IMPORTANT
- **Detection**: `Content-Security-Policy.*(?:'unsafe-inline'|'unsafe-eval')`
- **OWASP**: A02
Use nonce-based CSP: `script-src 'self' 'nonce-{SERVER_GENERATED}'`
### H3: Missing Strict-Transport-Security
- **Severity**: IMPORTANT
- **Detection**: Absence of `Strict-Transport-Security` header
- **OWASP**: A02
Value: `max-age=31536000; includeSubDomains; preload`
### H4: Missing X-Content-Type-Options
- **Severity**: IMPORTANT
- **Detection**: Absence of `X-Content-Type-Options: nosniff`
- **OWASP**: A02
### H5: Missing X-Frame-Options
- **Severity**: IMPORTANT
- **Detection**: Absence of `X-Frame-Options` header
- **OWASP**: A02
Value: `DENY`. Also set `Content-Security-Policy: frame-ancestors 'none'`.
### H6: Permissive Referrer-Policy
- **Severity**: SUGGESTION
- **Detection**: `Referrer-Policy.*(?:unsafe-url|no-referrer-when-downgrade)`
- **OWASP**: A02
Use: `strict-origin-when-cross-origin`
### H7: Missing Permissions-Policy
- **Severity**: SUGGESTION
- **Detection**: Absence of `Permissions-Policy` header
- **OWASP**: A02
Value: `camera=(), microphone=(), geolocation=(), payment=()`
### H8: CORS Wildcard with Credentials
- **Severity**: CRITICAL
- **Detection**: `(?:cors|Access-Control-Allow-Origin).*\*`
- **OWASP**: A02
```typescript
// GOOD
app.use(cors({
origin: ['https://app.example.com', 'https://staging.example.com'],
credentials: true,
}));
```
---
## Frontend Anti-Patterns (FE1-FE8)
### FE1: Unsanitized HTML Rendering
- **Severity**: CRITICAL
- **Detection**: `(?:innerHTML|v-html|dangerouslySetInner)` without DOMPurify
- **OWASP**: A05
Always sanitize with DOMPurify before rendering user-controlled HTML. See I4.
### FE2: Dynamic Code Evaluation with User Input
- **Severity**: CRITICAL
- **Detection**: `eval\s*\(`
- **OWASP**: A05
Use structured data parsers (JSON.parse) instead.
### FE3: postMessage Without Origin Validation
- **Severity**: IMPORTANT
- **Detection**: `addEventListener\s*\(\s*['"]message['"].*(?!.*origin)`
- **OWASP**: A01
```typescript
window.addEventListener('message', (event) => {
if (event.origin !== 'https://trusted.example.com') return;
processData(event.data);
});
```
### FE4: Prototype Pollution
- **Severity**: IMPORTANT
- **Detection**: `(?:__proto__|constructor\.prototype|Object\.assign)\s*.*(?:req\.|body\.|query\.)`
- **OWASP**: A05
Validate and filter keys from user input before merging into objects.
### FE5: Open Redirect
- **Severity**: IMPORTANT
- **Detection**: `(?:window\.location|location\.href|router\.push)\s*=\s*(?:req\.|params\.|query\.)`
- **OWASP**: A01
```typescript
// GOOD — relative paths only
const redirect = new URLSearchParams(window.location.search).get('redirect');
if (redirect?.startsWith('/') && !redirect.startsWith('//')) {
window.location.href = redirect;
}
```
### FE6: Sensitive Data in localStorage
- **Severity**: IMPORTANT
- **Detection**: `localStorage\.setItem\(.*(?:token|session|credit|ssn|password)`
- **OWASP**: A07
Use httpOnly cookies for tokens.
### FE7: Missing CSRF Token
- **Severity**: IMPORTANT
- **Detection**: POST/PUT/DELETE forms without CSRF token or SameSite cookie
- **OWASP**: A01
Use double-submit cookie or synchronizer token. Next.js Server Actions have built-in CSRF via Origin header.
### FE8: Client-Only Input Validation
- **Severity**: IMPORTANT
- **Detection**: Form validation only in frontend
- **OWASP**: A05
ALWAYS validate on server too. Use zod, joi, or class-validator.
---
## Dependencies Anti-Patterns (D1-D5)
### D1: Known Vulnerable Dependency
- **Severity**: CRITICAL
- **Detection**: `npm audit --audit-level=high` exits non-zero
- **OWASP**: A03
### D2: Lockfile Out of Sync
- **Severity**: IMPORTANT
- **Detection**: `npm ci` fails
- **OWASP**: A08
### D3: Typosquatting Risk
- **Severity**: IMPORTANT
- **Detection**: Manual review of new dependency names
- **OWASP**: A03
### D4: Postinstall Scripts in New Dependency
- **Severity**: IMPORTANT
- **Detection**: `"postinstall"` in new dependency's package.json
- **OWASP**: A03
### D5: Unpinned Versions in Production
- **Severity**: SUGGESTION
- **Detection**: `":\s*["']\*["']|":\s*["']latest["']`
- **OWASP**: A03
---
## API Anti-Patterns (AP1-AP6)
### AP1: New Endpoint Without Rate Limiting
- **Severity**: IMPORTANT
- **OWASP**: A05
### AP2: GraphQL Without Depth Limiting
- **Severity**: IMPORTANT
- **Detection**: `new ApolloServer` without depth/complexity limits
- **OWASP**: A05
```typescript
import depthLimit from 'graphql-depth-limit';
const server = new ApolloServer({
schema,
validationRules: [depthLimit(5)],
introspection: process.env.NODE_ENV !== 'production',
});
```
### AP3: File Upload Without Validation
- **Severity**: IMPORTANT
- **Detection**: `multer|formidable|busboy` without type/size checks
- **OWASP**: A05
```typescript
const upload = multer({
dest: 'uploads/',
limits: { fileSize: 5 * 1024 * 1024 },
fileFilter: (req, file, cb) => {
const allowed = ['image/jpeg', 'image/png', 'image/webp'];
cb(null, allowed.includes(file.mimetype));
},
});
```
### AP4: Webhook Without Signature Verification
- **Severity**: CRITICAL
- **OWASP**: A08
Always verify webhook signatures (Stripe, GitHub HMAC, etc.).
### AP5: API Exposing Internal Info
- **Severity**: IMPORTANT
- **Detection**: `(?:stack|trace|query|sql).*(?:res\.json|res\.send)`
- **OWASP**: A10
### AP6: Missing Request Body Size Limit
- **Severity**: IMPORTANT
- **Detection**: `express\.json\(\)` without `limit`
- **OWASP**: A05
```typescript
app.use(express.json({ limit: '100kb' }));
```
---
## AI/LLM Security Anti-Patterns (AI1-AI3)
### AI1: Prompt Injection via User Input
- **Severity**: CRITICAL
- **Detection**: User input concatenated into LLM prompts without sanitization
- **OWASP**: A05 (Injection)
```typescript
// BAD — user input directly in prompt
const response = await llm.complete(`Summarize this: ${userInput}`);
// GOOD — structured input with system/user message separation
const response = await llm.complete({
system: "You are a summarization assistant. Only summarize the provided text.",
user: userInput,
});
```
### AI2: LLM Output Used in SQL/Shell Without Sanitization
- **Severity**: CRITICAL
- **Detection**: LLM response passed to `db.query()`, `exec()`, or template literals without validation
- **OWASP**: A05 (Injection)
Never trust LLM output as safe. Treat it as untrusted user input — parameterize queries, escape shell arguments, sanitize HTML before rendering.
### AI3: Missing Output Validation from LLM Responses
- **Severity**: IMPORTANT
- **Detection**: LLM response rendered or executed without schema validation
- **OWASP**: A08 (Software or Data Integrity Failures)
Validate LLM output against expected schemas (Zod, JSON Schema) before using in application logic. Reject responses that don't match expected structure.
---
## Logging Anti-Patterns (L1-L4)
### L1: Security Events Not Logged
- **Severity**: IMPORTANT
- **OWASP**: A09
Log: auth failures, access denied, rate limit hits, input validation failures, password changes.
### L2: Sensitive Data in Logs
- **Severity**: CRITICAL
- **Detection**: `(?:log|logger)\.\w+\(.*(?:password|token|secret|ssn|credit)`
- **OWASP**: A09
```typescript
import pino from 'pino';
const logger = pino({ redact: ['req.headers.authorization', 'req.body.password'] });
```
### L3: Missing Trace IDs
- **Severity**: SUGGESTION
- **OWASP**: A09
### L4: Log Injection
- **Severity**: IMPORTANT
- **Detection**: `console\.log\(.*\+.*(?:req\.|user\.|body\.)`
- **OWASP**: A09
Use structured logging (JSON, auto-escaped) instead of string concatenation.
---
## Framework-Specific: React / Next.js (RX1-RX4)
### RX1: Server Action Without Auth
- **Severity**: CRITICAL
- **Detection**: `'use server'` function without `auth()` or session check
- **OWASP**: A01
```typescript
'use server';
import { auth } from '@/auth';
export async function deleteUser(id: string) {
const session = await auth();
if (!session?.user || session.user.role !== 'admin') throw new Error('Unauthorized');
await db.user.delete({ where: { id } });
}
```
### RX2: process.env Without NEXT_PUBLIC_ in Client
- **Severity**: IMPORTANT
- **Detection**: `'use client'` file accessing `process.env` without `NEXT_PUBLIC_`
- **OWASP**: A02
### RX3: RSC Serialization Leaking Data
- **Severity**: IMPORTANT
- **OWASP**: A01
Pick only needed fields before passing DB objects to Client Components.
### RX4: middleware.ts Not Protecting API Routes
- **Severity**: IMPORTANT
- **Detection**: `config.matcher` not covering `/api/`
- **OWASP**: A01
---
## Framework-Specific: Angular (NG1-NG3)
### NG1: bypassSecurityTrustHtml with User Input
- **Severity**: CRITICAL
- **Detection**: `bypassSecurityTrust(?:Html|Script|Style|Url|ResourceUrl)`
- **OWASP**: A05
Sanitize with DOMPurify BEFORE calling bypassSecurityTrust.
### NG2: Template Expression Injection
- **Severity**: IMPORTANT
- **OWASP**: A05
Do not use JitCompilerFactory with user-controlled templates.
### NG3: HttpInterceptor Not Attaching Auth
- **Severity**: IMPORTANT
- **OWASP**: A07
Use a centralized `HttpInterceptorFn` for auth tokens.
---
## Framework-Specific: Express (EX1-EX4)
### EX1: Missing helmet.js
- **Severity**: IMPORTANT
- **OWASP**: A02
```typescript
import helmet from 'helmet';
app.use(helmet());
app.disable('x-powered-by');
```
### EX2: express.json() Without Body Size Limit
- **Severity**: IMPORTANT
- **OWASP**: A05
```typescript
app.use(express.json({ limit: '100kb' }));
```
### EX3: Cookie Without Secure Flags
- **Severity**: IMPORTANT
- **OWASP**: A07
```typescript
res.cookie('session', value, {
httpOnly: true, secure: true, sameSite: 'strict', maxAge: 3600000, path: '/',
});
```
### EX4: Error Handler Exposing Stack Trace
- **Severity**: IMPORTANT
- **OWASP**: A10
Only expose error details in development mode.
---
## Framework-Specific: Go (GO1-GO3)
### GO1: math/rand for Security Operations
- **Severity**: CRITICAL
- **Detection**: `math/rand` import in security-related files
- **OWASP**: A04
Use `crypto/rand` for cryptographically secure random values.
### GO2: TLS InsecureSkipVerify
- **Severity**: CRITICAL
- **Detection**: `InsecureSkipVerify:\s*true`
- **OWASP**: A04
Use system CA pool (default) instead.
### GO3: String Interpolation in SQL
- **Severity**: CRITICAL
- **Detection**: `fmt\.Sprintf\s*\(.*(?:SELECT|INSERT|UPDATE|DELETE|FROM|WHERE)`
- **OWASP**: A05
```go
// GOOD — parameterized
db.Where("id = ?", userID).Find(&user)
```
---
## Security Headers Template
### helmet.js (Express)
```typescript
import helmet from 'helmet';
app.use(helmet({
contentSecurityPolicy: {
directives: {
defaultSrc: ["'self'"],
scriptSrc: ["'self'"],
styleSrc: ["'self'"],
imgSrc: ["'self'", "data:", "https:"],
fontSrc: ["'self'"],
connectSrc: ["'self'"],
frameAncestors: ["'none'"],
objectSrc: ["'none'"],
baseUri: ["'self'"],
formAction: ["'self'"],
upgradeInsecureRequests: [],
},
},
hsts: { maxAge: 31536000, includeSubDomains: true, preload: true },
frameguard: { action: 'deny' },
referrerPolicy: { policy: 'strict-origin-when-cross-origin' },
crossOriginOpenerPolicy: { policy: 'same-origin' },
crossOriginResourcePolicy: { policy: 'same-origin' },
}));
app.disable('x-powered-by');
```
---
## JWT Validation Checklist
1. Verify signature with expected algorithm — reject `alg: none`
2. Enforce algorithm: `algorithms: ['RS256']` or `['ES256']`
3. Check `exp` — reject expired tokens
4. Check `iat` — reject tokens issued too far in the past
5. Check `aud` — reject tokens not intended for this service
6. Check `iss` — reject tokens from unknown issuers
7. Store in httpOnly cookie — not localStorage
8. Use short-lived access tokens (15 min) + refresh token rotation
9. Rotate signing keys periodically
---
## Secure Cookie Flags
```
Set-Cookie: session=value; HttpOnly; Secure; SameSite=Strict; Path=/; Max-Age=3600
```
| Flag | Purpose | When to use |
|------|---------|-------------|
| `HttpOnly` | Not accessible via JavaScript (prevents XSS token theft) | Always |
| `Secure` | Only sent over HTTPS | Always |
| `SameSite=Strict` | Only sent on same-site requests (strongest CSRF) | Auth/session cookies |
| `SameSite=Lax` | Sent on top-level navigations (moderate CSRF) | Cookies that need cross-site top-level nav (e.g., OAuth return) |
| `Path=/` | Limit cookie scope | Always |
| `Max-Age` | Explicit expiration (prefer over `Expires`) | Always |
---
## Security Checklist
### Authentication and Sessions
- [ ] Passwords hashed with Argon2id or bcrypt (cost >= 12)
- [ ] JWT signed with RS256/ES256, algorithm enforced on verify
- [ ] Access tokens expire in <= 15 minutes
- [ ] Refresh tokens: one-time use, rotated, stored in httpOnly cookie
- [ ] Rate limiting on login, registration, and password reset
- [ ] Session regenerated after authentication
- [ ] MFA available for privileged accounts
### Authorization
- [ ] Every API endpoint has auth middleware
- [ ] Ownership checks on all resource access (prevent IDOR)
- [ ] Server-side authorization (frontend guards are UX only)
- [ ] Mass assignment prevented (explicit field selection)
- [ ] Re-authentication required for sensitive operations
### Input and Output
- [ ] All user input validated server-side (zod/joi/class-validator)
- [ ] Parameterized queries for all database operations
- [ ] HTML output sanitized (DOMPurify) when rendering user content
- [ ] Error responses do not expose stack traces in production
### Secrets
- [ ] No hardcoded secrets in source code
- [ ] `.env` files in `.gitignore`
- [ ] Server secrets not exposed to client (no NEXT_PUBLIC_ on secrets)
- [ ] Environment variables validated at startup
### Headers
- [ ] Content-Security-Policy configured (nonce-based preferred)
- [ ] Strict-Transport-Security with preload
- [ ] X-Content-Type-Options: nosniff
- [ ] X-Frame-Options: DENY
- [ ] Referrer-Policy: strict-origin-when-cross-origin
- [ ] Permissions-Policy restricting unused APIs
- [ ] CORS restricted to known origins
### Dependencies
- [ ] `npm audit` (or equivalent) passing in CI
- [ ] Lockfile committed and verified with `npm ci`
- [ ] New dependencies reviewed for typosquatting and postinstall scripts
- [ ] No wildcard or "latest" versions in production
### Logging
- [ ] Security events logged (auth failures, access denied, rate limits)
- [ ] No sensitive data in logs (passwords, tokens, PII)
- [ ] Structured logging with correlation IDs
- [ ] Alerts configured for anomalous patterns