security（安全）

Security

Security Checklist

Security Basics:
- [ ] Authentication required for protected routes
- [ ] Passwords hashed (bcrypt/argon2), never stored plain text
- [ ] API keys in environment variables, not code
- [ ] HTTPS only in production
- [ ] Input validated on server side
- [ ] SQL injection prevented (use parameterized queries)
- [ ] XSS prevented (sanitize user input)
- [ ] CSRF tokens on forms
- [ ] Rate limiting on API endpoints
- [ ] User sessions expire (30min-1hr typical)

See COMMON-VULNS.md for detailed checks.

Critical: Never Store These in Code

Move to environment variables:

• Database passwords
• API keys (Stripe, SendGrid, etc)
• JWT secrets
• OAuth client secrets
• Encryption keys

Tell AI:

Store API keys in .env file, not in code.
Add .env to .gitignore.
Access via process.env.API_KEY

Authentication Basics

Minimum requirements:

• Passwords: 8+ chars, require number/symbol
• Hash passwords (bcrypt with 10+ rounds)
• Email verification for signups
• Password reset via email only
• Sessions expire (30-60 min idle)
• Logout clears session completely

Tell AI:

Add authentication:
- bcrypt for password hashing (12 rounds)
- Email verification required
- Session timeout: 30 minutes
- Password requirements: 8+ chars, 1 number, 1 symbol

See SECURITY-PROMPTS.md for implementation details.

Data Protection

Always encrypt:

• Passwords (hashed, not encrypted)
• Payment info (use Stripe, don't store cards)
• Personal identifiable information (PII)

Never log:

• Passwords (even hashed)
• Credit card numbers
• API keys
• Session tokens

Tell AI:

Never log sensitive data.
Replace passwords/tokens with "[REDACTED]" in logs.

API Security

Required for all API endpoints:

• Authentication check
• Rate limiting (prevent abuse)
• Input validation
• Error messages don't leak info

Tell AI:

Add to all API routes:
- Require valid auth token
- Rate limit: 100 requests/minute per IP
- Validate all inputs (reject invalid)
- Generic error messages (no stack traces to users)

Common Vulnerabilities

Most common in AI-built apps:

1. Exposed API keys - In code instead of.env
2. No rate limiting - APIs can be spammed
3. Missing auth checks - Routes accessible without login
4. SQL injection - Raw SQL with user input
5. XSS attacks - Unescaped user content displayed

See COMMON-VULNS.md for how to check.

Security Prompts for AI

Adding authentication:

Add authentication to this route.
Require valid JWT token.
Return 401 if missing/invalid.
Don't expose error details.

Rate limiting:

Add rate limiting:
- 100 requests/minute per IP
- Return 429 "Too many requests" if exceeded
- Use sliding window, not fixed

Input validation:

Validate all user inputs:
- Email: valid format
- Password: 8+ chars, 1 number, 1 symbol
- Username: alphanumeric only, 3-20 chars
Reject invalid input with clear error message

See SECURITY-PROMPTS.md for more.

Pre-Launch Security Review

Before deploying:

Production Security:
- [ ] All secrets in environment variables
- [ ] HTTPS enforced (no HTTP)
- [ ] Database backups configured
- [ ] Rate limiting on all APIs
- [ ] Error pages don't show stack traces
- [ ] Admin routes protected
- [ ] File uploads validated (type, size)
- [ ] CORS configured (not wildcard "*")

When to Get Security Audit

Signs you need expert review:

• Handling payments directly (not Stripe)
• Storing health/financial data
• Multi-tenant with data isolation
• Over 1,000 users
• Processing sensitive PII

For most MVPs: Following this checklist is sufficient.

Common Founder Mistakes

Quick Wins

Easy security improvements:

1. Add Helmet.js (Node) - Sets security headers
2. Use HTTPS everywhere - Force in production
3. Add rate limiting - Prevents abuse
4. Environment variables - Keep secrets safe
5. Update dependencies - Fix known vulnerabilities

Tell AI:

Add helmet.js for security headers.
Configure for production (HTTPS, CSP, XSS protection).

Testing Security

Quick checks:

Exposed secrets:

grep -r "api_key" src/
grep -r "password" src/
# Should only find references to env vars

No auth bypass:

• Try accessing protected routes without login
• Should redirect to login or return 401

Rate limiting works:

• Hit API endpoint 100 times quickly
• Should get 429 error

Success Looks Like

✅ No secrets in code (all in.env) ✅ Can't access protected routes without auth ✅ Passwords hashed, never stored plain text ✅ Rate limiting prevents abuse ✅ HTTPS enforced in production ✅ Input validated on server side