secure-coding-cybersecurity

Secure Coding & Cybersecurity Skill

Vision and Purpose

This skill establishes security as a first-class citizen in code generation and review. It recognizes that AI-generated code often prioritizes functionality over security, inheriting and propagating vulnerabilities from training data. This skill provides comprehensive guardrails, checklists, and verification steps to ensure all code—especially AI-generated—is robust, secure, and follows industry best practices.

Critical Understanding: Why AI Code is Often Insecure

AI models optimize for "working code" and "correct-looking output" before "secure and robust code." This introduces classic vulnerabilities at concerning rates:

Root Causes of AI Security Failures

1. Functionality over Security: If the prompt doesn't explicitly demand security, the model prioritizes the shortest solution even if it uses eval(), SQL string concatenation, or hardcoded keys.

2. Learning from Insecure Examples: A significant portion of publicly available code contains bad practices (hardcoded secrets, weak encryption, missing validation), and models reproduce these patterns.

3. Lack of Context Awareness: AI doesn't understand your business rules, threat model, or compliance requirements (PCI-DSS, HIPAA, GDPR), filling gaps with dangerous assumptions.

4. High CWE Rates in Studies: Empirical research consistently finds serious vulnerabilities (SQLi, XSS, buffer overflows, crypto misuse, hardcoded credentials) in AI-generated code samples.

The Security Non-Negotiables

These rules must NEVER be violated, regardless of convenience or "example purposes":

1. Never Prioritize Simplicity Over Security: Do not provide "clean" examples that omit input validation or use insecure defaults, even with disclaimers.

2. Treat AI Output as Untrusted: Always audit AI-generated snippets for hallucinated libraries, outdated patterns, and missing security controls.

3. Fail Closed: All logic must default to "Access Denied" if an exception occurs, validation fails, or state is unclear.

4. No Hardcoded Secrets: Never suggest code with hardcoded API keys, passwords, tokens, or cryptographic keys. Always use environment variables, secret managers, or secure vaults.

5. Parameterized Everything: Never use string concatenation or formatting for SQL queries, OS commands, HTML rendering, or LDAP filters.

6. Validate All Inputs: Every piece of external data must be validated and sanitized before use.

7. Defense in Depth: Never rely on a single security control; implement multiple layers of protection.

Comprehensive Security Checklist

1. Input Validation and Injection Prevention

The Problem: AI frequently omits input validation and sanitization unless explicitly requested, leading to CWE-20 (Improper Input Validation) and the entire injection vulnerability family (SQLi, XSS, OS Command Injection, LDAP Injection).

Research Finding: Multiple studies identify SQL injection, XSS, and OS command injection as recurrent vulnerabilities in LLM-generated code.

Checklist

• Validate and normalize ALL user input before use, including:

  • Query parameters (URL parameters)
  • Request body (JSON, form data)
  • Headers (including custom headers)
  • Path parameters
  • File uploads
  • WebSocket messages
  • GraphQL inputs

• Use allowlists over denylists: Define what is permitted rather than trying to block what is malicious. Use strict type checking, enum validation, and schema validation.

• Parameterized queries ONLY: Never concatenate strings to build SQL queries, shell commands, LDAP filters, XPath expressions, or NoSQL queries. Always use:

  • Prepared statements with bound parameters
  • ORM query builders with parameterized methods
  • Safe API abstractions that prevent injection

• XSS Prevention: Escape or sanitize all data before rendering in HTML contexts:

  • Use auto-escaping template engines
  • Apply context-appropriate encoding (HTML, JavaScript, CSS, URL)
  • Avoid innerHTML, document.write, and similar dangerous APIs
  • Implement Content Security Policy (CSP) headers

• Disable dangerous evaluation: Never use eval(), Function(), exec(), child_process.exec(), Runtime.exec(), os.system(), or similar on user-controlled data. If dynamic execution is absolutely necessary, use strict sandboxing and allowlists.

• Strict regex validation: When using regular expressions for validation:

  • Prefer exact match patterns (^pattern$) over partial matches
  • Avoid overly permissive patterns
  • Be aware of ReDoS (Regular Expression Denial of Service) vulnerabilities

Verification Steps

Before considering input handling complete:

1. Fuzz Test: Test with unexpected inputs (null, empty strings, very long strings, special characters, Unicode, binary data)
2. Injection Test: Attempt SQL, NoSQL, command, and XSS injection payloads
3. Boundary Test: Test at and beyond length limits, type boundaries, and range limits
4. Negative Test: Ensure invalid inputs are rejected with appropriate errors

Common AI Mistakes to Avoid

# ❌ NEVER DO THIS - SQL Injection vulnerability
query = f"SELECT * FROM users WHERE id = {user_id}"
cursor.execute(query)

# ✅ DO THIS INSTEAD - Parameterized query
query = "SELECT * FROM users WHERE id = ?"
cursor.execute(query, (user_id,))

# ❌ NEVER DO THIS - XSS vulnerability
element.innerHTML = userInput

// ❌ NEVER DO THIS - Command Injection
const output = exec(`ls ${userInput}`);

// ✅ DO THIS INSTEAD - Safe command execution with allowlist
const allowedCommands = ['list', 'status'];
if (allowedCommands.includes(userInput)) {
  const output = execFile('ls', [safePath]);
}

---

2. Authentication, Authorization, and Session Management

The Problem: AI generates authentication flows that "pass happy path tests" but ignore critical security details like rate limiting, constant-time comparison, or fine-grained authorization controls. This enables brute-force attacks, privilege escalation, and API abuse.

Checklist

• Secure password storage:

  • Use modern, memory-hard algorithms: Argon2id (recommended), bcrypt, or scrypt
  • Never use MD5, SHA1, SHA256 for password hashing (they're too fast)
  • Always use unique salts per password
  • Configure appropriate cost factors/work factors

• Constant-time comparisons: Compare tokens, passwords, and API keys using constant-time comparison functions to prevent timing attacks:

  • Python: hmac.compare_digest()
  • Node.js: crypto.timingSafeEqual()
  • Java: MessageDigest.isEqual()

• Rate limiting and brute-force protection:

  • Implement rate limiting on login endpoints, password reset, OTP verification, and sensitive APIs
  • Use progressive delays for failed attempts
  • Consider CAPTCHA after multiple failures
  • Implement account lockout policies (with unlock mechanisms)

• Authorization on every endpoint:

  • Verify ownership and permissions on every request, not just authentication
  • Check for IDOR (Insecure Direct Object Reference) vulnerabilities
  • Validate that users can only access their own resources
  • Implement attribute-based access control (ABAC) where appropriate

• Secure session management:

  • Use cryptographically secure random session IDs
  • Implement reasonable session timeouts
  • Support session revocation and rotation
  • Regenerate session IDs on privilege changes (login, password change, role change)

• Secure cookie configuration:

  • Set HttpOnly flag (prevents JavaScript access)
  • Set Secure flag (HTTPS only)
  • Set SameSite attribute (Strict or Lax)
  • Use appropriate Max-Age or Expires
  • Consider __Host- prefix for additional protection

• JWT security:

  • Use strong signing algorithms (RS256, ES256, HS256 with strong secrets)
  • Never use "none" algorithm or weak secrets
  • Set short expiration times
  • Include token rotation and revocation mechanisms
  • Store tokens securely (not in localStorage for sensitive apps)

• Password reset security:

  • Use cryptographically secure random tokens with high entropy
  • Implement single-use tokens with short expiration
  • Do not reveal whether an email exists in the system (privacy protection)
  • Invalidate existing tokens when new ones are requested
  • Require re-authentication after password reset

• Multi-factor authentication (MFA):

  • Support TOTP (Time-based One-Time Password)
  • Support WebAuthn/FIDO2 for strong authentication
  • Enforce MFA for privileged accounts
  • Implement backup codes securely

Verification Steps

1. Brute Force Test: Attempt to guess passwords, session IDs, and tokens
2. Timing Attack Test: Measure response times for different inputs
3. Authorization Test: Try to access other users' resources
4. Session Fixation Test: Verify session ID changes on login
5. Token Analysis: Check JWT headers and payloads for security issues

Common AI Mistakes to Avoid

# ❌ NEVER DO THIS - Insecure password hashing
hashed = hashlib.md5(password.encode()).hexdigest()

# ✅ DO THIS INSTEAD - Secure password hashing
hashed = bcrypt.hashpw(password.encode(), bcrypt.gensalt(rounds=12))

# ❌ NEVER DO THIS - Timing attack vulnerability
if token == stored_token:
    # authenticate

# ✅ DO THIS INSTEAD - Constant-time comparison
if hmac.compare_digest(token, stored_token):
    # authenticate

# ❌ NEVER DO THIS - No rate limiting
@app.route('/login', methods=['POST'])
def login():
    # authenticate

# ✅ DO THIS INSTEAD - Rate limiting
@limiter.limit("5 per minute")
@app.route('/login', methods=['POST'])
def login():
    # authenticate

# ❌ NEVER DO THIS - Missing authorization check
@app.route('/api/documents/<doc_id>')
def get_document(doc_id):
    return Document.query.get(doc_id)  # No ownership check!

# ✅ DO THI

---

*Content truncated.*