Security Audit Skill

Detect common security vulnerabilities during code review and development. Based on OWASP guidelines and common vulnerability patterns.

Design Principle

This skill is framework-generic. It provides universal security patterns:

• Covers OWASP Top 10 and common CWEs
• Works with Python, TypeScript, and other languages
• Project-specific security requirements go in project-specific skills

Variables

| Variable | Default | Description | |----------|---------|-------------| | SEVERITY_THRESHOLD | medium | Minimum severity to report | | SCAN_DEPTH | 3 | Directory depth for scanning | | INCLUDE_TESTS | false | Include test files in scan |

Instructions

MANDATORY - Follow the Workflow steps below in order.

1. Identify security-sensitive code areas
2. Check for common vulnerability patterns
3. Report findings with severity
4. Suggest remediation

Red Flags - STOP and Reconsider

If you're about to:

• Write SQL with string concatenation
• Use bare except: blocks
• Execute shell commands with user input
• Store secrets in code
• Disable security features "temporarily"

STOP -> Use parameterized queries -> Add specific exception handling -> Then proceed

Cookbook

SQL Injection Prevention

• IF: Writing database queries
• THEN: Read and execute ./cookbook/sql-injection.md

Bare Except Handling

• IF: Writing exception handlers
• THEN: Read and execute ./cookbook/bare-except.md

Shell Injection Prevention

• IF: Executing shell commands
• THEN: Read and execute ./cookbook/shell-injection.md

Vulnerability Patterns

SQL Injection (CWE-89)

BAD - String concatenation:

python
# VULNERABLE
query = f"SELECT * FROM users WHERE id = {user_id}"
cursor.execute(query)

query = "SELECT * FROM users WHERE name = '" + name + "'"

GOOD - Parameterized queries:

python
# SAFE
cursor.execute("SELECT * FROM users WHERE id = %s", (user_id,))

# SQLAlchemy
session.query(User).filter(User.id == user_id).first()

# Prisma
await prisma.user.findUnique({ where: { id: userId } })

Bare Except (CWE-754)

BAD - Catches everything:

python
# VULNERABLE - hides bugs, catches KeyboardInterrupt
try:
    risky_operation()
except:
    pass

# VULNERABLE - too broad
except Exception:
    log.error("Something failed")

GOOD - Specific exceptions:

python
# SAFE - specific exceptions
try:
    risky_operation()
except ValueError as e:
    log.warning(f"Invalid value: {e}")
except ConnectionError as e:
    log.error(f"Connection failed: {e}")
    raise

Shell Injection (CWE-78)

BAD - User input in shell:

python
# VULNERABLE
os.system(f"grep {user_input} /var/log/app.log")

import subprocess
subprocess.run(f"ls {directory}", shell=True)

GOOD - Avoid shell, use lists:

python
# SAFE - no shell
subprocess.run(["grep", user_input, "/var/log/app.log"])

# SAFE - validated input
if not re.match(r'^[a-zA-Z0-9_-]+$', directory):
    raise ValueError("Invalid directory name")
subprocess.run(["ls", directory])

Path Traversal (CWE-22)

BAD - User input in paths:

python
# VULNERABLE
path = f"/uploads/{user_filename}"
with open(path) as f:
    return f.read()

GOOD - Validate and sanitize:

python
# SAFE
from pathlib import Path

upload_dir = Path("/uploads").resolve()
requested = (upload_dir / user_filename).resolve()

if not requested.is_relative_to(upload_dir):
    raise ValueError("Path traversal attempt")

with open(requested) as f:
    return f.read()

Hardcoded Secrets (CWE-798)

BAD - Secrets in code:

python
# VULNERABLE
API_KEY = "sk-1234567890abcdef"
DB_PASSWORD = "super_secret_password"

GOOD - Environment variables:

python
# SAFE
import os

API_KEY = os.environ["API_KEY"]
DB_PASSWORD = os.environ["DB_PASSWORD"]

# Or with defaults for development
API_KEY = os.getenv("API_KEY", "dev-key-only")

XSS (CWE-79)

BAD - Unsanitized output:

html
<!-- VULNERABLE -->
<div>{{ user_input }}</div>

GOOD - Proper escaping:

html
<!-- SAFE - auto-escaped in most frameworks -->
<div>{{ user_input | e }}</div>

<!-- Or use textContent in JS -->
element.textContent = userInput;  // Safe

Severity Levels

| Severity | Impact | Examples | |----------|--------|----------| | CRITICAL | Data breach, RCE | SQL injection, shell injection | | HIGH | Data exposure, privilege escalation | Path traversal, hardcoded secrets | | MEDIUM | Information disclosure | Verbose errors, bare excepts | | LOW | Best practice violation | Missing input validation |

Detection Patterns

Python Patterns to Search

python
VULNERABLE_PATTERNS = {
    "sql_injection": [
        r'execute\([\'"].*%s.*[\'"].*%',  # % formatting in SQL
        r'execute\(f[\'"]',                 # f-string in SQL
        r'execute\([\'"].*\+',              # String concat in SQL
    ],
    "shell_injection": [
        r'os\.system\(',                    # os.system
        r'subprocess\..*shell=True',        # shell=True
        r'eval\(',                          # eval
        r'exec\(',                          # exec
    ],
    "bare_except": [
        r'except\s*:',                      # bare except
    ],
    "hardcoded_secrets": [
        r'password\s*=\s*[\'"]',            # password = "..."
        r'api_key\s*=\s*[\'"]',             # api_key = "..."
        r'secret\s*=\s*[\'"]',              # secret = "..."
    ],
}

TypeScript Patterns to Search

typescript
const VULNERABLE_PATTERNS = {
  sqlInjection: [
    /`SELECT.*\$\{/,        // Template literal in SQL
    /"SELECT.*" \+ /,       // String concat in SQL
  ],
  xss: [
    /innerHTML\s*=/,        // innerHTML assignment
    /dangerouslySetInnerHTML/, // React dangerous prop
  ],
  shellInjection: [
    /exec\([`'"]/,          // child_process.exec
    /spawn\(.*shell:\s*true/, // shell: true
  ],
};

Audit Workflow

Step 1: Identify Sensitive Areas

markdown
Check these high-risk areas first:
- Authentication/authorization code
- Database queries
- File operations
- External API calls
- User input handling
- Serialization/deserialization

Step 2: Run Pattern Scan

markdown
For each source file:
  Match against vulnerability patterns
  Record file, line, pattern matched
  Assess severity

Step 3: Generate Report

markdown
# Security Audit Report

## Summary
- CRITICAL: 2
- HIGH: 5
- MEDIUM: 12

## Critical Issues

### 1. SQL Injection in user_service.py:45
Pattern: f-string in execute()

cursor.execute(f"SELECT FROM users WHERE id = {user_id}")

**Fix**: Use parameterized query

cursor.execute("SELECT FROM users WHERE id = %s", (user_id,))

Integration

With /ai-dev-kit:execute-lane

Run security audit in code-related lanes:

markdown
Lane: SL-API

Post-implementation checks:
1. ✓ Tests pass
2. ✓ Lint clean
3. ⚠️ Security audit: 2 MEDIUM issues

Review security findings before merge.

CI Integration

yaml
- name: Security Audit
  run: |
    # Check for vulnerable patterns
    grep -rn "execute(f" --include="*.py" && exit 1 || true
    grep -rn "shell=True" --include="*.py" && exit 1 || true
    grep -rn "except:" --include="*.py" && echo "Warning: bare except found"

Best Practices

1. Defense in depth: Multiple layers of security
2. Least privilege: Minimum permissions needed
3. Input validation: Validate all external input
4. Output encoding: Escape output appropriately
5. Secrets management: Never hardcode secrets
6. Error handling: Don't expose internal details
7. Logging: Log security events (not sensitive data)
8. Updates: Keep dependencies updated