/security - Security Audit Workflow

Dedicated security analysis for sensitive code.

When to Use

• "Security audit"
• "Check for vulnerabilities"
• "Is this secure?"
• "Review authentication code"
• "Check for injection attacks"
• Before handling auth, payments, user data
• After adding security-sensitive features

Workflow Overview

┌─────────┐    ┌───────────┐
│  aegis  │───▶│ arbiter  │
│         │    │           │
└─────────┘    └───────────┘
  Security       Verify
  audit          fixes

Agent Sequence

#	Agent	Role	Output
1	aegis	Comprehensive security scan	Vulnerability report
2	arbiter	Verify fixes, run security tests	Verification report

Why Dedicated Security?

The /review workflow focuses on code quality. Security needs:

• Specialized vulnerability patterns
• Dependency scanning
• Secret detection
• OWASP Top 10 checks
• Authentication/authorization review

Execution

Phase 1: Security Audit

Task(
  subagent_type="aegis",
  prompt="""
  Security audit: [SCOPE]

  Scan for:

  **Injection Attacks:**
  - SQL injection
  - Command injection
  - XSS (Cross-Site Scripting)
  - LDAP injection

  **Authentication/Authorization:**
  - Broken authentication
  - Session management issues
  - Privilege escalation
  - Insecure direct object references

  **Data Protection:**
  - Sensitive data exposure
  - Hardcoded secrets/credentials
  - Insecure cryptography
  - Missing encryption

  **Configuration:**
  - Security misconfigurations
  - Default credentials
  - Verbose error messages
  - Missing security headers

  **Dependencies:**
  - Known vulnerable packages
  - Outdated dependencies
  - Supply chain risks

  Output: Detailed report with:
  - Severity (CRITICAL/HIGH/MEDIUM/LOW)
  - Location (file:line)
  - Description
  - Remediation steps
  """
)

Phase 2: Verification (After Fixes)

Task(
  subagent_type="arbiter",
  prompt="""
  Verify security fixes: [SCOPE]

  Run:
  - Security-focused tests
  - Dependency audit (npm audit, pip audit)
  - Re-check reported vulnerabilities
  - Verify fixes don't introduce regressions

  Output: Verification report
  """
)

Security Scopes

Full Codebase

User: /security
→ Scan entire codebase

Specific Area

User: /security authentication
→ Focus on auth-related code

Single File

User: /security src/api/auth.py
→ Deep dive on one file

Dependencies Only

User: /security --deps
→ Only dependency vulnerabilities

Example

User: /security the payment processing code

Claude: Starting /security audit for payment code...

Phase 1: Security audit...
[Spawns aegis]

┌─────────────────────────────────────────────────────────────┐
│ Security Audit Report                                       │
├─────────────────────────────────────────────────────────────┤
│ Scope: src/services/payment/                                │
│ Files scanned: 12                                           │
│ Lines analyzed: 2,847                                       │
├─────────────────────────────────────────────────────────────┤
│ CRITICAL (1)                                                │
│ ──────────                                                  │
│ [C1] SQL Injection in payment.py:89                         │
│      query = f"SELECT * FROM orders WHERE id = {order_id}"  │
│      Fix: Use parameterized queries                         │
│                                                             │
│ HIGH (2)                                                    │
│ ────────                                                    │
│ [H1] Hardcoded API key in stripe_client.py:12               │
│      STRIPE_KEY = "sk_live_..."                             │
│      Fix: Move to environment variable                      │
│                                                             │
│ [H2] Missing input validation in refund.py:45               │
│      amount = request.json['amount']  # No validation       │
│      Fix: Validate amount is positive number                │
│                                                             │
│ MEDIUM (1)                                                  │
│ ──────────                                                  │
│ [M1] Verbose error messages in error_handler.py:23          │
│      return {"error": str(e)}  # Leaks internal details     │
│      Fix: Return generic message, log details internally    │
│                                                             │
│ LOW (0)                                                     │
│ ───────                                                     │
│ None                                                        │
├─────────────────────────────────────────────────────────────┤
│ Summary: 1 critical, 2 high, 1 medium, 0 low                │
│ Status: BLOCKING - Fix critical issues before release       │
└─────────────────────────────────────────────────────────────┘

Fix the critical and high issues, then run:
/security --verify

After Fixes

User: /security --verify

Claude: Phase 2: Verifying fixes...
[Spawns arbiter]

✅ C1: SQL injection fixed - using parameterized queries
✅ H1: API key moved to environment variable
✅ H2: Input validation added
✅ M1: Error messages sanitized

All security tests passing.
Security audit: PASSED

OWASP Top 10 Coverage

Risk	Checked
A01 Broken Access Control	✅
A02 Cryptographic Failures	✅
A03 Injection	✅
A04 Insecure Design	✅
A05 Security Misconfiguration	✅
A06 Vulnerable Components	✅
A07 Auth Failures	✅
A08 Data Integrity Failures	✅
A09 Logging Failures	✅
A10 SSRF	✅

Flags

• --deps: Dependencies only
• --verify: Re-run after fixes
• --owasp: Explicit OWASP Top 10 report
• --secrets: Focus on secret detection