/security-audit

Multi-agent security audit with findings saved to timestamped report.

Usage

/security-audit yourbench           # Full security review
/security-audit coordinatr          # Audit specific project

Audit Dimensions

Five security-auditor agents run in parallel:

Agent	Focus Area	Checks
Agent 1: Auth & Access	Authentication, Authorization	JWT handling, session management, RBAC, privilege escalation
Agent 2: Input & Data	Injection, Validation	SQL injection, XSS, command injection, input sanitization
Agent 3: Crypto & Secrets	Cryptography, Secrets	Hardcoded credentials, weak crypto, key management, PII
Agent 4: Config & Deploy	Configuration, Infrastructure	CORS, CSRF, security headers, exposed endpoints, debug mode
Agent 5: Dependencies	Supply Chain, Libraries	Vulnerable packages, outdated deps, license issues

OWASP Top 10 Coverage

OWASP Risk	Coverage
A01 Broken Access Control	Agent 1
A02 Cryptographic Failures	Agent 3
A03 Injection	Agent 2
A04 Insecure Design	Agents 1, 4
A05 Security Misconfiguration	Agent 4
A06 Vulnerable Components	Agent 5
A07 Auth Failures	Agent 1
A08 Data Integrity Failures	Agents 2, 3
A09 Logging Failures	Agent 4
A10 SSRF	Agent 2

Execution Flow

1. Validate Project

ls spaces/[project]/

2. Launch Parallel Audits

5 security-auditor agents run concurrently with focused prompts.

3. Consolidate Findings

Aggregate by:

• Severity: Critical, High, Medium, Low, Info
• Category: OWASP classification
• Location: File path + line number
• Remediation: Specific fix guidance

4. Generate Report

Write: .claude/temp/security-audit-[project]-[timestamp].md

Report Structure

# Security Audit: [Project Name]
**Date**: YYYY-MM-DD HH:MM:SS

## Executive Summary
- Critical issues: X
- High severity: Y
- Total findings: Z

## Critical Issues
### [Issue Title]
- **Severity**: Critical
- **Category**: SQL Injection (CWE-89)
- **Location**: src/api/users.py:42
- **Description**: [What's wrong]
- **Impact**: [What could happen]
- **Remediation**: [How to fix]

## High Severity Issues
[...]

## Recommendations
- Priority actions
- Long-term improvements

## Scan Coverage
- Files scanned: X
- Technologies: Z

When to Use

• Before production deployments
• After major feature additions
• Monthly security reviews
• Before external security audits
• After dependency updates

Output Location

.claude/temp/security-audit-yourbench-2026-01-08-143022.md

Reports saved to

.claude/temp/

(gitignored) for review.

Notes

• Read-only: No code changes made
• Non-blocking: Doesn't prevent commits
• Parallel execution: Agents run concurrently
• False positives possible: Manual review recommended

Integration

Implement security feature → /security-audit → Fix issues → /commit