Security Audit Skill

Purpose

To act as an automated SAST (Static Application Security Testing) scanner. Goal: Identify vulnerabilities in the implementation phase, before the code reaches the review phase.

ROI Metric: Detecting a hardcoded secret here takes 5 seconds. Rotating a compromised key in production takes 5 hours.

When to Use

• •Trigger: When backend implements an endpoint involving users, auth, or payments.
• •Trigger: When frontend uses dangerouslySetInnerHTML or similar raw HTML rendering.
• •Trigger: When pm-agent flags a task as "High Risk."
• •Agent: Primary skill for the security agent; secondary skill for backend/review modes.

The Audit Protocol (The Scanner)

1. Pattern Scanning (The Grep Check)

Scan the code for these specific "smells":

2. Logic Analysis (STRIDE)

Ask these questions about the flow:

• •Spoofing: Does this verify who the user is? (Check: Auth Middleware presence).
• •Tampering: Can I modify the payload? (Check: Zod/Joi validation strip() on unknown keys).
• •Information Disclosure: Does the error return the stack trace? (Check: try/catch blocks).
• •Elevation of Privilege: Does it check user.id === resource.ownerId? (Check: Authorization logic).

3. Dependency Check (Supply Chain)

• •If package.json was modified:
  • •Are versions pinned? (Good: 1.2.3, Bad: ^1.2.3).
  • •Are there known CVEs? (Trigger tavily search).

Execution Template

Use this scratchpad to document the audit findings.

## 🔐 Security Audit Report: [File/Component]

### 1. Static Scan
- [ ] Checked for hardcoded secrets.
- [ ] Checked for raw SQL/HTML injection.
- [ ] Checked for dangerous functions (`eval`, `exec`).

### 2. Logic Check (OWASP)
- **AuthZ**: [ Verify BOLA/IDOR protection ]
- **Input**: [ Verify Schema Validation ]
- **Output**: [ Verify Data Sanitization ]

### 3. Vulnerability Findings
| Severity | Line | Issue | CWE ID | Remediation |
| :--- | :--- | :--- | :--- | :--- |
| 🔴 **Critical** | 42 | SQL Injection (`${id}`) | CWE-89 | Use Param: `WHERE id = $1` |
| 🟡 **Medium** | 15 | Missing Rate Limit | CWE-799 | Add `RateLimitMiddleware` |

### 🏁 Verdict
[ PASS / FAIL ] - *If Fail, do not commit.*

Remediation Patterns (The "Fix It" Guide)

Problem: SQL Injection ❌ db.query("SELECT * FROM users WHERE name = '" + name + "'") ✅ db.query("SELECT * FROM users WHERE name = $1", [name])

Problem: XSS (React) ❌ <div dangerouslySetInnerHTML={{ __html: userInput }} /> ✅ <div>{userInput}</div> or use DOMPurify.sanitize(userInput)

Problem: Broken Auth (IDOR) ❌ return await db.orders.find(req.params.id) ✅ return await db.orders.find({ id: req.params.id, ownerId: req.user.id })

Integration with Agents

• •security agent: Uses this skill as its primary method of interaction.
• •backend agent: Must invoke security-audit before marking an API task as "Complete."
• •review mode: Uses the Findings Table as the basis for "Critical" comments.