Klemenska Security Auditor

Scan and audit installed skills for security risks, suspicious patterns, and permission overreach. Use when: (1) before installing a new skill; (2) periodically reviewing installed skills; (3) before granting skill permissions; (4) when explicitly asked to audit skills or check for vulnerabilities.

Security Auditor

Audit skills for security risks before installing or using them.

Purpose

Skills can request permissions and access that may be:

• Overreaching (accessing data they shouldn't need)
• Suspicious (behaving oddly, phoning home, etc.)
• Outdated (known vulnerabilities in dependencies)

This skill helps you audit them.

When to Run

Trigger	Action
Before installing a new skill	Full audit
Periodic review	Quick scan of installed skills
Suspicious behavior	Deep analysis
Permission review	Check requested permissions

Audit Workflow

Step 1: Quick Scan

bash

python3 scripts/audit.py --scan

Checks:

• File access patterns
• Network access requests
• Suspicious API usage
• Permission requests

Step 2: Detailed Audit

bash

python3 scripts/audit.py --audit <skill-path>

Performs deep analysis:

• Code pattern analysis
• Dependency checking
• Permission mapping
• Risk scoring

Step 3: Generate Report

bash

python3 scripts/audit.py --report <skill-path> --output report.md

Creates detailed security report.

Step 4: Compare Skills

bash

python3 scripts/audit.py --compare <skill1-path> <skill2-path>

Compare security posture of two skills.

Risk Levels

Level	Meaning	Action
🟢 LOW	Minimal risk, standard permissions	Safe to install
🟡 MEDIUM	Some overreach, review recommended	Read code before install
🔴 HIGH	Significant risks, careful review required	Do not install without review
⛔ CRITICAL	Dangerous patterns detected	Do not install

Red Flags to Watch For

File Access

• Accessing ~/.ssh/ or ~/.aws/
• Reading *password*, *secret*, *key* files
• Writing to system directories
• Accessing other users' directories

Network

• Exfiltrating data to unknown servers
• DNS rebinding patterns
• Encrypted payloads to unfamiliar domains

Permissions

• Requesting exec with no scope limitation
• Reading memory or process info
• Keylogging or screenshot capabilities
• Accessing other installed skills' data

Code Patterns

• Obfuscated code
• Dynamic code generation
• Shell commands without sanitization
• Credential harvesting patterns

Files

• scripts/audit.py — Main audit script
• scripts/scan_skill.py — Skill-specific scanner
• references/rules.md — Security rules and patterns
• references/permissions.md — Permission reference guide