Multi-Agent AI Setup Guide

• Outcome: Review
• Latest audit: May 29, 2026 at 9:38 PM UTC
• Version:
  1.0.1

Overview

This documentation-only setup guide is coherent and not malicious, but it merits Review because it broadly advises disabling sandboxing while setting up persistent multi-agent memory and WhatsApp console mirroring.

Install only if you intentionally want this kind of powerful multi-agent setup. Keep sandboxing enabled unless you have a specific external-volume problem, use a dedicated memory vault with no secrets, review any related skills before installing them, and enable WhatsApp mirroring only for channels where logs and prompts are safe to forward.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (2)

Missing User Warnings

Medium

• Confidence: 94% confidence
• Finding: The skill recommends a WhatsApp bidirectional mirroring setup without an explicit privacy, retention, or data-sharing warning. This can expose sensitive console content, operational logs, prompts, secrets, or personal data to a third-party messaging service and to any device/account linked to that WhatsApp session.

Natural-Language Policy Violations

Medium

• Confidence: 89% confidence
• Finding: The guide gives a blanket instruction to never enable sandbox mode on macOS with an external drive, which encourages disabling an important isolation control without requiring a risk assessment or safer alternatives. In a setup guide for multi-agent systems that can access shared memory and external tools, normalizing sandbox disablement increases the blast radius of prompt injection, tool misuse, or agent error.

VirusTotal

62/62 vendors flagged this skill as clean.