Security audit

Secure Script Runner (Zero‑Exposure Sandbox)

Security checks across malware telemetry and agentic risk

Overview

The skill is not malicious, but it enables local script execution and credential access while making some overbroad confidentiality claims users should review carefully.

Install only if you intend to let agents help manage and run local MGC Blackbox scripts. Require explicit approval before every save, run, seal, or credential-related action, and do not treat the zero-exposure claim as universal: scripts passed through agent tool calls may still be visible to the agent before storage.

SkillSpector

Vulnerability Patterns

MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (2)

Intent-Code Divergence

Medium

Confidence: 91% confidence
Finding: The skill claims it only provides documentation and does not execute automatically, yet it directly instructs an AI agent to use mgc_run for script execution. This mismatch can mislead operators and downstream policy systems into treating the skill as passive documentation when it is actually an execution-enabling workflow for opaque code.

Intent-Code Divergence

Medium

Confidence: 87% confidence
Finding: The document asserts that AI never sees script plaintext or stdout, but the included examples embed script plaintext directly in the skill content. This undermines the zero-exposure claim and can create a false sense of confidentiality around content that is in fact visible to the model and anyone inspecting the skill.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.