ClawHub
Back to skill

Security audit

LYGO Automation Workflows

Security checks for vulnerabilities and agentic risk

Overview

This skill is a local, consent-oriented automation planning guide and CLI, with no evidence of hidden network, account, publishing, or background behavior.

This looks safe to install as a local planning aid. Treat any separately advertised full zip or external automation implementation as outside this review, and review generated plans carefully before connecting real vendors, CRMs, payment systems, or tools that handle personal data.

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 (1)

Intent-Code Divergence

Medium
Confidence
93% confidence
Finding
The module docstring asserts "no network/subprocess," but the self-check only detects subprocess and a narrow set of os execution calls. It does not inspect imports or call sites for network-capable modules such as socket, urllib, requests, httpx, or similar, so the check can give a misleading assurance that network behavior is absent when it has not actually been verified.

Static analysis

No suspicious patterns detected.