ClawHub
Back to skill

Security audit

Multi-Agent Collaboration Communication

Security checks across malware telemetry and agentic risk

Overview

This skill is a coherent multi-agent system design guide with local helper scripts and no evidence of hidden, destructive, or exfiltrating behavior.

Reasonable to install for multi-agent design work. Before running the scripts, review input and output paths, avoid putting real secrets in placeholder auth fields, and treat the broad capability tags as over-inclusive rather than as behavior shown in the artifacts.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • 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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (1)

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The example message schema and output explicitly include identifiable author metadata such as an email address and commit attribution, but provide no privacy notice, minimization guidance, or retention/access controls. In a multi-agent system, this increases unnecessary propagation of personal data across components and logs, raising privacy, compliance, and insider-exposure risks if the data is stored, forwarded, or leaked.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.