ClawHub

Agent Team

Security checks across malware telemetry and agentic risk

Overview

This skill is a plausible multi-agent manager, but it should be reviewed because it can spawn ongoing child-agent sessions and includes an under-documented helper that calls an external model API with confusing credential handling.

Review before installing if you do not want delegated child-agent sessions or local persona files sent to model providers. Use only trusted SOUL.md/config.json agent profiles, monitor and stop spawned sessions as needed, and do not place real API keys directly into greetings.py.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Intent-Code Divergence

High
Confidence
98% confidence
Finding
The file claims API keys are read securely from environment variables, but the actual request path uses per-agent api_key fields embedded in configuration objects. This creates a risky pattern where developers may hardcode secrets in source control or believe environment-based protection is in effect when it is not, leading to credential exposure and misuse of paid model APIs.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
Interactive chat sends both user input and the loaded SOUL.md persona content to an external model process without any explicit consent, warning, or data-minimization step. In this context, SOUL.md and user prompts may contain private workspace data or sensitive instructions, so silent forwarding creates a real confidentiality and governance risk.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal