ClawHub

Emblem Ai Agent Wallet

Security checks across malware telemetry and agentic risk

Overview

This is a disclosed crypto wallet integration, but users should review it carefully because agent-mode transaction commands are high-impact and the approval boundary is not fully demonstrated in the submitted artifact.

Install only if you trust EmblemVault and the npm package. Keep limited funds in wallets used through agents, protect EMBLEM_PASSWORD and ~/.emblemai, prefer browser auth for interactive use, and do not allow unattended agent-mode transfers or swaps unless you independently verify that every transaction requires explicit human approval.

SkillSpector

By NVIDIA
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

High
Confidence
95% confidence
Finding
The documentation presents conflicting safety guarantees: it says wallet-modifying actions always require explicit confirmation, yet elsewhere advertises single-shot agent automation for sends, swaps, and other financial actions. In a wallet/trading skill, this ambiguity is dangerous because integrators may assume human approval gates exist when the CLI or downstream agent can execute state-changing actions autonomously.

Intent-Code Divergence

High
Confidence
97% confidence
Finding
The skill claims safe mode requires explicit user confirmation for all wallet-modifying actions, but it also instructs agents to use one-shot commands like sending SOL or swapping assets. For a crypto wallet skill, this mismatch can directly lead to unauthorized transactions if host agents rely on the docs rather than an enforced approval boundary.

VirusTotal

60/60 vendors flagged this skill as clean.

View on VirusTotal