Security audit

stablecoin-depeg-scanner

Security checks across malware telemetry and agentic risk

Overview

This is a disclosed crypto depeg analysis skill that gives risky trading-style suggestions, but it does not access wallets, credentials, or execute trades.

Use this only as a research aid for speculative stablecoin depeg analysis. Verify exploit facts, collateral status, liquidity, and team statements independently before acting, treat all BUY/WATCH/AVOID output as non-authoritative, and provide an approximate capital range rather than exact portfolio details.

SkillSpector

By NVIDIA

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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

Findings (5)

Description-Behavior Mismatch

Medium

Confidence: 98% confidence
Finding: The skill does not merely analyze a depeg event; it outputs direct trading advice such as BUY/WATCH/AVOID, suggested position sizing, profit/loss projections, and execution steps. In this context, that is dangerous because the recommendation engine relies on incomplete signals and explicitly leaves exploit verification to manual follow-up, yet still produces actionable instructions that could drive risky financial behavior.

Intent-Code Divergence

Low

Confidence: 88% confidence
Finding: The docstring promises an 'Automated 3-minute assessment,' but the code later states that exploit analysis must be manually verified by the agent. This mismatch is risky because users or downstream agents may overtrust the automation and act before the most important security facts—such as whether collateral was stolen or the protocol is unrecoverable—are confirmed.

Missing User Warnings

Medium

Confidence: 97% confidence
Finding: The code generates concrete financial recommendations and loss/profit scenarios without any explicit warning that the output is not financial advice and may be wrong or incomplete. In a crisis-arbitrage context involving hacks and depegs, omission of such warnings materially increases the chance that users treat a heuristic output as authoritative and suffer losses.

Vague Triggers

Medium

Confidence: 92% confidence
Finding: The trigger list is broad enough to activate on generic investing or news-discussion phrases such as 'should I buy', 'black swan', and depeg-related terms without sufficient narrowing to explicit stablecoin-arbitrage requests. In this context, that can cause the skill to intervene in ordinary financial conversations and steer users toward rapid speculative trading recommendations during crisis events, increasing the chance of unsuitable or harmful advice.

Natural-Language Policy Violations

Medium

Confidence: 89% confidence
Finding: Forcing a Chinese-only output format without checking user language can miscommunicate time-sensitive risk information such as stop conditions, exploit status, and position sizing. In a high-volatility trading scenario, any language mismatch increases the risk that users misunderstand the recommendation or act on incomplete comprehension.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.