Security audit

政策资产联动日报

Security checks across malware telemetry and agentic risk

Overview

This skill coherently generates a market-policy HTML report, with one disclosed external market-data widget but no evidence of hidden persistence, credential access, exfiltration, or destructive behavior.

Install only if you want bilingual policy-to-market HTML reports that use current web research. Be aware that generated reports include a TradingView real-time ticker, so opening the HTML may contact TradingView and run its embedded widget script; use a browser or network controls appropriate for your privacy needs.

SkillSpector

By NVIDIA

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection

Findings (5)

Context-Inappropriate Capability

Medium

Confidence: 91% confidence
Finding: The template loads a third-party TradingView widget into what is described as a single-file report. This introduces external script execution, network dependency, tracking/supply-chain risk, and means the report is no longer self-contained or safe for offline/local use.

Vague Triggers

Medium

Confidence: 89% confidence
Finding: The description and trigger language are broad enough to match common finance and reporting requests such as 'daily macro briefing' or 'generate topic page,' which can cause the skill to activate outside its narrowly intended use. Over-broad routing can steer users into unsolicited web-driven report generation, increasing the chance of incorrect tool use, unnecessary data access, or confusing outputs that do not match user intent.

Natural-Language Policy Violations

Medium

Confidence: 92% confidence
Finding: The skill metadata hard-codes bilingual Chinese/English output without indicating that the user can choose a preferred language. This can override user expectations, create unnecessary disclosure or formatting of content, and reduce usability for users who only requested a single-language output.

Natural-Language Policy Violations

Medium

Confidence: 94% confidence
Finding: The body instructions repeatedly require a bilingual report and top-right language toggle, but do not document any opt-in or override based on user preference. This creates a persistent behavior mismatch where the skill may generate more content than requested and ignore localization expectations, which is a policy/quality weakness even if not a direct code-execution risk.

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: The page silently performs a network request to a third-party domain for market data/script content without informing the user. This creates privacy and transparency issues, and exposes report viewers to remote-content risks that are unexpected for a local HTML report.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.