AI 趋势分析

• Outcome: Pass
• Latest audit: May 29, 2026 at 8:29 PM UTC
• Version:
  0.1.3

Overview

This is a disclosed AI-industry research skill that makes narrow read-only API requests with a configured token, with no evidence of hidden persistence, destructive behavior, or unrelated data access.

Install this only if you are comfortable configuring a JQZX API token and sending AI-industry search queries to the disclosed service. Keep the token revocable, avoid confidential query terms, and verify the BASE_URL is the intended provider endpoint before use.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (2)

Intent-Code Divergence

Medium

• Confidence: 94% confidence
• Finding: The workflow instructs the agent to use 'LLM memory' to expand search keywords, which directly contradicts the stated prohibition on relying on model memory for factual AI-industry answers. This inconsistency weakens safety boundaries and can cause the agent to introduce hallucinated or unverified terms into retrieval, contaminating evidence chains and undermining traceability.

Vague Triggers

Medium

• Confidence: 91% confidence
• Finding: The skill claims mandatory use for essentially any AI-industry question and forbids answering from general model capability, creating an overly broad activation scope. In a multi-skill or assistant environment, such blanket routing can override user intent, force unnecessary external data access, and increase privacy, availability, and prompt-routing risk.

VirusTotal

65/65 vendors flagged this skill as clean.