Security audit

Agent 测试

Security checks across malware telemetry and agentic risk

Overview

This skill is a disclosed QA framework for testing AI agents, with only minor scoping and install-guidance caveats.

Installers should treat this as a QA planning skill, not an execution harness. Review the optional recommendation to install the full external QA skill set before running it, and clarify user intent when the request is merely about AI assistants or chatbots rather than testing them.

SkillSpector

By NVIDIA

Vulnerability Patterns

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

Vague Triggers

Medium

Confidence: 90% confidence
Finding: The activation conditions include broad terms such as AI助手, 聊天机器人, and LLM测试, which can match many general requests outside a narrowly scoped Agent-testing workflow. This can cause the skill to trigger unexpectedly, leading to irrelevant instructions being applied and increasing the chance of prompt-routing mistakes or unsafe tool-enabled behavior in contexts that did not request this specialized test process.

Vague Triggers

Medium

Confidence: 86% confidence
Finding: The startup instruction says to begin testing after the user raises an Agent testing need, but it does not define clear thresholds or disambiguation rules for mixed-purpose conversations. In a skill-routing system, this ambiguity can misapply the skill in borderline cases, causing the agent to switch into a testing/security workflow without sufficient user intent or context validation.

VirusTotal

62/62 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.