Skill Evaluator

Security checks across malware telemetry and agentic risk

Overview

This skill transparently reviews other skills with multiple sub-agents and shows no hidden, destructive, or credential-seeking behavior.

Install if you want a Chinese-language skill reviewer and are comfortable with it reading the full text contents of the skill directory you ask it to evaluate and sending those contents to several sub-agent review sessions. Use it deliberately on non-sensitive skill folders and expect higher runtime or token usage from the multi-agent workflow.

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 (2)

Vague Triggers

Medium

Confidence: 89% confidence
Finding: The description "帮我评估一下这个 skill。" is extremely broad and conversational, making the skill likely to trigger on ordinary user requests rather than only explicit invocations. This can cause the evaluator skill to activate in unintended contexts and process arbitrary skill contents, increasing the chance of prompt-scope confusion or unintended delegation behavior.

Natural-Language Policy Violations

Medium

Confidence: 80% confidence
Finding: The skill content is written entirely in Chinese and mandates Chinese output structure without offering language selection, which can cause mismatches with user intent and reduce the reliability of security-relevant analysis. While this is not a direct code-execution issue, it can lead to misunderstanding, hidden assumptions, or inaccessible results for users and reviewers who expect another language.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal