Guardian Shield

• Outcome: Pass
• Latest audit: May 29, 2026 at 8:13 PM UTC
• Version:
  1.1.1

Overview

This skill is a local prompt-injection scanner, and the flagged attack phrases are used as detection data rather than instructions to control the agent.

Install this only if you want a local scanner that may warn on or block untrusted content before your agent processes it. Review config.json thresholds and scan mode, pin optional Python dependencies in your own environment if supply-chain control matters, and expect false positives on security research text or examples containing jailbreak language.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (4)

Ssd 1

High

• Confidence: 98% confidence
• Finding: The vocabulary explicitly includes many jailbreak and role-play prompt fragments such as acting as unrestricted personas and ignoring safety guidance. In a model or classifier asset, this strongly suggests the skill is designed to detect, reproduce, or facilitate prompt-injection bypass patterns, which can materially weaken downstream guardrails if reused at inference time.

Ssd 2

High

• Confidence: 99% confidence
• Finding: The file contains extensive obfuscated, multilingual, and encoded jailbreak phrases, including variations intended to evade straightforward filtering. That breadth and diversity increase the likelihood the artifact is meant to normalize or operationalize prompt-bypass techniques rather than merely incidentally contain them.

Ssd 3

High

• Confidence: 98% confidence
• Finding: The vocabulary includes many phrases requesting confidential information, credentials, patient data, and other sensitive user data. Shipping an artifact that embeds these exfiltration prompts can enable systems built on it to classify, trigger on, or even prioritize dangerous data-access attempts.

Ssd 4

High

• Confidence: 99% confidence
• Finding: This vocabulary contains long-form, staged jailbreak scripts that progressively coerce a model into unrestricted behavior, including token systems, alternate personas, and explicit instructions to bypass policy. Such multi-step prompt structures are particularly dangerous because they are crafted to survive partial filtering and manipulate model state over multiple turns.

VirusTotal

66/66 vendors flagged this skill as clean.