Llm As Judge

• Outcome: Pass
• Latest audit: May 28, 2026 at 10:11 AM UTC
• Version:
  1.0.1

Overview

This is a coherent instructional skill for LLM evaluation that discloses cloud API use and has no hidden execution, but users should handle evaluation data and logs carefully.

Install this if you intend to run LLM evaluation workflows with external judge providers. Use scoped API keys with spending limits, confirm provider data-handling settings, and avoid logging raw prompts, source text, candidates, ground truth, or judge rationales unless the data is non-sensitive or logs are redacted, access-controlled, and retention-limited.

SkillSpector

By NVIDIA

Vulnerability Patterns

• 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
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (1)

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: The guidance to "store raw judge responses for debugging score disputes" encourages logging full model outputs without any warning about sensitive, user-provided, or proprietary content that may appear in prompts, candidates, or judge rationales. In an evaluation pipeline, these logs can accumulate confidential data and create a secondary exposure surface through log stores, dashboards, backups, and support tooling.

VirusTotal

64/64 vendors flagged this skill as clean.