Advanced Evaluation

Security checks across malware telemetry and agentic risk

Overview

This is a documentation-only skill for LLM evaluation workflows, with one wording issue around chain-of-thought prompting but no code execution or hidden data access.

Reasonable to install as a low-risk evaluation guide. When using it, revise prompts to ask judges for concise evidence, rubric-grounded rationale, scores, and confidence, and avoid asking models to reveal hidden chain-of-thought. For sensitive evaluation data, use only approved model providers and pipelines.

SkillSpector

By NVIDIA

Vulnerability Patterns

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
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (2)

Ssd 2

Medium

Confidence: 98% confidence
Finding: The skill explicitly instructs evaluators to require justification before scoring and labels this as a chain-of-thought requirement, which encourages elicitation of hidden reasoning. In production systems, prompting for internal reasoning can increase leakage of sensitive intermediate analysis, expose policy-related reasoning patterns, and conflict with safer best practices that request brief evidence or concise rationale instead of full internal chain-of-thought.

Ssd 2

Medium

Confidence: 98% confidence
Finding: The guideline reiterates that prompts should require justification before scores and explicitly endorses chain-of-thought prompting, reinforcing a pattern of soliciting internal reasoning. This makes the issue more dangerous because it is presented as a normative rule for all uses of the skill, increasing the chance downstream users will operationalize unsafe prompting patterns across evaluation pipelines.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal