Skill Sharpener

Security checks across malware telemetry and agentic risk

Overview

This skill is a disclosed local quality auditor for other skills, with optional user-approved edits rather than hidden automatic modification.

Install if you want a skill that can audit and help improve other skills. Point it only at the skill directory you intend to review, inspect proposed changes before accepting fixes or rewrites, and keep backups until you verify the edited skill.

SkillSpector

By NVIDIA

Vulnerability Patterns

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
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (2)

Description-Behavior Mismatch

Medium

Confidence: 93% confidence
Finding: The skill is described as an auditor that assesses and recommends, but it later asks whether to apply fixes, rewrite descriptions, and fully restructure the target skill. This expands behavior from analysis into modification, creating a scope mismatch that can lead to unintended file changes and user confusion about what the skill is authorized to do.

Intent-Code Divergence

Medium

Confidence: 91% confidence
Finding: The documentation frames the workflow as producing an assessment report, then transitions into editing behavior after report generation. This hidden state change increases the chance that an operator invokes a read-only review workflow but gets a write-capable one, which is dangerous in systems that rely on skill descriptions for trust and approval decisions.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal