Jean-Claw Van Damme

Security checks across malware telemetry and agentic risk

Overview

This security skill is purpose-aligned and does not show exfiltration or destructive behavior, though users should understand its local audit logging and helper scripts before installing.

Install only if you are comfortable with a security skill influencing agent authorization decisions and writing local audit/grant records. Review the two shell scripts first, keep the data directory out of shared repos or backups when needed, and periodically delete or rotate audit files that may reveal sensitive activity.

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)

Intent-Code Divergence

Medium

Confidence: 88% confidence
Finding: The README makes a strong trust-building claim that the skill is 'pure markdown' with 'no external dependencies,' but later documents shell scripts and runtime state files as part of the implementation. This discrepancy can mislead users about the executable surface area and operational behavior, causing them to under-scrutinize the skill before installation.

Missing User Warnings

Low

Confidence: 76% confidence
Finding: The README states that audit, grants, policy, and threat logs are created at runtime but does not warn users that potentially sensitive local data will be retained on disk. In a security-focused skill, undocumented retention can expose operational history, approval decisions, and possibly sensitive scopes or identifiers to other local users, backups, or accidental commits.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal