Security audit

OSNK Trainer

Security checks across malware telemetry and agentic risk

Overview

This is a disclosed educational practice skill with local question files, local progress storage, and a limited GitHub fallback for question text, with no evidence of hidden exfiltration or destructive behavior.

Install if you are comfortable with a Bash-based training helper that stores practice stats locally and may contact GitHub for missing question text. For stricter offline use, keep the bundled/local question files available or block network access. Verify important answer keys against official sources, as the skill itself warns they may not be fully accurate.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (4)

Tp4

High

Category: MCP Tool Poisoning
Confidence: 85% confidence
Finding: The finding indicates the skill fetches content from an external GitHub raw repository despite presenting itself as an offline/local trainer. Undisclosed remote content retrieval changes the trust boundary: remote question data could be altered, become unavailable, or be used to influence outputs unexpectedly, and users are not properly informed that network access may occur.

Context-Inappropriate Capability

Medium

Confidence: 89% confidence
Finding: The script silently pulls training content from a remote GitHub URL when local files are absent, creating an untrusted content dependency at runtime. Even though the fetched data is only displayed rather than executed, remote content can be changed without user awareness, enabling deceptive, malicious, or policy-violating output and creating privacy/network-exposure concerns in an otherwise local-looking skill.

Missing User Warnings

Low

Confidence: 84% confidence
Finding: Fetching remote content without any user-facing notice is a transparency and trust problem. In the context of a training skill expected to operate on local question banks, silent outbound requests can surprise users, disclose usage metadata, and expose them to unreviewed remote content.

Missing User Warnings

Low

Confidence: 84% confidence
Finding: The year-specific fallback also performs undisclosed remote fetches, again violating user expectations of a local educational tool. While the content is not executed, the behavior still leaks network metadata and permits delivery of unvetted content from an external source.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.