Wip Grok

Security checks across malware telemetry and agentic risk

Overview

This is a coherent xAI Grok integration, but it needs review because its agent-exposed image editing path can read arbitrary local file paths without containment or a clear consent boundary.

Review before installing, especially if exposing the MCP tools to an autonomous agent. Use a dedicated xAI API key with spending limits, avoid confidential prompts or private media, and do not allow untrusted prompts to choose local image paths or output paths.

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 (6)

Context-Inappropriate Capability

Medium

Confidence: 97% confidence
Finding: The image editing path accepts non-URL, non-data inputs as filesystem paths and reads them from local disk, expanding the skill from API mediation into arbitrary local file access. In an agent context, this can expose sensitive local files for subsequent transmission to the external xAI API, especially because the feature is not tightly constrained to an approved directory or file type beyond MIME inference by extension.

Intent-Code Divergence

Low

Confidence: 94% confidence
Finding: The docstring states that edit_image accepts image URLs or base64 data URIs, but the implementation silently also reads local file paths from disk. This mismatch is security-relevant because callers may assume no local file access occurs, when in fact arbitrary readable files can be ingested and later sent to a third-party API.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The skill advertises web and X search features but does not clearly warn that user queries and possibly related context will be transmitted to xAI/X-controlled external services. This can cause unintentional disclosure of sensitive prompts, proprietary data, or personal information when users treat the skill like a local capability rather than a third-party networked service.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The image/video generation and editing sections describe uploading prompts, URLs, file paths, base64 images, and seed images to the external xAI API without any privacy or data-handling warning. Users may unknowingly send sensitive images, copyrighted material, or confidential prompts off-platform, creating privacy, compliance, and data leakage risk.

Missing User Warnings

High

Confidence: 98% confidence
Finding: This code reads local files from arbitrary paths and converts them to data URLs in preparation for image-edit requests, creating a path from local disk access to third-party transmission. In an agent skill, that combination materially increases data-exfiltration risk because a caller may provide a sensitive path and have its contents uploaded without a strong consent boundary.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: This MCP server exposes multiple network-backed tools that send user-supplied prompts, queries, images, and other inputs to xAI Grok services, but the tool definitions and handlers do not provide any explicit privacy or third-party disclosure warning. In an agent context, users may reasonably assume inputs stay local unless told otherwise; this can lead to unintended disclosure of sensitive data to an external provider.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal