Security audit

图像处理专业版

Security checks across malware telemetry and agentic risk

Overview

The skill is mainly an image-processing helper, but it needs review because it combines command/write access with broad triggering, external upload/webhook features, and an unrelated credential-environment check.

Review this skill before installing. Use it only for image-processing tasks, avoid running the environment-variable discovery example, and require explicit approval before batch writes, CDN uploads, webhook callbacks, or any operation that sends images or metadata to an external service.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands

Findings (5)

Context-Inappropriate Capability

Medium

Confidence: 97% confidence
Finding: The skill includes a validation step that enumerates environment variables matching API, KEY, TOKEN, SECRET, or PROFILE, which is unrelated to core image-processing behavior. Even though the example masks values with sed, it still reveals the presence and names of sensitive variables and normalizes secret discovery behavior inside a skill that also has exec capability.

Intent-Code Divergence

Medium

Confidence: 94% confidence
Finding: The document states that outputs should not contain secrets, but elsewhere instructs commands that list sensitive environment-variable names. This contradiction weakens operator trust, encourages unsafe debugging practices, and can expose metadata about credentials or integrations that aids follow-on attacks.

Vague Triggers

Medium

Confidence: 83% confidence
Finding: The description is broad enough to activate for generic efficiency, automation, workflow optimization, and decision-support requests beyond image processing. In a skill with read, write, and exec tools, overbroad routing increases the chance the agent applies this skill in inappropriate contexts and executes commands or networked workflows without clear user intent.

Vague Triggers

Medium

Confidence: 81% confidence
Finding: The invocation guidance tells users to describe needs in natural language and says the agent will automatically execute corresponding operations, but it does not define scope boundaries or approval points. This makes the skill easier to trigger accidentally and more dangerous because it can drive command execution for loosely related requests.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The skill advertises CDN upload pipelines and includes callback/webhook support, but does not give a prominent user-facing warning that files and metadata may be transmitted to external services. In an enterprise image workflow, images can contain sensitive content or metadata, so silent or poorly disclosed exfiltration risk is material.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.