Security audit

Model Routing Middleware

Security checks across malware telemetry and agentic risk

Overview

The skill mostly does what it claims, but it needs review because it logs prompt excerpts by default and can activate on broad, generic terms.

Review and change the logging settings before installing, especially if prompts may contain secrets, private data, proprietary code, or regulated information. Narrow the activation trigger if possible, and verify cloud routes or escalation targets match your data-handling policy before connecting this router to real model providers.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
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
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (4)

Description-Behavior Mismatch

Medium

Confidence: 94% confidence
Finding: The router logs a preview of the user prompt, which can capture secrets, personal data, credentials, proprietary code, or other sensitive inputs. Because this component's role is routing rather than content retention or auditing, storing prompt content in logs expands the data exposure surface and can leak information to operators, log processors, or third-party observability systems.

Vague Triggers

Medium

Confidence: 84% confidence
Finding: The trigger phrase is broad and generic, covering common terms like model routing, LLM selection, cost optimization, middleware, and agent. In an agent ecosystem, this can cause the skill to activate in unrelated contexts, increasing the chance of unintended invocation and execution of its logic or instructions when the user did not explicitly request it.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The configuration explicitly enables prompt preview logging, which can capture portions of user prompts and other sensitive context into a local log file. In an AI middleware skill that routes varied tasks and context, those previews may include secrets, personal data, credentials, or proprietary content, creating an unnecessary data exposure risk if logs are accessed, retained, or forwarded.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: Prompt content is written to logs without any visible disclosure or consent mechanism. In practice, users may submit confidential material to a router expecting transient processing, but the implementation persists excerpts into logs, creating a silent confidentiality and privacy risk.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.