Security audit

Azure Cli Toolkit

Security checks across malware telemetry and agentic risk

Overview

This Azure admin skill is mostly purpose-aligned, but it mixes in unrelated review-grading instructions and shows high-impact Azure deletion and credential workflows without enough safeguards.

Review this skill carefully before installing. Use it only for Azure environments where the agent is allowed to administer resources, and require explicit subscription and resource-group scope, list-first/dry-run checks, and human confirmation before any delete, deallocate, deployment, or cross-subscription action. Prefer managed identity or tightly scoped service principals, store secrets in a secret manager, and ignore the unrelated grading-style sections because they do not match the Azure administration purpose.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Intent-Code Divergence

High

Confidence: 96% confidence
Finding: The documented input/output contract switches from Azure administration to a generic code-review grading workflow, creating a dangerous semantic mismatch between what the skill claims to do and how an agent may interpret requests. In an MD+EXEC skill, this kind of spec confusion can cause the agent to apply the wrong operational logic, mishandle user input, or execute unintended actions while presenting misleading results.

Intent-Code Divergence

Medium

Confidence: 93% confidence
Finding: The error-handling section instructs the agent to treat failures like content-review or network-diagnostic issues rather than Azure operational errors, including suggestions to run ping commands unrelated to the declared function. This increases the chance of unsafe fallback behavior, unnecessary command execution, and operator confusion during incidents.

Intent-Code Divergence

High

Confidence: 95% confidence
Finding: The usage examples show grading 'content' and producing evaluation scores instead of Azure management tasks, reinforcing a contradictory execution model inside a skill that exposes exec capability. This mismatch can mislead users and agents about intended behavior, making unsafe or incorrect command execution more likely under ambiguous prompts.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The skill documents bulk deletion and cleanup pipelines using xargs and force flags without prominent safety gates, dry-run guidance, scope validation, or confirmation requirements. In an agent-executed environment, these examples normalize destructive behavior and could lead to mass resource deletion across resource groups or subscriptions if adapted from user prompts.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The skill instructs users to handle service principal credentials and export secrets into environment variables without a clear upfront warning about secret exposure, logging, shell history, and least-privilege practices. Because the skill is designed for automation and exec usage, careless adoption could expose high-value Azure credentials and enable account compromise.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.