ClawHub
Back to skill

Security audit

Cloud Architect Free

Security checks across malware telemetry and agentic risk

Overview

This cloud architecture advice skill is not clearly malicious, but it asks for host command execution without enough limits for an advisory-only use case.

Install only if you are comfortable with this advisory cloud-architecture skill having command execution available. Prefer using it for design guidance only, and do not allow it to run diagnostics unless the target and command are explicit and intentional.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (4)

Context-Inappropriate Capability

Medium
Confidence
95% confidence
Finding
This skill is presented as an advisory/documentation skill for cloud architecture, but it declares the `exec` tool and repeatedly frames itself as able to 'execute' operations. That creates an unnecessary capability/expectation mismatch that could let user-supplied input be turned into host-level command execution, especially since the document later references running diagnostic commands like `ping`.

Intent-Code Divergence

Low
Confidence
83% confidence
Finding
The file claims it does not directly access user cloud accounts, yet it instructs execution of network diagnostic commands such as `ping`. While `ping` is lower risk than arbitrary shell execution, it still causes external network interaction from the agent host and contradicts the stated non-operational scope, which can be abused for probing or unintended outbound activity.

Vague Triggers

Medium
Confidence
80% confidence
Finding
The invocation guidance tells the agent to execute tasks based on broad references to the skill’s capabilities, without concrete trigger boundaries, approval requirements, or limits on when execution-capable behavior should be used. In a skill that has `exec` available, vague routing increases the chance that unrelated or attacker-shaped prompts get interpreted as permission to perform actions rather than provide advice.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The skill metadata exposes command execution capability in a context that reads like cloud consulting guidance, but it does not warn users or the orchestrator about possible host-level side effects. That omission makes it easier for the skill to be selected in low-risk-seeming contexts while still possessing system-impacting behavior, increasing the chance of unsafe execution or trust confusion.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.