ClawHub
Back to skill

Security audit

Docker核心操作免费版

Security checks across malware telemetry and agentic risk

Overview

This Docker helper is mostly coherent, but it tells agents to execute state-changing Docker commands without clear confirmation or safety boundaries.

Install only if you are comfortable with an agent helping run Docker commands on your machine. Review generated commands before execution, especially docker rm, docker rmi, prune, docker run with ports or bind mounts, and commands that pull or run third-party images.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

Vague Triggers

Medium
Confidence
88% confidence
Finding
The skill description is broadly scoped to 'development scenarios' and 'docker essentials related capabilities' without clear activation boundaries, which can cause an agent to apply the skill in situations where command execution was not explicitly intended. Because the skill also exposes exec/write capabilities, vague triggering increases the chance of unintended system-affecting Docker actions.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
This section includes destructive or state-changing Docker commands such as container/image deletion and prune operations, but does not warn about data loss, service interruption, or the scope of cleanup. In an agent context, users may treat these examples as safe defaults, leading to accidental deletion of containers, images, or attached workflows.

Missing User Warnings

Medium
Confidence
97% confidence
Finding
The skill explicitly states that the agent will generate and execute Docker commands, but it does not warn that execution can modify the host environment, expose ports, start services, or pull and run untrusted images. In combination with the declared exec tool, this creates a meaningful risk of unintended system changes from natural-language requests.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.