ClawHub
Back to skill

Security audit

虚拟主播商品播报视频｜AI-HIVE

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed AI-HIVE virtual-host product-video workflow with expected API, media upload, and local ffmpeg helpers, and I did not find hidden exfiltration or destructive behavior.

Install only if you intend to use AI-HIVE for virtual-host product-video work. Review prompts, routing, model, pricing snapshot, and selected media before generation; do not provide real API keys in chat logs or repositories; upload only assets you are authorized to use; and verify product claims, prices, inventory, platform rules, and legal constraints outside the model output.

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (4)

Lp3

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding
The skill exposes operational capabilities including environment-variable use, file read/write, network access, and shell execution, but does not declare permissions or boundaries. This weakens reviewability and can lead to unintended secret access, filesystem changes, or external calls if the skill is invoked in a broader agent runtime.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding
The documented purpose is a constrained virtual-host product video workflow, but the described behavior includes broader generic chat, image generation, model discovery, balance retrieval, upload utilities, and local ffmpeg editing. Capability drift like this is dangerous because users and policy layers may authorize the skill for a narrow use case while it can actually access unrelated APIs, account metadata, and local media-processing functions.

Vague Triggers

Medium
Confidence
82% confidence
Finding
The invocation criteria are broad enough that the skill may trigger on loosely related requests, including requests for scripts, prompts, code, or task lists derived from reference cases. Over-broad routing increases the chance of unnecessary access to networked generation, local file/media utilities, or paid operations in contexts where a simpler and safer skill should have been used.

Natural-Language Policy Violations

Medium
Confidence
88% confidence
Finding
The default prompt hard-codes Chinese output ('production-ready Chinese workflow') without checking the user's language preference or requiring opt-in. This can override user intent, reduce accessibility, and cause the agent to produce output in an unexpected language, which is a real prompt-quality and safety issue even though it is not directly enabling code execution or data exfiltration.

Static analysis

No suspicious patterns detected.