AI大模型专家｜HappyHorse 视频生成与编辑

• Outcome: Pass
• Latest audit: Aug 24, 2026 · 2:33 AM UTC
• Version: 1.0.0

Overview

This skill is a disclosed AI-HIVE video-generation helper that handles API keys, media upload, paid task submission, polling, and downloads in ways that match its stated purpose.

Install only if you trust AI-HIVE with the prompts and media you provide, and confirm cost-sensitive or private-media jobs before the agent submits them. Keep the API key out of chats and repositories; prefer the 0600 config file or environment variable.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands

Findings (7)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 87% confidence
• Finding: The skill documents and encourages capabilities including network access, shell execution, reading environment variables, and reading/writing local files, but it does not declare permissions or clearly bound those capabilities. This weakens user and platform visibility into what the skill can do and increases the risk of over-privileged execution, especially because it handles API keys, opens browser flows, writes config files, and downloads remote content.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 93% confidence
• Finding: The documented purpose is a narrowly scoped HappyHorse video skill, but the analyzed behavior indicates a broader AI-HIVE client that can access unrelated capabilities such as generic chat, image generation, model enumeration, and user/account information including wallet balance. That mismatch is dangerous because users may invoke the skill expecting limited video operations while the implementation can access broader account data and platform functions than disclosed.

Description-Behavior Mismatch

Medium

• Confidence: 95% confidence
• Finding: The file is packaged as a HappyHorse video-generation/editing skill, but the implementation exposes a broader generic AI Hive client with chat, image generation, model enumeration, uploads, and task management. This scope expansion violates least privilege and increases the attack surface, enabling callers to use capabilities unrelated to the declared skill purpose and potentially bypass policy or review assumptions tied to the skill metadata.

Context-Inappropriate Capability

Medium

• Confidence: 93% confidence
• Finding: The skill includes a user-info endpoint that returns account information and wallet balance, which is unnecessary for a video-generation/editing skill. Exposing account metadata broadens data access beyond task execution and can leak sensitive billing or identity details to users or upstream agents that should only be able to create and manage video jobs.

Context-Inappropriate Capability

Medium

• Confidence: 95% confidence
• Finding: The generic text-chat capability is unrelated to the declared HappyHorse video/editing scope and materially expands what the skill can do. In an agent environment, this can be abused as an unintended general LLM interface, potentially enabling policy bypass, unreviewed prompt handling, or exfiltration of user-supplied content to a service not expected by the skill contract.

Context-Inappropriate Capability

Medium

• Confidence: 94% confidence
• Finding: Standalone image generation falls outside the stated HappyHorse video-generation/editing purpose and creates an unnecessary extra capability surface. This mismatch makes the skill more dangerous because consumers may trust it as a video-only tool while it can also perform unrelated model operations and send prompts or assets to different backend paths.

Vague Triggers

Medium

• Confidence: 95% confidence
• Finding: The skill enables implicit invocation while exposing a very broad description and a default prompt that can trigger real AI-HIVE video-generation workflows, including uploading user-provided media, submitting jobs, polling progress, and downloading outputs. Without narrow trigger constraints or explicit user confirmation, unrelated conversations could cause the agent to invoke this skill and perform unintended external actions, potentially sending sensitive files or incurring cost.

VirusTotal

65/65 vendors flagged this skill as clean.

Static analysis

No suspicious patterns detected.