Security audit

数字人试穿展示视频｜AI-HIVE

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed AI-HIVE virtual try-on video workflow whose API use, media uploads, local outputs, and optional API-key setup are aligned with its stated purpose.

Install only if you are comfortable using AI-HIVE for virtual try-on videos. Provide only media you have rights to use, review model/routing/price details before generation, and remember that selected files may be uploaded to AI-HIVE and results may be saved locally. Use the optional API-key config only on a machine where storing that key in your home directory is acceptable.

SkillSpector

Vulnerability Patterns

Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
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 (8)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 92% confidence
Finding: The skill advertises executable workflows that use environment variables, local file reads/writes, network access, and shell-level tools like ffmpeg, but no permissions are explicitly declared. That creates a transparency and governance gap: operators and users cannot accurately assess what the skill may access or modify before use, increasing the risk of unintended data exposure, local file modification, or external API calls.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 94% confidence
Finding: The skill is presented as a narrowly scoped digital try-on video tool, but the documented behavior includes broader capabilities such as generic AI chat/image generation, model and wallet queries, standalone uploads, ffmpeg media manipulation, and local credential setup/storage. This mismatch is dangerous because it can cause reviewers or users to trust and invoke a much more powerful tool than advertised, enabling unexpected data handling, billing-related actions, or abuse of local/media-processing functionality outside the stated purpose.

Description-Behavior Mismatch

Medium

Confidence: 94% confidence
Finding: The file presents itself as a specialized avatar try-on video skill, but actually exposes a broad general-purpose AI Hive client with chat, image, video, upload, and account-related operations. This scope expansion increases attack surface and enables capability misuse beyond the declared purpose, which is risky in agent ecosystems that rely on skill descriptions for trust and permission expectations.

Context-Inappropriate Capability

Low

Confidence: 84% confidence
Finding: The skill can query user account information and wallet balance even though that capability is unrelated to producing avatar try-on videos. In a skill-based environment, unnecessary account inspection expands data exposure and could reveal sensitive billing or identity information to users or upstream orchestrators without a business need.

Context-Inappropriate Capability

Medium

Confidence: 90% confidence
Finding: The generic text chat endpoint is unrelated to the declared avatar try-on video purpose and provides a much broader interaction surface than users would expect from this skill. That mismatch can be abused to route arbitrary prompts and external media through the provider using credentials provisioned for a narrowly scoped workflow.

Context-Inappropriate Capability

Medium

Confidence: 90% confidence
Finding: The script exposes generic image generation beyond the stated try-on video role, allowing use of the API key and model routing for unrelated content generation. This broadening of functionality weakens least-privilege expectations and can lead to policy bypass, abuse of paid resources, or generation of unrelated content under the guise of the try-on skill.

Intent-Code Divergence

Medium

Confidence: 88% confidence
Finding: The module and CLI documentation describe a general-purpose AI tool rather than a narrowly scoped avatar try-on skill. In agent marketplaces or orchestration systems, misleading documentation can cause the skill to be granted trust, invocation contexts, or user expectations that do not match its real capabilities.

Vague Triggers

Medium

Confidence: 93% confidence
Finding: The default prompt and display wording are broad enough to trigger this skill for a wide range of user requests about virtual try-on, ecommerce video, or AI model content without a strong authorization or safety gate. Because this skill can produce runnable AI-HIVE commands and media-generation workflows, unintended activation could facilitate misuse such as unauthorized likeness use, deceptive marketing assets, or policy-evasive content generation.

Static analysis

No suspicious patterns detected.