AI服装虚拟试穿效果图｜AI-HIVE

• Outcome: Pass
• Latest audit: Aug 24, 2026 · 3:54 PM UTC
• Version: 1.0.0

Overview

The skill appears to match its advertised AI-HIVE virtual try-on image workflow, but users should expect third-party uploads, API-key use, and possible generation charges.

Install only if you are comfortable giving AI-HIVE an API key, uploading authorized person and garment images to AI-HIVE/object storage, and reviewing parameters before any billable generation. Keep generated logs and task records free of real API keys, and remember the init command stores the key locally under your home directory.

SkillSpector

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 90% confidence
• Finding: The skill advertises executable workflows that use environment variables, local file access, shell commands, and network calls, yet no explicit permission model is declared. This creates a capability/expectation gap: a host or reviewer may treat the skill as low-privilege content while it can drive potentially billable external API calls, read local inputs, and write outputs. In a skill ecosystem, undeclared capabilities reduce transparency and make misuse or accidental overreach harder to detect.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 95% confidence
• Finding: The documented scope is narrow—virtual try-on preview generation with authorized person and garment references—but the described behavior appears to expose broader generic AI platform functions such as arbitrary model listing, generic uploads, video generation, chat, and account or wallet data access. That mismatch can let users or downstream agents invoke more powerful operations than reviewers expect, increasing the risk of privacy exposure, unexpected charges, policy bypass, or use outside the consent boundaries implied by the skill description.

Description-Behavior Mismatch

Medium

• Confidence: 90% confidence
• Finding: The file is a generic AI-Hive client exposing chat, image, video, model listing, upload, and account-related functions, while the advertised skill is narrowly scoped to apparel virtual try-on preview generation. This scope mismatch weakens containment and allows callers to use the skill as a broader capability broker than users and reviewers would reasonably expect.

Description-Behavior Mismatch

Medium

• Confidence: 95% confidence
• Finding: The parser and command map expose generic commands such as user-info, models, chat, video, task, and upload that are unnecessary for a specialized virtual try-on preview skill. In an agent environment, these extra entry points expand the attack surface and can be abused to access account metadata, upload arbitrary content, or invoke unintended model capabilities.

Natural-Language Policy Violations

Medium

• Confidence: 86% confidence
• Finding: The default prompt hard-codes Chinese output ('production-ready Chinese workflow'), which can override or conflict with a user's language preference without explicit opt-in. This is primarily a prompt-quality and user-consent issue rather than a direct security exploit, but it can cause unintended behavior, reduce transparency, and make downstream review harder for users expecting another language.

Static analysis

No suspicious patterns detected.