Security audit

参考动作迁移短视频｜AI-HIVE

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed AI-HIVE video-generation workflow that uses user-provided media and an API key, with some non-exposed generic helper code bundled in the script.

Before installing, be aware that this skill sends selected media and prompts to AI-HIVE and can run billable generation after confirmation. Use only media you are authorized to upload, keep the API key out of shared logs, and review the persisted key at ~/.ai-hive/config.json if you use init.

SkillSpector

Vulnerability Patterns

Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
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
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration

Findings (6)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 94% confidence
Finding: The skill exposes substantial capabilities—environment variable access, file read/write, network access, and shell execution—yet declares no permissions or capability boundaries. This creates a trust and review gap: users and the platform may invoke a skill that can access secrets, manipulate local files, and make outbound requests without explicit disclosure or gating.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 91% confidence
Finding: The documented purpose is narrowly scoped to reference-motion-transfer workflows, but the skill appears to include broader functions such as generic chat/image generation, wallet lookup, model enumeration, standalone uploads, browser-based API-key setup, and local ffmpeg operations. This mismatch increases the attack surface and can mislead users or reviewers about what the skill may do, enabling unexpected data exposure, unauthorized account actions, or execution of risky local operations under a narrower-looking description.

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The script is a general-purpose AI Hive client exposing chat, model listing, image generation, uploads, task polling, and account-related functions, while the declared skill is narrowly framed as reference-motion-transfer video generation. This expands the skill's effective capability surface beyond user expectations and policy scope, increasing the chance of misuse or unauthorized invocation of unrelated functions.

Context-Inappropriate Capability

Low

Confidence: 84% confidence
Finding: The user-info endpoint retrieves account and wallet data that is not necessary to generate motion-transfer videos. Even if not directly exposed automatically, bundling this capability into the skill creates unnecessary access to sensitive account metadata and broadens the privacy and abuse surface.

Context-Inappropriate Capability

Medium

Confidence: 93% confidence
Finding: The text chat capability is unrelated to the stated motion-transfer video purpose and allows arbitrary prompt forwarding to external AI models. This violates least privilege for skills and can be abused as an unexpected generic LLM proxy under the cover of a narrower media-generation skill.

Context-Inappropriate Capability

Medium

Confidence: 92% confidence
Finding: Standalone image generation exceeds the declared purpose of a reference-motion-transfer video skill and introduces another unrelated generative pathway. This capability broadening increases misuse potential and makes policy enforcement harder because the skill can perform actions outside its advertised scope.

Static analysis

No suspicious patterns detected.