Security audit

口播视频去废话精剪｜AI-HIVE

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed AI-HIVE video workflow helper that uses local video tools and explicit API calls, with no evidence of hidden exfiltration or destructive behavior.

Install only if you are comfortable with a skill that can store an AI-HIVE API key, upload user-selected media to AI-HIVE, submit potentially billable generation jobs, and write downloaded outputs locally. Confirm parameters and file paths before running generation or upload commands.

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 (6)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 92% confidence
Finding: The skill exposes substantial operational capabilities—environment variable handling, file read/write, network access, and shell execution—without declaring permissions or constraining when they are used. This weakens reviewability and consent, and could enable unexpected local actions or external data transfer if the associated scripts are invoked in broader agent contexts.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 96% confidence
Finding: The documented purpose is narrow video silence/filler cutting, but the skill also advertises generic AI-HIVE chat, image/video generation, API-key initialization, config management, and media upload workflows. This mismatch expands the effective attack surface and can mislead users or reviewers into approving a skill that can perform broader networked and content-generation actions than expected.

Description-Behavior Mismatch

High

Confidence: 96% confidence
Finding: The skill is advertised as a specialized talking-head silence-cut/editing tool, but the code is actually a generic AI-Hive client that can generate arbitrary videos. This mismatch is dangerous because users and orchestrators may grant broader trust, permissions, or media inputs under the assumption of bounded editing behavior, while the implementation can perform unrelated generation tasks and send content to external APIs.

Intent-Code Divergence

Medium

Confidence: 88% confidence
Finding: The module docstring and CLI describe a broad "generic AI capability" tool, contradicting the skill's narrow declared purpose. This creates a supply-chain and trust-boundary problem: reviewers or users may believe they are enabling a constrained editing utility when they are actually installing a general-purpose remote AI client.

Context-Inappropriate Capability

Medium

Confidence: 92% confidence
Finding: The file exposes unrelated text-chat and image-generation commands that are unnecessary for a talking-video silence-cut skill. Extra capability surface increases abuse potential, weakens least-privilege assumptions, and may let callers use the skill as a generic content-generation proxy rather than the declared editing tool.

Vague Triggers

Medium

Confidence: 91% confidence
Finding: Enabling implicit invocation without tight trigger constraints can cause the skill to activate in broader contexts than intended, including adjacent requests about video editing or marketing content. In this skill, that raises the chance of unintended routing into workflows that generate runnable AI-HIVE commands and media-processing steps, which can amplify unsafe or policy-sensitive user requests without explicit confirmation.

Static analysis

No suspicious patterns detected.