Security audit

商品演示B-roll自动补镜｜AI-HIVE

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed AI-HIVE product-demo B-roll workflow with expected network, upload, ffmpeg, and API-key use, but users should confirm files and costs before running generation.

Install only if you are comfortable using AI-HIVE with an API key and sending selected reference media to its service. Review generated commands before execution, confirm any billable generation step, avoid uploading sensitive or unauthorized files, and keep the API key out of logs and repositories.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
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

Findings (9)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 90% confidence
Finding: The skill advertises executable scripts that use environment variables, local file access, shell-adjacent tooling (ffmpeg), and network calls, but it declares no permissions or capability boundaries. This creates a transparency and least-privilege problem: a caller or platform may invoke a skill with broader operational access than users expect, increasing the chance of unintended data exposure, filesystem modification, or outbound requests.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 96% confidence
Finding: The documented purpose is a narrowly scoped B-roll gap-filling workflow, but the detected behaviors include broader capabilities such as generic chat/image generation, account and wallet retrieval, model enumeration, generic media upload, and local ffmpeg editing. This mismatch is dangerous because it obscures the true attack surface and can enable misuse of credentials, unexpected billing actions, privacy leakage through uploads, or execution of operations unrelated to the user’s stated intent.

Description-Behavior Mismatch

Medium

Confidence: 89% confidence
Finding: The file is presented as a product-demo B-roll generation skill, but it embeds a broad general-purpose AI client with chat, image generation, model enumeration, user-info, and upload capabilities. This violates least-privilege for skills: an agent invoking this skill for video generation would implicitly gain unrelated powers that could be abused for unauthorized data access, billing visibility, or off-scope AI actions.

Context-Inappropriate Capability

Medium

Confidence: 87% confidence
Finding: The skill exposes account-information and wallet-balance retrieval even though its declared purpose is generating product-demo B-roll footage. In an agent setting, this unnecessarily broadens access to potentially sensitive account metadata and spend information, creating avoidable privacy and reconnaissance risk.

Context-Inappropriate Capability

Medium

Confidence: 85% confidence
Finding: A generic text-chat capability is exposed despite being unrelated to the stated B-roll generation workflow. That gives the skill an unnecessary free-form prompt channel to a remote model, which expands the attack surface for data exfiltration, policy bypass attempts, and unintended billable actions beyond the user's expected task.

Context-Inappropriate Capability

Medium

Confidence: 84% confidence
Finding: The standalone image-generation path is broader than the declared purpose of filling B-roll video gaps. In a skill ecosystem, exposing extra generation modalities can enable unauthorized use, extra billing, and model misuse outside the user-approved workflow.

Vague Triggers

Medium

Confidence: 91% confidence
Finding: The default prompt invokes the skill with a broad, generic phrase and lacks clear scoping conditions, which can cause the agent to activate in contexts the user did not explicitly intend. Because this skill can generate production workflows and runnable AI-HIVE commands, over-broad invocation increases the chance of unintended API use, content generation, or workflow steering.

Natural-Language Policy Violations

Medium

Confidence: 95% confidence
Finding: The default prompt hard-codes Chinese output ('production-ready Chinese workflow') without checking the user's language preference. This can misalign with user intent, reduce transparency for users who cannot read Chinese, and increase the risk that the agent presents commands or operational steps in a language the user did not request or understand.

Missing User Warnings

Medium

Confidence: 79% confidence
Finding: The upload flow sends arbitrary local files to remote storage and then to the API workflow without an explicit user-facing warning or confirmation at the skill interface. In an agent context, this increases the chance that sensitive local media is transmitted off-device without the user's informed consent.

Static analysis

No suspicious patterns detected.