Security audit

桌面俯拍商品演示视频｜AI-HIVE

Security checks for vulnerabilities and agentic risk

Overview

The skill is a disclosed AI-HIVE tabletop product-demo video workflow with expected media upload, generation, polling, download, and local video-editing helpers, though users should understand its API-key and billable-generation behavior.

Install only if you intend to use AI-HIVE for product-demo video generation. Expect user-selected media to be uploaded to AI-HIVE, generation tasks may cost money, results may be downloaded under ~/Downloads/AiHive, and init may store an API key locally in ~/.ai-hive/config.json with restricted permissions. Review prompts, routing, and files before running generation or upload commands.

SkillSpector

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
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
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration

Findings (4)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 94% confidence
Finding: The skill exposes code-capable behaviors including environment access, file read/write, network access, and shell-style execution without declaring permissions or clearly constraining their use. This creates a confused-deputy risk where a user or downstream agent may invoke broader local or external actions than expected, including secret exposure, arbitrary file manipulation, or unreviewed outbound requests.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 97% confidence
Finding: The documented purpose is a narrow tabletop product-demo workflow, but the skill apparently includes broader generic AI chat, image generation, wallet/balance queries, model enumeration, media upload, and general ffmpeg editing utilities. This mismatch is dangerous because users, orchestrators, or policy layers may trust the narrower description while the implementation can perform materially broader actions, including account reconnaissance, unscoped uploads, and unintended content operations.

Vague Triggers

Medium

Confidence: 84% confidence
Finding: Overly broad trigger phrases can cause the skill to activate for generic requests such as tutorials, product videos, or creative conversions that may not actually require this tool. In an agentic environment, this increases the chance of inappropriate tool invocation, leading to unnecessary data collection, external API usage, file handling, or billable actions outside user intent.

Natural-Language Policy Violations

Medium

Confidence: 88% confidence
Finding: The default prompt hard-codes Chinese output ('production-ready Chinese workflow') without indicating that this depends on user preference. This can override or bias user intent, reduce usability for other language users, and create misleading behavior if the user expects the assistant to respond in another language.

Static analysis

No suspicious patterns detected.