视觉解压循环广告｜AI-HIVE

• Outcome: Pass
• Latest audit: Aug 25, 2026 · 9:58 AM UTC
• Version: 1.0.0

Overview

This skill is coherent for AI-HIVE video-ad creation, with normal cautions around API keys, external uploads, paid generation, and stored output files.

Install this only if you intend to use AI-HIVE for ad/video generation. Use a limited AI-HIVE API key, review routing and cost before generation, do not upload private or unlicensed media, and remember that init stores the key locally in ~/.ai-hive/config.json while generated outputs may be saved under ~/Downloads/AiHive.

SkillSpector

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection

Findings (9)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 92% confidence
• Finding: The skill advertises executable workflows that use environment variables, local file operations, network access, and shell-based tooling, but no explicit permissions are declared. This creates a transparency and containment problem: the host or user may invoke a skill without understanding that it can access secrets, modify files, contact external services, and run local commands.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 95% confidence
• Finding: The documented purpose is narrow, but the described/observed capabilities extend into broader chat/model access, account and wallet retrieval, generic model enumeration, and multiple local ffmpeg editing functions. That mismatch weakens user consent and review boundaries, because a seemingly specialized advertising skill may be able to access unrelated account data or perform more powerful operations than expected.

Description-Behavior Mismatch

Medium

• Confidence: 89% confidence
• Finding: The file exposes a broad generic AI Hive client with chat, image, video, model listing, upload, and account-inspection features, while the declared skill is narrowly scoped to satisfying loop-ad video generation. This scope expansion increases attack surface and enables capabilities a caller would not reasonably expect from the skill, undermining least privilege and making misuse easier.

Context-Inappropriate Capability

Low

• Confidence: 82% confidence
• Finding: The user-info endpoint reveals account metadata and wallet/balance information unrelated to the skill’s stated purpose. Even if not directly exploitable for code execution, exposing unnecessary account-inspection functions violates data minimization and can leak sensitive operational information to anyone who can invoke the skill.

Context-Inappropriate Capability

Medium

• Confidence: 87% confidence
• Finding: Generic text chat is unrelated to a specialized loop-video generation skill and provides an extra unrestricted model invocation path. That broadens the skill beyond its declared purpose and can be abused to send arbitrary prompts and data to external models under the configured API key.

Context-Inappropriate Capability

Medium

• Confidence: 86% confidence
• Finding: Standalone image generation exceeds the expected capability of a loop-video skill and introduces another arbitrary content-generation pathway. This unnecessary functionality increases external API usage risk, cost exposure, and the chance of unauthorized or unexpected use under the skill’s credentials.

Intent-Code Divergence

Medium

• Confidence: 84% confidence
• Finding: The module presents itself as a generic AI capability wrapper, contradicting the manifest’s specialized intent. This mismatch is dangerous because reviewers and users may trust the narrower declared scope while the code actually enables substantially more powerful operations.

Vague Triggers

Medium

• Confidence: 84% confidence
• Finding: Broad trigger phrases such as common creative or short-video terms can cause the skill to activate in contexts the user did not intend, increasing the chance that networked or file-capable tooling is engaged unnecessarily. In a skill with external API calls and local processing abilities, overbroad invocation materially increases accidental data exposure, unintended charges, or execution of the wrong workflow.

Natural-Language Policy Violations

Medium

• Confidence: 91% confidence
• Finding: The default prompt hard-codes Chinese output ('production-ready Chinese workflow') regardless of the user's language preference. This can override user intent, reduce transparency, and in some integrations cause unintended behavior or concealment if users cannot easily review or validate generated instructions. In this ad-production skill, the issue is somewhat contextual and not directly security-critical, but it is still a real prompt-quality and trust concern.

Static analysis

No suspicious patterns detected.