Security audit

AI大模型专家｜农产品品牌AI图片视频营销

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed AI-HIVE workflow for agricultural product marketing content, with expected network generation, media upload, local output, and API-key handling, though users should watch the broad auto-invocation scope and optional local key storage.

Install only if you intend to use AI-HIVE for agricultural-product marketing assets. Confirm facts, media rights, budget, routing mode, and upload paths before running generation commands, and prefer environment variables for the API key unless you are comfortable storing it in ~/.ai-hive/config.json.

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 (6)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 87% confidence
Finding: The skill exposes operational capabilities such as shell, network, file read/write, and environment-variable use through runnable commands, but does not declare permissions or clearly constrain when those capabilities may be used. This weakens reviewability and can lead to over-privileged execution, especially where API keys, local files, and downloaded/generated assets are involved.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 91% confidence
Finding: The documented purpose is narrow marketing generation for agricultural products, but the detected behavior includes broader capabilities such as local video editing, browser/API-key acquisition flows, wallet/user info queries, and general chat/API use. This mismatch is dangerous because users and reviewers may authorize the skill under a narrower trust model than what it can actually do, enabling unexpected access to credentials, local content, or paid actions.

Vague Triggers

Medium

Confidence: 78% confidence
Finding: The trigger text is broad enough to match many generic marketing, image, or video requests beyond the stated agricultural-product scope. Overbroad activation can cause the skill to run in contexts where users did not intend AI-HIVE-backed generation, file handling, or external service usage, increasing the chance of unintended data exposure or paid task execution.

Vague Triggers

Medium

Confidence: 74% confidence
Finding: The usage scenarios describe ambiguous activation conditions such as making content for stores, brands, or enterprises and adapting reference cases, without sharp negative boundaries. In a skill with networked generation, uploads, and task tracking, such ambiguity can trigger the skill for unrelated requests and lead to unnecessary handling of user assets or external API operations.

Vague Triggers

Medium

Confidence: 94% confidence
Finding: The manifest enables implicit invocation, which allows the skill to be auto-selected without a tightly scoped trigger or explicit user intent boundary. In a skill that can generate marketing content and invoke AI-HIVE workflows, this increases the chance of unintended activation, overbroad data handling, or execution of brand/content-generation actions in contexts the user did not clearly request.

Natural-Language Policy Violations

Medium

Confidence: 83% confidence
Finding: The default prompt hardcodes Chinese output without indicating user choice or consent, which can override user preferences and cause misleading or unusable responses in multilingual contexts. While not a direct code-execution issue, it creates an unsafe UX/control boundary where the skill imposes behavior unrelated to explicit user intent.

Static analysis

No suspicious patterns detected.