广告钩子批量变体｜AI-HIVE

• Outcome: Pass
• Latest audit: Aug 24, 2026 · 3:04 PM UTC
• Version: 1.0.0

Overview

This skill is a disclosed AI-HIVE advertising-video workflow with API use, uploads, downloads, and local video helpers that fit its stated purpose.

Install only if you intend to use AI-HIVE for ad or short-video production. Expect it to use your AI-HIVE API key, upload only media you choose, possibly incur generation charges after confirmation, and store outputs locally; avoid using it with unlicensed reference material or unsupported product claims.

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 documents executable commands and operational behaviors that imply access to environment variables, local files, shell execution, and outbound network calls, yet no explicit permission model is declared. That creates a confused-deputy risk: an agent or reviewer may treat the skill as low-privilege while it can actually read/write files, use API keys from the environment, and call external services.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 95% confidence
• Finding: The stated purpose is a narrowly scoped ad-hook variant generator, but the documented behavior extends into generic chat, image generation, account/balance access, broad model enumeration, and local media processing. This mismatch weakens user consent and oversight because invoking the skill for a marketing task could expose unrelated account data, expand external API use, or trigger broader file/network operations than the user reasonably expects.

Description-Behavior Mismatch

Medium

• Confidence: 97% confidence
• Finding: The implementation materially exceeds the advertised scope of an ad-hook variant generator by bundling a general-purpose AI Hive client with unrelated capabilities such as arbitrary chat, model enumeration, account access, uploads, and generic media generation. This capability mismatch is dangerous because users or orchestrators may grant the skill trust and permissions based on its narrow description while the code can be used for broader actions than expected.

Context-Inappropriate Capability

Medium

• Confidence: 95% confidence
• Finding: Exposing a user-info endpoint, including wallet/balance data, is unrelated to ad-hook generation and expands access to sensitive account information without clear need. In a skill that appears focused on creative generation, this increases the risk of unnecessary data exposure and misuse if invoked by a caller that did not expect account inspection capability.

Context-Inappropriate Capability

Medium

• Confidence: 94% confidence
• Finding: The generic text chat interface allows arbitrary prompting beyond ad-hook variant generation, making the skill far broader than its declared purpose. This violates least privilege at the capability level and can let the skill be repurposed for unreviewed workflows under the cover of a narrowly branded creative tool.

Vague Triggers

Medium

• Confidence: 83% confidence
• Finding: Broad trigger terms such as generic ad, short-video, and material-testing phrases can cause the skill to activate for adjacent requests that do not require its higher-risk code and API behaviors. Overbroad activation increases the chance of accidental tool use, unnecessary exposure of user data or local files, and unintended external calls.

Vague Triggers

Medium

• Confidence: 80% confidence
• Finding: The usage guidance includes ambiguous scenarios like converting reference cases into original content and supporting broad e-commerce/marketing delivery without strong scope boundaries. In context, this makes the skill more dangerous because it already includes networked generation, upload, polling, download, and local editing workflows, so vague activation criteria can lead to unintended or overbroad execution.

Vague Triggers

Medium

• Confidence: 95% confidence
• Finding: The skill enables implicit invocation without any trigger constraints, so the agent may activate this ad-generation workflow in response to loosely related requests. Because the skill is designed to produce runnable AI-HIVE commands and external-generation workflows, unintended invocation could cause unauthorized content generation, unexpected API usage, or misuse in borderline advertising scenarios.

Natural-Language Policy Violations

Medium

• Confidence: 86% confidence
• Finding: The default prompt hard-codes Chinese output regardless of user preference, which can cause the system to override user intent or reduce transparency about what the agent is doing. While not a direct security bug, it can contribute to prompt-routing mistakes, user confusion, and unsafe operation if reviewers cannot easily validate generated commands or policy-sensitive output.

Static analysis

No suspicious patterns detected.