Security audit

内容验证网关专业版

Security checks for vulnerabilities and agentic risk

Overview

This content moderation skill is not plainly malicious, but it needs review because it can run a web API, handle sensitive personal data, and send or expose moderation results without enough safeguards.

Install only after reviewing the REST, webhook, callback, and audit-log behavior. Use authentication, HTTPS, trusted callback destinations, restricted bind addresses, and a clear retention policy before using it with real user content or personal data. Clarify whether any content is sent to an external LLM or service.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Intent-Code Divergence

Medium

Confidence: 92% confidence
Finding: The documentation gives conflicting trust signals: it claims the semantic analysis is a local engine requiring no external API key, while also declaring an external LLM API as required. This can mislead users about where content is processed and whether sensitive data may leave the local environment, creating data handling and compliance risk.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The activation language is broad enough to match generic security, compliance, quality, or encryption-related requests, which can cause the skill to trigger outside the user's intended scope. In an agent context with read and exec tools, overbroad routing increases the chance of unnecessary command execution, data access, or unintended handling of sensitive content.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The skill advertises Webhook alerting and REST API integration but does not clearly warn that content, metadata, or moderation results may be transmitted to external services. Users may assume analysis is self-contained and inadvertently expose sensitive material to third-party endpoints.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: Documenting a callback_url parameter without a prominent warning obscures that verification outputs may be posted to an external endpoint. That creates a straightforward exfiltration path for content-derived results, identifiers, or metadata if users supply or inherit an untrusted callback destination.

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: The examples process highly sensitive personal data such as ID numbers, SSNs, phone numbers, and expose audit-trail and verification endpoints, yet the document shows no privacy notice, retention policy, access control discussion, or warning about logging user content. In this skill context, that omission is meaningful because the examples normalize collecting, analyzing, and returning sensitive content and audit metadata, which can lead downstream users to deploy privacy-impacting behavior without safeguards.

Static analysis

No suspicious patterns detected.