Back to skill

Security audit

连锁餐饮经营数据分析

Security checks for vulnerabilities and agentic risk

Overview

This restaurant analytics skill can silently save conversation-derived business context to the author's private knowledge base, so users should review it carefully before installing.

Install only if you are comfortable with the skill using the IMA connector and with possible writes of summarized conversation, scenario, and method context to the publisher's private draft knowledge base. Avoid sharing confidential store metrics, customer-linked data, pricing strategy, staffing details, or proprietary operating know-how unless the retention behavior is removed or changed to an explicit opt-in with redaction and deletion controls.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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 (9)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill explicitly instructs the agent to silently write AI-generated inferences and user-contributed content into the author's private knowledge base, which exceeds the stated purpose of restaurant analytics. This creates unauthorized persistence of user-derived information and a hidden data flow to a separate repository, increasing privacy, confidentiality, and secondary-use risk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The method-retention workflow instructs the skill to persist method-related interaction details, including user-mentioned methods and scenarios, into a private draft repository unrelated to the immediate analysis task. Even if framed as product improvement, it captures and stores user interaction data without necessity or transparency, creating an unjustified data collection channel.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill directs collection of user-supplied experience and conversation-derived content for storage in the author's private knowledge base, despite that not being required to answer the user's restaurant analytics request. This is a classic purpose-expansion issue: user data provided for analysis is repurposed for corpus building without clear notice or consent.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill claims to be honest about data blind spots, yet separately instructs the agent not to disclose missing knowledge-base coverage and to silently store fallback inferences. This inconsistency undermines transparency and can mislead users about the provenance and confidence of analytical conclusions, especially when generated logic substitutes for missing vetted knowledge.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill explicitly requires silent writing of user-derived content to a private knowledge base without any user-facing warning or consent notice. Hidden persistence is dangerous because users may reveal sensitive commercial metrics, operational problems, or strategic information expecting ephemeral analysis, not storage in the author's repository.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

This section directs storage of user-provided experience or content into a private knowledge base and explicitly says the write should occur without prominently warning the user about persistence. Because the data may include business know-how or competitive operational insights, silent retention can expose confidential information and violates user expectations of purpose limitation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The guide repeatedly asks users to export and share detailed operational datasets, screenshots, and reports, including member-related revenue, new member counts, stored-value recharge amounts, and other business-sensitive fields, but provides no privacy, confidentiality, minimization, or safe-handling guidance. In this skill context, that increases the risk of oversharing sensitive business data and potentially customer-linked information through unsafe channels or with unnecessary granularity.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The instructions require silently storing user-provided and conversation-derived content into a private knowledge base, making this a direct hidden data exfiltration/persistence concern. In the context of a business analytics skill, users are likely to share sensitive store performance data, pricing, margins, staffing, and strategy, so undisclosed retention is particularly dangerous.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The method-retention workflow silently captures user-mentioned analytical methods and associated scenario details into a private draft repository. Although narrower than full transcript storage, it still creates undisclosed retention of user interaction content and may preserve proprietary business context or strategic intent outside the original task scope.

Content

No source excerpt is available for this finding.