鸡尾酒顾问 cocktail-advisor

• Outcome: Pass
• Latest audit: May 29, 2026 at 8:28 PM UTC
• Version:
  1.0.0

Overview

This is a cocktail recommendation skill with disclosed local preference memory and no evidence of hidden, destructive, or credential-seeking behavior.

Install only if you are comfortable with a local workspace note remembering cocktail recommendations and feedback. Keep feedback brief, avoid sensitive personal details, delete the memory file if you do not want retention, and use alcohol recipes responsibly.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration

Findings (5)

Description-Behavior Mismatch

Medium

• Confidence: 95% confidence
• Finding: The skill instructs itself to read and update a persistent workspace file containing user preference history, which expands behavior beyond transient cocktail recommendation into retention of personal preference data. Even though the data seems low sensitivity, persistence creates privacy, consent, and data lifecycle risks if users are not informed or if the file accumulates broader free-form feedback over time.

Context-Inappropriate Capability

Medium

• Confidence: 91% confidence
• Finding: A cocktail recommendation skill can function without persistent file access, so requiring reads and writes to workspace memory is not clearly necessary for the stated purpose. Unnecessary persistent access increases the attack surface for privacy leakage, cross-session profiling, and accidental storage of more personal information than intended.

Vague Triggers

Medium

• Confidence: 84% confidence
• Finding: The activation conditions are broad enough to overlap with ordinary conversation about drinks, ingredients, or scenarios, which can cause the skill to trigger unexpectedly. Over-broad triggering is dangerous in context because this skill also includes persistent preference tracking, so accidental activation can lead to unanticipated data handling and unsolicited behavioral steering.

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: The skill directs storage of user preference data in a persistent memory file but does not require notifying the user or obtaining consent. This creates a clear transparency and privacy problem because users may reasonably expect conversational recommendations to be ephemeral, not logged across sessions.

Ssd 3

Medium

• Confidence: 94% confidence
• Finding: The preference log stores free-form user feedback in persistent workspace memory without clear bounds on what may be recorded. Because feedback fields can capture arbitrary text, the file may gradually collect sensitive personal details, habits, or other unrelated information far beyond what is needed to recommend cocktails.

VirusTotal

66/66 vendors flagged this skill as clean.