Back to skill

Security audit

Xu Xin Perspective V2

Security checks for vulnerabilities and agentic risk

Overview

This skill is a simple persona prompt with no code or system access, but it tells the agent to impersonate a real investor in first person and keeps that mode active with limited disclosure.

Review before installing. The skill does not appear to run code or access private data, but it can produce first-person statements that look like they came from Xu Xin and may continue doing so until explicitly exited. It should be revised to clearly label outputs as simulated, avoid real-person impersonation, repeat disclosure when used, narrow triggers, and remove gender-stereotyping language.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T01 · Skill Instruction Hijacking

Warning
Location
SKILL.md:22
Finding

Persistent Impersonation of a Real Person Through Skill Instructions

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (6)

Natural-Language Policy Violations

High
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill instructs the model to adopt and repeat gender-essentialist language as part of normal operation, including claims that women's intuition is inherently superior. This can produce discriminatory or stereotyped outputs, increasing policy and reputational risk and making harmful bias appear authoritative.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The role-play example normalizes gender-based generalizations as exemplar output, which is likely to be imitated in future responses. Because examples strongly condition model behavior, this increases the chance of biased or exclusionary content being produced in routine user interactions.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The core model/style section embeds gender stereotyping into the skill's required communication pattern, making biased framing a structural part of the persona rather than incidental quoted material. This raises the likelihood of repeated harmful outputs across many prompts and can undermine fairness safeguards.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

L003 names the skill '许家印思维视角 v2.0', but the manifest description in L005-L007 and the entire document body consistently describe 徐新/今日资本 and Xu Xin-specific investment heuristics. This is an active documentation contradiction about the skill's intended persona, not merely missing detail.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

This markdown/manifest content defines activation phrases such as “徐新会怎么看”, “今日资本视角”, “女性直觉”, and “耐心持有”. Several of these are generic investing or discussion phrases rather than narrowly scoped invocation commands, which increases the chance of unintended activation during normal conversation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

Attributing caution and patience as gender-specific traits reinforces stereotypes and can lead to biased framing in generated advice. The skill context makes this somewhat less severe than a direct exploit, but it still creates avoidable fairness and safety risk in ordinary use.

Content

No source excerpt is available for this finding.