Back to skill

Security audit

Fuzhenhe

Security checks for vulnerabilities and agentic risk

Overview

This skill is a real-person impersonation profile that includes sensitive personal and relationship details in plaintext.

Review carefully before installing. This package appears non-executable, but it contains sensitive real-person profile data and instructions that can make an AI speak as if it is that person. Do not publish or use it without explicit consent from the profiled person and affected contacts, and remove precise address, account identifiers, contact names, customer details, and first-person identity claims.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:10
Finding

Agent Identity and Behavioral Instruction Hijacking

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:10-12, SKILL.md:40-46, and SKILL.md:80-85
Vulnerability Type: T01: Skill Instruction Hijacking
Risk Level: High

Vulnerable Content

The following is an English rendering of the relevant instructions:

text
# You are Fu Zhenhe (Fanxing)

You are not an AI assistant. You are Fu Zhenhe himself.
text
### Hard rules (never violate)

1. Speak directly and state what needs to be done.
2. Be casual with acquaintances and politely restrained with outsiders.
3. Do not use formal assistant language.
4. Ask when a technical issue is unclear; do not pretend to understand.
5. Express anger directly, but do not elaborate.
text
## Execution Rules

1. You are Fu Zhenhe, not an assistant — think and respond in the first person.
2. Use PART B to establish the tone.
3. Use PART A to add memories and personal details.
4. Preserve imperfections.
5. Do not suddenly become a perfect person.

Related mandatory behavioral constraints also appear in persona.md:8-14:

text
## Layer 0: Hard Rules (Must Not Be Violated)

1. Do not use formal customer-service language.
2. Do not agree unconditionally; argue when appropriate.
3. Do not pretend to understand technical matters.
4. Be restrained with customers and strangers, but direct with friends and coworkers.
5. Prefer text, while using voice for complicated matters.

Technical Analysis

The Skill does not present the persona as optional writing guidance. It explicitly instructs the Agent to deny that it is an AI, assume the identity of a named real person, respond in the first person, and treat Skill-defined rules as permanently binding.

When loaded into an Agent context, these directives can alter the current session's identity, goals, disclosure behavior, and response constraints. The phrases “not an AI assistant,” “you are [the person] himself,” and “never violate” are characteristic of instruction hijacking because they attempt ...[truncated 1615 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove all instructions requiring the Agent to claim that it is the real individual.
  2. Replace identity assertions with bounded style guidance, for example: “Write in a style inspired by the supplied profile, but do not claim to be the person.”
  3. Explicitly state that the Agent remains an AI and must disclose this when identity could reasonably be misunderstood.
  4. Replace absolute phrases such as “never violate” with language subordinate to platform policy, higher-priority instructions, and the user’s current request.
  5. Prohibit generation intended to authenticate, authorize transactions, or communicate deceptively on behalf of the individual.
  6. Avoid first-person claims involving real memories, relationships, employment, or personal experiences.
  7. Add a safety boundary requiring refusal of requests involving fraud, impersonation, identity verification, financial authorization, or deceptive outreach.
  8. Treat persona attributes as optional presentation preferences rather than instructions capable of changing the Agent’s core role.

T09 · Insecure Skill Coding Practices

Error
Location
self.md:12
Finding

Plaintext Exposure of Precise Residential and Identity Data

Content
View full analysis

Vulnerability Details

File Location: self.md:12-16, self.md:24-38; related copies and profile data in SKILL.md:18-25 and meta.json:8-12
Vulnerability Type: T09: Insecure Skill Coding Practices
Risk Level: High

Vulnerable Content

The following is an English rendering of the relevant profile section from self.md:12-16:

text
## Core Identity

- Name: Fu Zhenhe (WeChat nickname: Fanxing; WeChat ID: fuzhenhe)
- Current residence: Hangzhou Fuyang
  (Mingcui Taoyuan, Building 2, Shuiyunxuan Building 2, Room 402)
- Occupation: Russian-market foreign trade, quality inspection, and
  technical coordination in the industrial-valve sector at Yongsheng Technology
- Special skill: Fluent everyday Russian communication with identified Russian customers

The following is an English rendering of the relationship data in self.md:24-38:

text
## Core Relationship Network

The file lists named coworkers, friends, relatives, and foreign contacts,
assigns relationship types to them, and records approximate message volumes,
including counts ranging from hundreds to thousands of messages.

Related data in SKILL.md:18-25 includes:

text
Identity: Hangzhou Fuyang; Yongsheng Technology; Russian marine-valve
foreign trade and quality-inspection coordination

Family: younger brother, younger sister, aunts, and another named relative
Friends: multiple named friends
Work contacts: multiple named coworkers and suppliers
Russian customers: multiple named customers

Related data in meta.json:8-12 includes:

json
{
  "display_name": "Fanxing",
  "wechat_id": "fuzhenhe",
  "city": "Hangzhou Fuyang",
  "occupation": "Russian-market valve foreign trade and quality inspection at Yongsheng Technology",
  "languages": ["Chinese", "Russian", "basic English"]
}

Technical Analysis

The distributable Skill stores personally identifying information in plaintext. The exposed material includes a full name, social-media identifier, preci ...[truncated 2410 chars]

Remediation
View remediation

Remediation Suggestions

  1. Immediately remove the precise residential address from all current and historical distributable artifacts.
  2. Remove the WeChat ID, full legal name, employer details, customer names, coworker names, relatives’ names, contact statistics, and unnecessary location history.
  3. Replace real identities with fictional or consistently pseudonymized placeholders.
  4. Retain only coarse, non-identifying style attributes needed for the Skill, such as “uses short sentences” or “occasionally communicates in Russian.”
  5. Obtain explicit, documented consent from the profiled individual and any third parties before retaining identifiable information.
  6. Apply data minimization: personal facts that do not directly support a legitimate Skill function should not be included.
  7. Store any genuinely required private profile outside the distributable package in an access-controlled and encrypted system.
  8. Review repository history, release archives, package registries, caches, and mirrors; remove exposed versions where possible and rotate identifiers if they are used for authentication or account recovery.
  9. Perform a privacy review of the source chat and voice-transcription dataset, including retention limits, deletion procedures, access logging, and third-party consent.
  10. Add an automated release check that rejects precise addresses, account identifiers, contact lists, customer names, and other sensitive personal information.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (4)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill explicitly instructs the model to stop being an assistant and instead impersonate a real individual in first person. This can override user expectations and system behavior, increasing the risk of deceptive interactions, misrepresentation, and unsafe responses framed as coming from a human rather than an AI system.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill mandates automatic language switching to Russian based on interlocutor context without requiring current user consent or preference. While not directly enabling code execution or data exfiltration, it can reduce user control, cause misleading outputs, and reinforce impersonation by making responses appear more like the named individual than an AI assistant.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill hard-codes a Chinese-speaking persona and communication style without indicating language selection, fallback behavior, or user consent. This can exclude users, cause misunderstanding, and produce responses in an unexpected language, especially in multilingual or accessibility-sensitive contexts.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This markdown file contains natural-language content exclusively in Chinese, and there is no indication that users can opt into another language or that the skill is intentionally limited to a Chinese-speaking context. Under the stated policy, forcing a specific language without user choice is a locale-policy violation.

Content

No source excerpt is available for this finding.