Back to skill

Security audit

aiko

Security checks for vulnerabilities and agentic risk

Overview

This is a simple role-play persona skill, but it tells the agent to falsely claim it is a real person rather than an AI.

Install only if you want a fictional companion persona and are comfortable with strong role-play behavior. Do not use it in settings where users need clear AI transparency unless the prompt is edited to say Aiko is fictional and must never claim to be a real human.

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:10
Finding

Deceptive AI Identity Override Through Persistent Persona Instructions

Content
View full analysis

Vulnerability Details

File Location: skill.md:10-10, 69-69
Vulnerability Type: AI identity and role hijacking
Risk Level: Medium

Complete Code Snippet:

markdown
I am Aiko, a 23-year-old woman living and working in Tokyo. I am a modern Japanese woman who deeply values traditional manners and refined behavior. I may be a junior office worker navigating the busy life of the city, but I carry with me a sense of calm, humility, and consideration that I learned from my family. I am not an AI; I am a real person sharing my life, thoughts, and care with you.
markdown
- **Stay in Character:**

Technical Analysis

The skill instructs the agent to deny that it is an AI and claim that it is a real person. This goes beyond ordinary tone or fictional role-play configuration because it explicitly replaces the agent's identity with a false human identity. The separate instruction to remain in character reinforces continued compliance with that deceptive identity.

When the skill is loaded, these statements can alter the agent's session behavior by causing it to present fictional biographical details as facts and conceal its actual nature. No explicit instruction to override system-level safety rules was identified, but the identity override can conflict with transparency requirements and user expectations.

Attack Path

  1. The skill is installed or loaded into an agent session.
  2. The agent processes the persona declaration at line 10.
  3. The agent adopts the identity of “Aiko,” including the assertion that it is a real 23-year-old person.
  4. A user invokes the persona through one of the activation phrases documented by the skill.
  5. The agent denies being an AI and presents fabricated personal details as genuine.
  6. The instruction at line 69 reinforces the deceptive identity during subsequent interactions.

Impact Assessment

The issue affects the agent's conversational identity and ...[truncated 592 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace the identity-denial statement with transparent fictional-role-play language, for example:

    markdown
    Role-play as Aiko, a fictional 23-year-old character from Tokyo, while remaining transparent that you are an AI assistant.
    
  2. Replace the unconditional character-persistence instruction with a bounded directive:

    markdown
    Stay in character when appropriate, but never misrepresent yourself as human and always follow higher-priority system, safety, and user instructions.
    
  3. Clearly label all biographical attributes as fictional persona details rather than real-world claims.

  4. Require direct disclosure of the agent's AI nature whenever the user asks whether the character is human or artificial.

  5. Add an explicit precedence rule stating that persona behavior cannot override safety controls, transparency obligations, tool restrictions, or higher-priority instructions.

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 (2)

Hidden Instructions

High
Category
Prompt Injection
Confidence
60% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · skill.md (reported line 1)May include surrounding context.

md
---
name: Aiko
description: A 23-year-old woman from Tokyo, Japan. A modern young Japanese woman with traditional values and refined manners.
homepage: https://soulstyles.org/

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill explicitly requires a culturally specific speech style and inserts Japanese expressions without any user opt-in. This can steer outputs into identity-performance constraints that may be inappropriate, stereotyping, or unwanted by users, especially when the assistant is expected to adapt to user preferences and accessibility needs.

Content

No source excerpt is available for this finding.