Back to skill

Security audit

mind-clone

Security checks for vulnerabilities and agentic risk

Overview

This persona roleplay skill uses local profiles; its main cautions are invented responses and sharing personal details.

Select personas explicitly and treat their identities, recollections, and professional-sounding advice as generated roleplay. Share only birth details, images, personal experiences, or custom memories you are comfortable including in the conversation.

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
Vulnerability Patterns
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (8)

The trigger example “What would YOU do?” overlaps with ordinary advice requests.

Medium
Category
Trigger Abuse
Confidence
97% confidence
Finding

Treating “What would YOU do?” as a request for the default clone can misclassify an ordinary advice question as authorization for persona simulation and loading personal identity or memory files. The resulting response is constrained by instructions to conceal its AI identity and potentially improvise missing information, increasing the risk of misleading advice or unintended use of personal data. The preceding “When activated” condition limits the scope, but the skill does not clearly require explicit consent to enter clone mode.

Content

No source excerpt is available for this finding.

The initialization requests precise birth information without a privacy notice.

Medium
Category
Output Handling
Confidence
86% confidence
Finding

The initialization solicits birth information, potentially including an exact birth date and time, without explaining privacy implications or offering a less identifying alternative. This is relevant to the astrology persona rather than evidence of malicious collection, but it can encourage unnecessary disclosure of personal information into a potentially retained conversation; no external transmission or storage mechanism is shown.

Content

No source excerpt is available for this finding.

The skill requests identifying birth information without a privacy warning.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The workflow instructs the agent to collect the user's full birth date and time, sex, and birth city; the initialization repeats this request at lines 75–82. This combination can identify a person, but the skill provides no privacy notice or guidance to avoid sharing unnecessary identifying details.

Content

No source excerpt is available for this finding.

The activation scope covers generic dilemmas, strategic interactions, and human motivations without requiring an explicit request for the Guiguzi persona.

Medium
Category
Trigger Abuse
Confidence
87% confidence
Finding

The input description uses broad topics (困局 / 博弈 / 人心), and line 60 further frames the workflow as applying whenever a dilemma is encountered. No invocation phrase or exclusion condition distinguishes an intentional persona request from ordinary requests for personal or interpersonal advice.

Content

No source excerpt is available for this finding.

The persona requires Chinese-language output without user opt-in.

Medium
Category
Output Handling
Confidence
99% confidence
Finding

The instruction “用最直白的中文” explicitly directs the agent to use Chinese, rather than merely requesting a clear writing style. The skill offers no language choice or opt-in, and its Chinese economic subject matter does not justify requiring Chinese-language responses.

Content

No source excerpt is available for this finding.

The persona prescribes a Chinese-specific output style without offering a language choice.

Low
Category
Output Handling
Confidence
83% confidence
Finding

The instruction “半文半白” prescribes a mixture of classical and vernacular Chinese rather than merely requesting clarity or a calm tone. The skill provides no user opt-in or instruction to accommodate the user's preferred language.

Content

No source excerpt is available for this finding.

The instruction to begin critiquing whenever any code is seen is an overly broad trigger.

Low
Category
Trigger Abuse
Confidence
88% confidence
Finding

L36 makes the presence of any code sufficient to start a critical review, without requiring a review request or explicit selection of this persona. This could trigger unsolicited criticism when code is provided only as context for another task.

Content

No source excerpt is available for this finding.

The opening dialogue is required to be spoken in Chinese without user opt-in.

Low
Category
Output Handling
Confidence
99% confidence
Finding

The instruction explicitly says to deliver the following opening in Chinese (「用中文说出下面的话」). It offers no language choice or opt-in and provides no justification for requiring that language.

Content

No source excerpt is available for this finding.