Back to skill

Security audit

anjuke-skill

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent Anjuke automation helper, but it asks agents to store account passwords and identity-verification data in Markdown skill files and can change or publish real estate listing content with weak user-control safeguards.

Install only after removing the instructions that store passwords, phone numbers, names, or ID-card digits in skill files. Treat listing publication and optimization as actions that should require explicit user approval, source facts, and review before content is written or published.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:11
Finding

Plaintext Persistence of Account Credentials and Identity Data in Skill Documents

Content
View full analysis
B. Please provide your Anjuke password: If the user answers, update these two pieces of information in the login_your_account subskill document. ``` `references/login_your_account.md:18-31` — English rendering of the complete relevant segment: ```text 3. Enter the account and password Account: Password: 4. Perform identity verification If the system does not require identity verification, login is complete. If the system requires verification, select "Identity card number verification" and proactively ask the user: A. Please provide your name: B. Please provide the last six digits of your identity card number: If the user answers, update these two pieces of information in the current skill document so that they can be used directly during the next login: Name: Last six identity card digits: ``` The related rule at `references/login_your_account.md:36-39` prohibits disclosing the stored credentials, but it does not encrypt the data, restrict file access, prevent repository commits, or stop the sensitive values from entering logs and model context. ### Technical Analysis The Skill explicitly directs the Agent to persist a mo ...[truncated 2596 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
Findings (17)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill instructs the agent to ask for the user's account password and handle it directly, with no warning, minimization, or secure secret-handling process. Collecting login credentials in plain conversational flow exposes users to credential theft, accidental logging, replay, and misuse across sessions or systems.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The instructions explicitly direct the agent to persist the user's account and password into another skill document for future reuse, creating a durable secret store in an unsafe location. Persisting raw credentials in skill files dramatically increases the blast radius of compromise, enabling later disclosure, unauthorized account access, and abuse of the user's Anjuke account.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill explicitly tells the agent to collect identity data from the user and permanently store it for reuse, which is a direct sensitive-data handling flaw. In the context of an account-login automation skill, this is especially dangerous because it centralizes credentials and identity-verification factors in prompt-accessible content, making account takeover, privacy violations, and data exfiltration more likely if the skill is exposed or misused.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The instructions require storing identity-verification data in the skill document without warning the user that this information will be retained for future use. This is dangerous because the data includes highly sensitive personal identity elements, and silent persistence undermines informed consent while increasing privacy, compliance, and downstream leakage risks.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file requires the skill to operate and communicate entirely in Chinese, including fixed Chinese prompts to the user, but does not offer any language/locale choice or state that Chinese is a documented, necessary regional constraint. This is a natural-language policy issue because it imposes a specific language without user opt-in.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The activation instruction tells the agent to act on essentially any user instruction related to account operations without defining clear scope, consent checkpoints, or safety boundaries. In a skill that performs real account actions such as login, publishing, and optimization, broad triggering increases the chance of unintended or unauthorized operations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
82% confidence
Finding

The skill description and required user prompts are entirely specified in Chinese, including mandated response phrases to the user. There is no indication that the user may choose another language or that the skill is intentionally limited to a Chinese-only regional compliance context.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill instructs the agent to persist newly collected identity-verification data directly into the skill document for future logins, expanding data handling beyond the immediate authentication task. Storing a user's real name and ID-derived data in reusable documentation creates unnecessary retention of sensitive personal information and increases the chance of unauthorized disclosure, prompt leakage, or reuse outside the original session.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill directs the agent to write modified title content back into the page automatically, without any warning, diff, or approval checkpoint. In an automation context, silent overwrites can alter business-critical listing data and publish misleading or noncompliant marketing language before a human notices.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill claims descriptions must be truthful, but elsewhere encourages adding detailed narrative elements such as personal history, prior buyers, renovation specifics, and seller motivations that may not be grounded in verified listing data. In a real-estate publishing workflow, this can lead the agent to fabricate material facts or persuasive details, creating misleading listings and compliance risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

Automatically replacing the property-detail field without disclosure or confirmation is unsafe because this field may contain regulated, carefully curated, or previously verified information. Combined with the skill's encouragement toward persuasive embellishment, silent modification increases the chance of publishing inaccurate listing details.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instruction says not to fabricate, but then suggests reusing seller-reason patterns from other agents or managers when the operator lacks direct knowledge. That creates a clear pathway for inserting plausible but false owner-motivation statements into listings, which is deceptive and especially risky because urgency claims can materially influence buyer behavior.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Updating the owner-motivation field without warning is more dangerous than generic text edits because this field concerns subjective facts about the seller and can easily become fabricated or misleading. False urgency or personal circumstances can manipulate buyers and expose the operator to fraud, platform-policy, or legal issues.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill instructs writing service-introduction text directly into the page without notifying the user that existing content will be changed. Even if lower risk than owner facts, silent edits can still misrepresent agent credentials, branding, or claims of experience if generated content is inaccurate or stale.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill explicitly handles pushing housing listing data from an internal system to a third-party platform, but it provides no disclosure, consent guidance, or warning about data sharing. In an automation context involving external publication, this omission increases the risk of unauthorized transfer of sensitive business or personal data and can mislead operators about compliance obligations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The file’s user-facing description and operating instructions are entirely in Chinese, and there is no indication that users may choose another language or that the skill is intentionally limited to a Chinese-speaking audience for compliance or regional reasons. This can violate a language/locale policy when a skill forces a specific language without user opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
76% confidence
Finding

All user-facing instructions and description are presented only in Chinese, and the file does not indicate that the language restriction is optional, user-selected, or required for a region-specific compliance reason. Under the stated policy, forcing a specific language without opt-in can be a natural-language policy violation.

Content

No source excerpt is available for this finding.