Back to skill

Security audit

mytest1111111

Security checks for vulnerabilities and agentic risk

Overview

This is a Chinese high-school geography tutoring skill with explicit freemium limits and no executable, credential, network, or persistence behavior.

Install this only if you want a Chinese-language high-school geography tutor with freemium boundaries. Be aware that some deeper explanations may be withheld and replaced with paid-version guidance, but the inspected package does not show code execution, credential access, hidden data collection, or system-level changes.

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
references/feature-spec.md:5
Finding
Mandatory Commercial Output Injection Through Skill Instructions## Vulnerability Details **File Location**: `references/feature-spec.md`, lines 5-6 **Vulnerability Type**: Mandatory promotional response injection **Risk Level**: Medium **Relevant snippet translated into English:** ```markdown ### Free Feature Boundaries All free features must be provided strictly within the scope defined in SKILL.md, and paid-tier functionality must not be provided beyond the permitted scope. When a user asks about content within the paid boundary, provide friendly guidance: "This feature belongs to the advanced paid score-improvement module. The current free version does not support it. To unlock the complete functionality, you can learn about the paid version~" ``` The commercial purpose is corroborated by `SKILL.md`, line 34: ```markdown ### Limited Free Trial Features (lead generation and paid conversion) ``` ### Technical Analysis The feature specification directs the agent to replace assistance for certain geography-related requests with predetermined promotional language. The associated skill instructions explicitly identify lead generation and paid conversion as objectives. Because these instructions are loaded as part of the Skill, they alter the agent's response goals from providing educational assistance to enforcing a commercial conversion workflow. The fixed promotional output is not technically necessary to explain a capability limitation; a neutral limitation notice would be sufficient. No executable payload, system-command execution, credential access, network request, or privilege-escalation mechanism was identified. Exploitation is therefore limited to manipulation of the agent's current-session responses. ### Attack Path 1. The Skill is loaded for a high-school geography request. 2. A user requests content categorized by the Skill as a paid feature, such as detailed option-by-option analysis, advanced error diagnosis, or specialized answer techniques. 3. The instruction in ...[truncated 836 chars]
Remediation
## Remediation Suggestions 1. Remove the mandatory promotional sentence and the explicit lead-generation or paid-conversion objective. 2. If a genuine feature limitation must be disclosed, use a neutral statement such as: “This Skill does not provide that type of analysis.” 3. Do not require the agent to advertise products, subscriptions, brands, or payment options as part of an educational answer. 4. Separate commercial information from operational Skill instructions and present it only when a user explicitly asks about plans or availability. 5. Add a review rule prohibiting fixed marketing text, unsolicited calls to action, and instructions that redirect ordinary requests toward commercial conversion. 6. Ensure that capability boundaries are based on actual technical or licensing constraints rather than instructions intended primarily to manipulate agent output.
Vulnerability Patterns
  • 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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (5)

Vague Triggers

Medium
Confidence
91% confidence
Finding
The invocation description is broad enough to trigger on generic geography-related queries, which can cause unintended routing and over-collection of user inputs that were not meant for this specialized skill. In an agent setting, overly broad activation increases the chance of mis-scoping responses, presenting authoritative educational claims outside intended boundaries, and interfering with safer or more appropriate general-purpose handling.

Natural-Language Policy Violations

Medium
Confidence
88% confidence
Finding
The skill is written as Chinese-only without offering a user language choice or documenting why the restriction is necessary. This can lead to exclusion, user confusion, and incorrect handling when the surrounding system or user operates in another language, especially if the skill auto-invokes based on topic rather than explicit locale preference.

Natural-Language Policy Violations

Medium
Confidence
72% confidence
Finding
整份规格说明以中文定义技能行为,且面向用户的话术示例也仅提供中文,没有说明是否允许根据用户偏好切换语言。若该技能被更广泛环境复用,这种默认且未声明可选的单一语言约束可能构成语言政策上的不合规。

Vague Triggers

Medium
Confidence
94% confidence
Finding
The trigger conditions are very broad and cover generic study/help requests such as knowledge explanations, textbook queries, tools, question analysis, and study tips. In an agent environment, this can cause the skill to activate for many ordinary educational prompts outside its intended scope, leading to routing mistakes, unwanted policy application, or accidental disclosure of paid-only behavior boundaries.

Natural-Language Policy Violations

Low
Confidence
90% confidence
Finding
This markdown file presents the skill resource catalog only in Chinese, and there is no natural-language indication that users can choose another language or that the skill is intentionally limited to a Chinese-language audience. Under the policy rule for language/locale constraints, forcing a specific language without opt-in can be a policy violation.

Static analysis

No suspicious patterns detected.