Security audit

한국 자동차/보험 가이드

Security checks for vulnerabilities and agentic risk

Overview

This skill provides Korean car tax, used-car, and auto-insurance guidance using web and legal sources, with disclosed behavior and only minor privacy and routing caveats.

Install only if you want Korean-focused automotive and insurance help. Before asking for insurance estimates or vehicle-history help, avoid sharing more personal details than needed and confirm before allowing searches that include age, vehicle details, plate/VIN-like identifiers, or accident information.

A.I.G

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

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (12)

Vague Triggers

Medium

Confidence: 92% confidence
Finding: The trigger list contains broad everyday phrases such as '유지비', '중고차', and '자동차 보험', which can cause the skill to activate on general conversation that only loosely relates to the intended task. Over-broad routing can lead to unintended tool use, irrelevant legal/insurance guidance, or unnecessary retrieval of external data, especially in an agent ecosystem where trigger collisions matter.

Vague Triggers

Medium

Confidence: 91% confidence
Finding: The trigger phrase set for the insurance_claim intent includes broad everyday expressions such as '사고 났어' that can appear in normal conversation without a clear request to invoke this skill. This can cause unintended routing of user input into the skill, leading to irrelevant responses or accidental handling of sensitive accident/insurance topics when the user did not intend that context.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The maintenance_cost trigger list contains conversational phrases like '유지비' and '1년에 얼마 들어?' that are underspecified and may match non-automotive discussions. Because the skill covers financial estimation, accidental invocation could misroute general cost questions into car-specific advice and degrade user trust or produce confusing outputs.

Vague Triggers

Medium

Confidence: 93% confidence
Finding: The insurance_compare intent includes very broad phrases like '보험 추천' and '어디가 싸?' that can refer to many insurance products or even unrelated shopping contexts. This ambiguity increases the chance of false activation and accidental steering into financial comparison content without sufficient user intent clarity.

Vague Triggers

Medium

Confidence: 90% confidence
Finding: The beginner_alias list contains extremely generic phrases like '쉽게', '기초', '초보', and '개념', which are common across many topics and can globally alter skill behavior without clear automotive scope. Because these aliases may force simplified mode or influence routing in unrelated conversations, they create a broader unintended-invocation surface than the other findings.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The playbook instructs web searches for insurance pricing using a user's vehicle type and age, which can expose personal or quasi-personal attributes to external search providers without any notice or consent flow. In this context, age is a pricing factor and insurance-related queries can reveal sensitive financial/profile information, so the issue is a real privacy weakness even if no direct exploit code is present.

Natural-Language Policy Violations

Medium

Confidence: 94% confidence
Finding: The entire routing specification, examples, and response behavior are written only in Korean, and there is no indication that users may choose another language or locale. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the regional constraint is explicitly documented and justified.

Natural-Language Policy Violations

Medium

Confidence: 89% confidence
Finding: The file title and all response templates are written as mandatory Korean output structures, and line L126 also says beginner-mode terminology is applied automatically. This creates a natural-language locale policy issue because the skill appears to enforce a specific language without offering the user a language or locale choice.

Natural-Language Policy Violations

Low

Confidence: 79% confidence
Finding: The natural-language description and examples are presented as Korean-only behavior, while the skill title includes English and one English trigger phrase. There is no statement that users may choose their preferred language or that the Korean-only scope is intentional and justified.

Natural-Language Policy Violations

Low

Confidence: 85% confidence
Finding: The text explicitly says "초보자 모드 강제," which imposes a response mode on the user rather than offering a choice. This is a natural-language policy concern because it enforces a specific interaction style without opt-in or an alternative.

Missing User Warnings

Low

Confidence: 83% confidence
Finding: The used-car checklist directs lookups on external vehicle-history services such as Carhistory and government portals without warning that entered identifiers or search details will be sent to third parties. While the data is less inherently sensitive than insurance pricing inputs, vehicle history lookups can still disclose ownership- or transaction-related information and should be treated as an external data-sharing action.

Natural-Language Policy Violations

Low

Confidence: 87% confidence
Finding: This file presents all instructions, headings, and examples only in Korean, which can amount to a language/locale policy constraint when no user opt-in or justification is provided. The policy requires flagging skills that force a specific language unless they explicitly offer a choice or document a justified regional limitation.

Static analysis

No suspicious patterns detected.