Security audit

한국 세금/절세 가이드

Security checks for vulnerabilities and agentic risk

Overview

This is a Korean tax guidance skill that is mostly coherent and disclosed, with some routing and connector scoping issues users should understand before relying on it.

Install this only if you want Korea-focused tax guidance. Treat outputs as general information, verify current rates and deadlines against official sources, avoid sharing more personal financial detail than needed, and be aware that broad Korean triggers may cause the skill to enter tax mode when a clarification would be better.

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

Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
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 (12)

Vague Triggers

Medium

Confidence: 92% confidence
Finding: The file mixes a wide trigger list with 10 different intents but does not specify clear invocation boundaries or precedence rules, making activation scope ambiguous. This can cause the orchestrator to select the skill for loosely related queries and increase the chance of incorrect tax guidance or unnecessary connector/tool invocation.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The file mixes a wide trigger list with 10 different intents but does not specify clear invocation boundaries or precedence rules, making activation scope ambiguous. This can cause the orchestrator to select the skill for loosely related queries and increase the chance of incorrect tax guidance or unnecessary connector/tool invocation.

Vague Triggers

Medium

Confidence: 95% confidence
Finding: The beginner_alias values are very generic Korean phrases such as '쉽게', '기초', and '모르겠어'. In a routing system, these can cause the tax skill to activate on non-tax conversations whenever the broader domain triggering is loose, leading to unintended collection of user context and incorrect tax guidance in unrelated chats. The skill context makes this somewhat more plausible because the domain is broad and the aliases are explicitly meant to force beginner routing.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The flagged trigger set for the year_end_settlement intent includes broad terms like '환급', '소득공제', and '세액공제', which are not unique to year-end settlement and may map unrelated tax or finance questions into the wrong intent. In this skill, misrouting is the main risk: users may receive incomplete or misleading procedural advice for the wrong tax workflow.

Vague Triggers

Medium

Confidence: 93% confidence
Finding: The trigger '절세' is highly generic and overlaps with many tax intents, including deductions, inheritance planning, property tax, and general tax advice. In a tax skill that provides actionable financial guidance, such ambiguity can reliably misroute users and surface unsuitable recommendations or omit important caveats.

Vague Triggers

Medium

Confidence: 89% confidence
Finding: The trigger '가족 간 돈 이체' is ambiguous because many ordinary personal-finance conversations use that phrase without any inheritance or gift-tax relevance. In this skill context, that increases the chance of unsolicited tax framing, potentially causing users to receive legal-tax implications they did not ask about or to disclose unnecessary family/asset details.

Vague Triggers

Medium

Confidence: 86% confidence
Finding: The trigger '4대보험' is broader than withholding tax and often refers to payroll, benefits, enrollment, or labor compliance rather than tax withholding itself. In this tax guide, that can misclassify employment-administration questions as tax questions and produce incorrect compliance guidance or omit non-tax distinctions that matter to the user.

Natural-Language Policy Violations

Medium

Confidence: 93% confidence
Finding: The playbook is entirely written in Korean and includes Korean-only required response/disclaimer text, with no indication that users may choose another language or locale. This is a natural-language policy concern because it imposes a specific language by default rather than documenting a justified locale restriction or offering opt-in.

Context-Inappropriate Capability

Medium

Confidence: 88% confidence
Finding: The manifest describes a tax guidance skill that routes tax intents and produces reports from legal/tax knowledge sources. However, the playbook explicitly invokes `bash skills/law-search/scripts/law_search.sh ...`, introducing shell execution capability rather than merely describing use of tax/legal references. Spawning shell commands is a materially broader capability than ordinary tax-answer generation.

Natural-Language Policy Violations

Medium

Confidence: 95% confidence
Finding: The entire skill specification, including headings, trigger phrases, and routing instructions, is written in Korean with no indication that users may choose another language or that the skill is intentionally restricted to Korean-language interactions. This creates a natural-language locale constraint that may violate policy requiring user opt-in or clear justification for language restrictions.

Natural-Language Policy Violations

Medium

Confidence: 93% confidence
Finding: The file title and all example output text are written in Korean, and line L123 indicates automatic beginner-mode terminology explanations, suggesting the skill's outputs are expected to be produced in Korean by default. There is no indication in this file that users may opt into another language or that the Korean-only behavior is a clearly documented region-specific requirement.

Natural-Language Policy Violations

Low

Confidence: 79% confidence
Finding: The natural-language description and examples are written exclusively in Korean, and the file does not state that users may choose another language or locale for responses. This can constitute a language/locale policy issue when a skill implicitly fixes output behavior to one language without opt-in or explicit justification.

Static analysis

No suspicious patterns detected.