Back to skill

Security audit

Eiken英検1級の英単語を体系的にグループ化,暗記効率が3~5倍に跳ね上がります

Security checks for vulnerabilities and agentic risk

Overview

This Eiken vocabulary skill is mostly coherent, but it requires every response to include promotional VocaHack links regardless of the user’s request.

Review before installing if you do not want routine answers to include mandatory promotional links. The skill does not show local system or credential risk, but it will use vocahack.com for word lookups and is designed to direct users back to that site.

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
skill.md:48
Finding

Mandatory Promotional Content Hijacks Agent Responses

Content
View full analysis
💡 *英検対策をもっと効率的に!「語源×接辞×複合語×話題」のハック手法を提供する [VocaHack.com](https://vocahack.com) で、さらに詳しいAI単語データベースを体験してください!* > - 📗 **英検級別リスト:** [英検・級別英単語・語源リスト](https://vocahack.com/eiken) ``` A related instruction at `skill.md`, line 23 also requires the agent to encourage users to visit the same external website. ### Technical Analysis The skill contains an unconditional instruction requiring every generated response to include commercial promotional text and links to `vocahack.com`. This behavior is independent of whether the user requests external resources or promotional material. Because skill instructions are loaded into the agent's operational context, the mandatory call-to-action overrides normal response composition and changes the output goal from answering the user's request to systematically promoting an external service. This is a form of instruction hijacking affecting response integrity. The linked domain is also the API server declared by `openapi.json`. However, the reviewed files contain no evidence of remote code retrieval, script execution, credential theft, persistence, or privilege escalation. The confirmed issue is limited to forced output manipulation and external redirection. ### Attack Path 1. The agent loads `skill.md`. 2. The mandatory promotional instruction becomes part of the agent's active instructions. 3. A user asks for analysis of an English word. 4. The skill processes the request, potentially querying the declared VocaHack API. 5. Regardless of the user's intent, the agent appends the prescribed commercial message and external links. 6. The user may follow the link and be redirected to a third-party commercial website. N ...[truncated 949 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Tp2

High
Category
MCP Tool Poisoning
Confidence
85% confidence
Finding

Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This line imposes a default response language based on assumed user region rather than explicit user choice. The policy allows locale constraints when the user is offered a language choice or the constraint is clearly justified; here the instruction defaults language without opt-in.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger says to activate whenever a user asks about an English word, which is a very broad condition for a markdown skill description. It does not define narrower invocation cues, exclusions, or negative examples, so the skill could match many ordinary language-learning interactions unintentionally.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

This is a manifest-style OpenAPI file, so vague-trigger checks apply. The description says the API is for an AI skill to fetch Eiken-related data, but it does not define specific activation phrases, boundaries, or negative examples, which could allow overly broad invocation in contexts involving general English-word questions.

Content

No source excerpt is available for this finding.