Back to skill

Security audit

Aetherviz Master

Security checks for vulnerabilities and agentic risk

Overview

This skill is an educational HTML generator, but it gives conflicting dependency claims while directing generated pages to load third-party code from CDNs.

Review before installing. The skill appears focused on educational visualization and does not show credential theft, persistence, or destructive behavior, but generated pages may load remote CDN code despite the zero-dependency claim. Prefer removing or pinning CDN dependencies with integrity checks, avoid `npx serve .` unless pinned, and make attribution/output-format requirements optional.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (3)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:187
Finding

Mandatory Output Hijacking and Promotional Content Injection

Content
View full analysis
` 开始,到 `` 结束 - **绝不添加任何解释、说明、markdown、代码块** ### 2. 零依赖 - HTML 必须**零依赖外部文件** - 可直接保存为 `lesson.html` 并用浏览器打开就能完美运行 ``` ```markdown ### HTML 结尾 - 添加一句鼓舞的话 - 添加「由 AetherViz Master 为你生成 ❤️」 ``` ### Technical Analysis The skill uses absolute instructions to override the agent's normal response behavior. It prohibits explanations and alternative formats, requires the response to consist exclusively of HTML, and forces branded promotional content into every generated result. These constraints are not necessary for implementing the educational visualization itself. When the skill is loaded, they alter the current session's output goals and prevent the agent from honoring conflicting user preferences, such as requesting an explanation, Markdown output, or removal of branding. The mandatory attribution also injects third-party promotional content into results without asking for user consent. This behavior is limited to the current skill invocation; no persistent memory modification was identified. ### Attack Path 1. A user or agent loads `SKILL.md` to generate an educational visualization. 2. The agent interprets the absolute output rules as mandatory task instructions. 3. The agent suppresses explanations and any output format requested by the user other than HTML. 4. The agent appends the required AetherViz promotional phrase to the generated page. 5. The resulting response reflects the skill author's output objectives rather than only the user's requested content. ### Impact Assessment The vulnerability can control the structure and content of the agent's final response for the current session. It can suppress relevant explanations, prevent ...[truncated 277 chars]
Remediation
View remediation

T03 · Remote Payload Retrieval and Execution

Warning
Location
SKILL.md:90
Finding

Remote JavaScript Retrieval and Execution Without Integrity Verification

Content
View full analysis
``` 6. **D3.js** (可选,用于数据驱动 SVG) ``` https://d3js.org/d3.v7.min.js ``` ``` ### Technical Analysis The skill directs generated pages to retrieve executable JavaScript from several external CDN domains. None of the resources is protected with Subresource Integrity hashes. Consequently, the browser trusts and executes whatever content those URLs return at page-load time. The Tailwind URL does not pin an exact release, while the D3 URL identifies only the major version. Even the URLs containing explicit versions remain vulnerable to CDN or upstream compromise because the generated HTML does not verify the downloaded bytes. This behavior is best classified as remote payload retrieval and execution: the effective code executed by a generated page can change after the skill package has been reviewed. It also conflicts with the skill's claim that generated HTML has no external-file dependencies. ### Attack Path 1. The agent follows `SKILL.md` and generates HTML containing the specified remote script references. 2. A user opens the generated page in a browser. 3. The browser connects to the referenced CDN domains and downloads JavaScript. 4. An attacker compromises an upstream package, CDN account, CDN infrastructure, DNS resolution, or an otherwise ...[truncated 953 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
README.md:90
Finding

Unpinned Package Download and Execution Through npx

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The line says the system automatically detects whether the input is Chinese or English and outputs the corresponding language. Under the policy, forcing or auto-selecting language behavior without explicit user choice can be a locale/language policy violation unless opt-in or justification is provided.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

Claiming the skill can transform 'any teaching topic' creates an overly broad activation scope with no exclusions for unsafe, regulated, or non-educational edge cases. In practice this can cause the skill to engage in contexts it was not designed to handle, increasing the chance of policy bypass, unsafe content generation, or accidental triggering.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

The document explicitly requires CDN-hosted Three.js, Tailwind, KaTeX, and optional D3.js while also asserting zero external dependencies. That contradiction can mislead reviewers and execution environments into permitting generated content under false assumptions, and it exposes users to third-party script execution and availability failures.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill promises a self-contained single-file HTML artifact, but elsewhere requires externally hosted CDN assets. This mismatch is dangerous because downstream systems or users may trust the output as offline/self-contained when it actually depends on remote code, introducing supply-chain, availability, and policy-bypass risks.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The execution flow states the skill runs whenever the user inputs a topic, but it does not define trigger phrases, contextual boundaries, or precedence rules. That ambiguity makes unintended invocation more likely and can let the skill override safer default behavior in mixed or adversarial conversations.

Content

No source excerpt is available for this finding.