Back to skill

Security audit

Qizheng Brand Marketing

Security checks for vulnerabilities and agentic risk

Overview

The marketing skill itself is mostly coherent, but the bundled HTML includes an unrelated iframe inspection script that can expose page and form data to any framing parent.

Review before installing. The core brand-marketing instructions are not malicious by themselves, but the HTML artifact should have the iframe inspection/highlighting script removed or restricted with strict origin validation, no form-value collection, and clear disclosure. Also clarify consent, access control, and retention for client materials, public-source enrichment, emergency contacts, and permanent delivery links.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
index.html:594
Finding

Unrestricted iframe messaging exposes DOM content and user input to an untrusted parent

Content
View full analysis
50 ? text.substring(0, 50) + "..." : text; } ``` Captured element information is transmitted to the parent window using the unrestricted wildcard target origin: ```javascript var elementInfo = { tagName: target.tagName.toLowerCase(), rect: { left: rect.left, top: rect.top, right: rect.right, bottom: rect.bottom, width: rect.width, height: rect.height, x: rect.x, y: rect.y }, selector: selector, text: text, attributes: attributes, url: window.location.href, path: window.location.pathname, timestamp: Date.now() }; try { window.parent.postMessage({ type: "iframe-element-hover", data: elementInfo, source: "iframe-highlight-injector" }, "*"); } catch (error) { console.warn("无法发送消息到父窗口:", error); } ``` The click handler performs the same unrestricted disclosure: ```javascript var elementInfo = { tagName: target.tagName.toLowerCase(), rect: { left: rect.left, top: rect.top, right: rect.right, bottom: rect.bottom, width: rect.width, height: rect.height, x: rect.x, y: rect.y }, selector: selector, text: text, attributes: attributes, url: window.location.href, path: window.location. ...[truncated 5559 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (21)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

This file embeds a large iframe-inspection and cross-window messaging script that is unrelated to a brand-marketing landing page. The script enumerates DOM elements, captures text, attributes, selectors, and page URL, and communicates them to a parent window, which creates surveillance capability inconsistent with the declared skill purpose and strongly suggests covert data collection or unauthorized page introspection.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The hover handler collects element metadata including visible text, attributes, selector, page URL, and geometry, then posts it to the parent window with no origin restriction or user disclosure. In the context of a marketing skill, this is unnecessary and dangerous because it can leak sensitive page structure or user-visible content from embedded contexts to an external controller.

Content

No source excerpt is available for this finding.

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · index.html (reported line 83)May include surrounding context.

html
</div>

<div class="container">
  <!-- TIERS -->
  <div class="section">
    <div class="st">三档套餐,为不同阶段的企业设计</div>
    <div class="ss">所有套餐均包含品牌诊断+核心叙事,区别在于执行深度和陪跑时长</div>

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · packages.html (reported line 83)May include surrounding context.

html
</div>

<div class="container">
  <!-- TIERS -->
  <div class="section">
    <div class="st">三档套餐,为不同阶段的企业设计</div>
    <div class="ss">所有套餐均包含品牌诊断+核心叙事,区别在于执行深度和陪跑时长</div>

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · index.html (reported line 83)May include surrounding context.

html
</div>

<div class="container">
  <!-- TIERS -->
  <div class="section">
    <div class="st">三档套餐,为不同阶段的企业设计</div>
    <div class="ss">所有套餐均包含品牌诊断+核心叙事,区别在于执行深度和陪跑时长</div>

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · packages.html (reported line 83)May include surrounding context.

html
</div>

<div class="container">
  <!-- TIERS -->
  <div class="section">
    <div class="st">三档套餐,为不同阶段的企业设计</div>
    <div class="ss">所有套餐均包含品牌诊断+核心叙事,区别在于执行深度和陪跑时长</div>

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest description embeds multiple high-level trigger words like '品牌营销', '爆款预测', and '危机公关' that are common across many benign user requests. Because manifest metadata may be used for routing, these broad terms can make the skill over-match unrelated prompts and activate without sufficient user intent specificity.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The repeated trigger-word section reinforces ambiguous activation cues without clarifying exclusion conditions, making accidental or excessive activation more likely. Repetition of generic trigger phrases increases routing sensitivity while providing no guardrails, which is especially risky in a skill that can initiate structured analysis and simulation workflows.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The quick-call trigger phrases are broad, generic user utterances such as '帮我做品牌定位' and '帮我制定推广方案', with no explicit activation boundaries or disambiguation logic. In an agent environment, this can cause the skill to activate on ordinary conversation and steer execution into skill-specific workflows unexpectedly, increasing the risk of overreach, unintended data collection, or tool/script invocation.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · build.py (reported line 4)May include surrounding context.

python
#!/usr/bin/env python3
"""Build the full HTML page for the brand marketing skill package."""

SKILL_MD = open('/workspace/skills/qizheng-brand-marketing/SKILL.md').read()

HTML = '''<!DOCTYPE html>
<html lang="zh-CN">

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The embedded comments explicitly describe this as an injection script for use on target websites to support cross-origin iframe highlighting, yet it is bundled directly into this page and auto-enabled. That mismatch between stated purpose and actual inclusion increases suspicion that the page is being used as a carrier for generic inspection tooling rather than only presenting marketing content.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The script silently transmits element data and the current page URL to the parent window during hover events, without any visible notice or consent flow. Even if not immediately exploitative, undisclosed telemetry about page contents and structure is a privacy and security risk, especially when the skill does not need such functionality.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The click handler captures selected element contents and full attributes, then posts them to the parent window without disclosure. Click-derived element data can include user-entered or context-sensitive information, making this a meaningful exfiltration risk in embedded scenarios.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The message listener accepts arbitrary control messages from a parent window to toggle inspection, update selected elements, and manipulate behavior without validating event.origin or authenticating the sender. This allows any embedding parent to drive DOM inspection features and potentially exfiltrate page information or interfere with user interaction.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill promises delivery via permanently accessible web links and elsewhere includes emergency contact rosters with reserved phone numbers, yet it does not mention authentication, expiry, access restrictions, or handling of embedded personal contact data. In this context, exposing crisis plans, internal roles, and contact details through long-lived links creates a realistic risk of unauthorized access, data leakage, and social-engineering abuse.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill explicitly states it will collect client-provided materials and, if those are incomplete, gather additional information from public sources, but it provides no limits on what data may be collected, no consent model, and no retention or sharing safeguards. In a marketing/brand-advisory context this can lead to over-collection of personal or business-sensitive information and downstream misuse in generated deliverables.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

This markdown file contains end-user-facing natural language exclusively in Chinese, but it does not state that the skill is region-specific or provide any opt-in/choice for language. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale constraint is explicitly justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

This code sets the generated page language to zh-CN, which enforces a specific locale in the user-facing output. The file contains no visible indication that users can choose another language or opt in to this locale, which can violate language/locale policy requirements.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The document declares lang="zh-CN", and the visible content is entirely Chinese, with no indication that users may choose another language or that the skill is intentionally limited to a China-specific audience. Under the stated policy, forcing a specific language or locale without opt-in can be a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

All user-facing content in the file is presented only in Chinese, with no indication that users may choose another language or that the skill is intentionally limited to a Chinese-language audience for a documented reason. This can constitute a language-policy issue under the stated rule when no opt-in or justification is provided.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The HTML document is explicitly marked with lang="zh-CN", and the page content is entirely presented in Chinese. Under the policy rule for natural-language violations, forcing a specific language without user opt-in or a documented region-specific justification is a locale policy issue.

Content

No source excerpt is available for this finding.