Back to skill

Security audit

全能旅行规划师

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent travel-planning skill with disclosed web research and optional HTML output, but users should be careful with generated HTML because it can load a third-party chart script and should escape untrusted content.

Install only if you want a Chinese-oriented travel planner that performs web searches for current travel information. For HTML output, treat user input and web-sourced names/links as untrusted, escape dynamic values before generating the page, and consider removing or pinning the Chart.js CDN dependency before opening or sharing generated files.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
references/html_template.md:40
Finding

Unescaped Dynamic Content in Generated HTML

Content
View full analysis
{出发地} → {目的地} | 全能旅行规划师 ``` ```html

{出发地} → {目的地}

``` ```html {出发地} → {目的地} ``` The template usage instruction at line 709 states: ```text 1. **数据填充**:将模板中所有 `{变量}` 替换为实际搜集的数据 ``` This instruction requires every template variable to be replaced with collected data, but it does not require contextual encoding, validation, or sanitization. ### Technical Analysis The HTML-generation workflow inserts values obtained from users and web searches directly into HTML contexts. The same unsafe substitution pattern is used in document titles, text elements, tables, attributes, and other sections throughout the template. HTML treats substituted values as markup rather than plain text. For example, a malicious origin or destination value could contain: ```html

``` If this value is inserted without encoding, it closes the existing element and creates an executable `script` element. Values sourced from search results present a similar risk if compromised or attacker-controlled pages cause executable markup to be incorporated into the itinerary. The template also permits inline JavaScript, so there is no effective Content Security Policy preventing injected inline scripts from running. The generated file is explicitly intended to be opened through a file-presentation tool or browser preview, creating an execution point for the injected content. ### Attack Path 1. An attacker supplies a crafted travel field, such as an origin, destination, hotel name, restaurant name, or itinerary note containing HTML or Jav ...[truncated 1321 chars]

Remediation
View remediation
` as `>`. - Encode `"` as `"`. - Encode `'` as `'`. 2. Use a template engine that enables automatic HTML escaping by default rather than performing unrestricted string replacement. 3. Validate fields according to their expected formats: - Permit only reasonable city and destination characters and lengths. - Parse prices, percentages, durations, and ratings as numeric values. - Permit only explicitly approved URL schemes and domains for links. - Reject control characters and executable URL schemes such as `javascript:`. 4. Treat all user input and web-search content as untrusted. Add explicit Skill instructions prohibiting the direct inclusion of HTML, scripts, event-handler attributes, or executable URLs from those sources. 5. Avoid inserting dynamic values into inline JavaScript. If data must be exposed to JavaScript, serialize it with a safe JSON encoder and place it in a non-executable data element. 6. Add a restrictive Content Security Policy. Prefer external or hashed first-party scripts and disallow arbitrary inline execution. 7. Add security tests using payloads in every template field, including: ```html Open ``` The generated output should display these values only as inert text. ]]>

T08 · Insecure Dependencies

Note
Location
references/html_template.md:38
Finding

Unpinned Third-Party JavaScript Dependency

Content
View full analysis
{出发地} → {目的地} | 全能旅行规划师 ``` The Skill requires this dependency at `SKILL.md:329`: ```text 6. **图表依赖**:预算饼图使用 Chart.js CDN(`https://cdn.jsdelivr.net/npm/chart.js`),行程时间线使用纯 CSS 实现 ``` The package URL does not specify an exact Chart.js version and does not include a Subresource Integrity hash. This also conflicts with the statements that all CSS and JavaScript are inline and that the generated file has no external dependencies. ### Technical Analysis The generated HTML executes JavaScript retrieved from: ```text https://cdn.jsdelivr.net/npm/chart.js ``` Because the dependency is not pinned to an exact immutable version, the code returned for the generated document may change after the Skill has been audited. The browser trusts and executes the returned script with the privileges available to scripts in the itinerary page. The absence of a Subresource Integrity attribute means the browser has no cryptographic mechanism for confirming that the downloaded script matches a reviewed artifact. The generated page therefore inherits risk from the package publication account, package registry, CDN, DNS resolution, and delivery infrastructure. Loading the script also creates an external network request when the itinerary is opened. This undermines the documented single-file and offline behavior and exposes ordinary connection metadata, such as the user's IP address and request headers, to the CDN. ### ...[truncated 1255 chars]
Remediation
View remediation
``` Replace `EXACT_VERSION` and `REVIEWED_HASH` with a verified release and its correct integrity value. 3. Use an immutable URL and maintain a controlled dependency-update process that includes: - Release-note review. - Vulnerability scanning. - Integrity-hash regeneration. - Functional and security regression testing. 4. Add a restrictive Content Security Policy that permits scripts only from the selected source and blocks unexpected script origins. 5. Update the documentation so it accurately discloses: - The external dependency. - The network request made when the file is opened. - The loss of chart functionality when offline. - The trust relationship with the CDN and package registry. 6. If the output must truly be a standalone single file, require all script content to be embedded from a reviewed local artifact and prohibit runtime dependency retrieval. ]]>
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (8)

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 · references/html_template.md (reported line 339)May include surrounding context.

md
<body>
  <div class="container">

    <!-- 1. 方案概览 -->
    <div class="overview-card">
      <h1>{出发地} → {目的地}</h1>
      <div class="overview-meta">

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 · references/html_template.md (reported line 364)May include surrounding context.

md
<div class="card">
      <div class="card-title"><span class="icon">🚀</span> 城际交通方案</div>

      <!-- 飞机方案 -->
      <div class="transport-card flight">
        <div class="transport-header">
          <span class="emoji">✈️</span>

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest description embeds a long list of ambiguous activation keywords such as '旅游攻略' and '行程规划', which are common phrases in normal conversation and not sufficiently specific for reliable routing. If the platform uses this description for discovery or invocation, the skill may be selected too often, causing overreach and potentially exposing user context to unnecessary processing and external search steps.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger conditions are very broad and include generic travel-planning phrases that can match many ordinary travel-related requests, increasing the chance of unintended invocation. This can cause the skill to activate in contexts where the user did not explicitly want this planner, leading to incorrect tool usage, unnecessary web queries, or response hijacking away from a more appropriate skill.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

文件标题和全文说明均以中文固定描述流程与模板,未见允许用户选择语言、地区或本地化偏好的说明。根据规则,若技能强制特定语言且没有用户选择或合理限定,属于自然语言层面的语言/locale 政策问题。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

该 markdown 模板以“HTML 输出格式模板”中文内容开头,并在示例 HTML 中固定 lang="zh-CN",同时说明价格遵循中国惯例。这构成了默认强制特定语言/区域设置的自然语言策略问题,因为文件没有说明这是仅限中文用户或中国场景的专用模板,也没有提供语言/locale 的可选机制。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

<html lang="zh-CN"> 将输出页面语言直接限定为简体中文(中国),属于显式 locale 约束。文件中未看到让用户选择语言/地区的说明,也未将这一限制描述为特定合规或地区专用要求,因此符合语言/locale policy violation 条件。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

“价格标注遵循中国惯例(¥ 人民币,涨红跌绿)”是对输出 locale 的自然语言强约束。若该技能可用于一般旅行规划,这会在未征得用户同意的情况下默认施加中国本地化格式。

Content

No source excerpt is available for this finding.