Back to skill

Security audit

Zero-to-Launch

Security checks for vulnerabilities and agentic risk

Overview

The product-coaching skill itself is mostly coherent, but the package includes unrelated WeChat publishing utilities with committed API credentials and draft-creation capability.

Review before installing. The SKILL.md prompt can be used as a product-ideation assistant, but the package should not include the WeChat publisher directory or committed config secret. Rotate the exposed WeChat secret, remove config.json from the package, and avoid previewing untrusted HTML with the included preview server.

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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
wechat-publisher/config.json:2
Finding

Hardcoded WeChat API Credentials Committed to the Repository

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
wechat-publisher/preview.js:40
Finding

Unsanitized HTML Execution Through a Network-Exposed Preview Server

Content
View full analysis
${compressed} ``` The generated document is served by an HTTP server without an explicit loopback address: ```js const server = http.createServer((req, res) => { res.writeHead(200, { 'Content-Type': 'text/html; charset=utf-8' }); res.end(previewHTML); }); server.listen(PORT, () => { console.log(`\n📱 公众号预览已启动: http://localhost:${PORT}`); console.log(' 这是文章经过 HTML 压缩后在公众号环境中的模拟效果'); console.log(' 按 Ctrl+C 退出\n'); }); ``` ### Technical Analysis The preview tool accepts a user-selected HTML file through `process.argv[2]`, extracts content, performs whitespace compression, and inserts that content directly into the preview page. The sanitizer only removes a `font-family` declaration. It does not remove or neutralize active content such as: - `
Remediation
View remediation
{ console.log(`Preview available at http://127.0.0.1:${PORT}`); }); ``` 2. Reject requests whose socket address is not loopback. 3. Sanitize article content with a maintained, allowlist-based HTML sanitizer. Permit only the elements and attributes required for WeChat article formatting. 4. Explicitly remove script elements, event-handler attributes, iframes, active SVG content, forms, and dangerous URL schemes. 5. Apply a restrictive Content Security Policy, for example by disabling script execution and limiting external resources: ```js res.writeHead(200, { 'Content-Type': 'text/html; charset=utf-8', 'Content-Security-Policy': "default-src 'none'; img-src 'self' data: https:; style-src 'unsafe-inline'; script-src 'none'; frame-src 'none'; form-action 'none'; base-uri 'none'" }); ``` 6. Add `X-Content-Type-Options: nosniff`, `Referrer-Policy: no-referrer`, and a suitable framing restriction. 7. Serve only a fixed preview route and return `404` for unrelated requests. 8. Clearly warn users not to preview untrusted HTML until sanitization is implemented. 9. Consider rendering the article in a sandboxed iframe without `allow-scripts`. 10. Add security tests covering script tags, event handlers, `javascript:` links, SVG payloads, iframes, and access through non-loopback interfaces. ]]>
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
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (63)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description presents an interactive product ideation and validation assistant for independent developers and founders. However, the code shown is purely a utility script for analyzing inline styles in an HTML file, likely for WeChat publishing or HTML cleanup/optimization. It does not implement any product idea exploration, user profiling, competitor analysis, market sizing, MVP planning, or validation workflows. Its primary purpose, inputs, and outputs are materially different from the declared purpose, so this is a clear mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是一个产品探索/创业分析助手,核心能力应围绕想法澄清、用户研究、竞品分析、市场规模、验证方案和 MVP 定义等。实际代码与这些功能无关,只是在处理微信公众号/博客文章的 HTML 内容,进行字符串清洗、压缩和保存调试结果。代码既没有任何产品分析逻辑,也没有与用户想法、创业验证或可视化状态图相关的实现。其主要目的与声明严重不符,因此应判定为明显不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是产品战略/创业辅导类 Skill,核心能力应围绕想法澄清、市场分析、MVP 定义和验证方案生成。给出的代码却完全没有任何与产品管理、用户研究、竞品分析、市场规模估算或 Launch 规划相关的逻辑。相反,它操作本地 HTML 文件,按注释标记提取文章片段,移除注释和部分 div wrapper,替换/删除样式,压缩 HTML,并将结果写回新文件,明显属于微信文章发布前的内容清洗调试工具。其主要用途、处理对象、资源访问方式和触发场景都与声明严重不符,因此应判定为明显不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description presents a brainstorming/product-strategy assistant for taking vague startup ideas to launch planning artifacts. The supplied code does not implement any such functionality. Instead, it is a technical utility script for parsing and cleaning HTML content from a local file and saving the result. This is a materially different primary purpose and capability set. The behavior is unrelated to the listed triggers and promised outputs, so this should be flagged as a clear mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description is about interactive product ideation, market validation, and startup planning workflows. The supplied code does none of that. Instead, it is a utility script for processing a blog/article HTML file for WeChat publishing or debugging. Its primary purpose, inputs, outputs, and triggers are unrelated to the declared skill. This is a clear material mismatch between description and actual behavior.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description presents an ideation-to-launch product strategy assistant for entrepreneurs. The supplied code does something entirely different: it loads a local HTML document, extracts content between comment markers, strips wrapper divs, rewrites div tags to section tags, removes or normalizes styles/attributes, compresses the HTML, logs statistics, and saves the result to a new file. This is a content-processing/debug publishing script, not a product discovery or launch-planning skill. The behavior is materially different from the declared purpose, and it also uses filesystem read/write operations that are not reflected in the declared permissions.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是产品探索/创业辅助类能力,面向用户进行产品构思、市场分析与验证设计;而实际代码完全没有任何与产品想法分析、用户画像、竞品、市场规模、MVP 定义或原型生成相关的逻辑。相反,它执行的是文件级 HTML 后处理:读取固定路径的 blog-post-wechat.html,替换特定

和

样式,移除 id/class/font-family/position/z-index 等属性,替换字符实体,提取文章内容区间,压缩输出并检查是否低于 20000 字符。这是一个微信文章发布优化工具,与声明的创业产品规划用途在主要目的、操作对象、触发场景和能力上都明显不一致。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是一个面向产品探索与创业验证的分析/规划型 Skill,预期行为应包括产品思路梳理、用户画像、竞品分析、市场规模、验证方案、MVP 定义等。实际代码完全没有任何与产品策略、创业分析、用户交互式探索或可视化状态图相关的逻辑。相反,它是一个针对 blog-post-wechat.html 文件的静态内容优化脚本:读取文件、替换内联 CSS、删除冗余样式、压缩 HTML、统计字符数,并写回源文件。这不仅主目的不同,还涉及未声明的文件系统读写能力与微信 HTML 发布优化场景,和给定触发词/用途明显不一致。因此应判定为明显不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description is for an interactive product-planning assistant focused on idea validation, market analysis, MVP definition, and launch preparation. The supplied code does none of that. Instead, it is a content-processing utility for a WeChat publishing workflow: it reads a specific HTML file, applies regex-based style removals and substitutions, writes the result back, and checks character-count compliance. This is a materially different primary purpose and involves undeclared local file manipulation unrelated to the stated triggers or business-analysis function.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是产品探索/创业辅导类能力,预期应围绕产品想法分析、市场验证、竞品与 MVP 规划等展开。而实际代码完全没有处理用户输入、产品分析、创业建议或可视化状态图;它只是在本地读取并重写一个 HTML 文件,删除若干样式、压缩内容并检查字符长度限制。这属于微信文章发布优化工具,与声明的核心用途、功能范围和触发场景都明显不一致,因此应判定为描述与行为严重不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是一个产品探索/创业辅助类技能,核心能力应围绕产品构思、市场分析、验证方案和 MVP 规划。实际代码完全没有实现任何与产品发现、创业建议、用户画像、竞品分析或 Launch 规划相关的逻辑。相反,它实现的是一个微信公众号 HTML 文章本地预览与压缩检查工具,主要处理文件、HTML 内容和本地服务器预览。这属于 materially different primary purpose,且代码具备的文件访问、本地服务和微信文章预览能力均未在声明中体现,因此应判定为明显不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是产品探索/创业辅助类能力,面向产品想法澄清、市场分析和 MVP 规划;而代码没有实现任何相关分析、用户画像、竞品研究、市场规模测算、验证方案生成或原型输出。相反,代码的核心目的完全是微信公众号文章发布与格式适配,涉及本地文件读取、微信公众号认证、媒体上传和草稿创建。这属于 materially different primary purpose,并且包含多项未声明的外部资源访问与发布能力,因此存在明显描述-行为不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是一个产品发现/创业规划类技能,核心能力应围绕用户访谈式引导、市场与竞品分析、验证方案设计、MVP 定义和原型输出。实际代码却是一个独立的内容处理脚本:它只操作本地 HTML 文件,提取文章片段,移除 class/id/font-family/position/z-index 等样式,替换 rgba 颜色值,压缩 HTML,并打印兼容性统计信息。这与声明用途在目标用户、输入类型、输出结果和核心能力上都明显不一致,属于实质性功能错配。

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 · blog-post-wechat.html (reported line 42)May include surrounding context.

html
</head>
<body>

<!-- Toolbar - NOT part of the copied content -->
<div class="toolbar">
  <div>
    <div class="toolbar-title">公众号排版预览</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 · blog-post-wechat.html (reported line 42)May include surrounding context.

html
</head>
<body>

<!-- Toolbar - NOT part of the copied content -->
<div class="toolbar">
  <div>
    <div class="toolbar-title">公众号排版预览</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 · blog-post-wechat.html (reported line 60)May include surrounding context.

html
<div style="max-width:578px;margin:0 auto;padding:0;font-family:-apple-system,BlinkMacSystemFont,'PingFang SC','Hiragino Sans GB','Microsoft YaHei',sans-serif;font-size:16px;line-height:1.85;color:#2D2A3E;">

<!-- HERO -->
<div style="background:#1B1642;color:#fff;padding:40px 28px 36px;text-align:center;">
  <div style="display:inline-block;background:rgba(255,255,255,0.12);border:1px solid rgba(255,255,255,0.2);padding:3px 14px;border-radius:20px;font-size:12px;color:rgba(255,255,255,0.8);margin-bottom:20px;">Product Sense Coach</div>
  <div style="font-size:23px;font-weight:700;line-height:1.5;margin-bottom:14px;color:#fff;">AI 时代,产品经理怎么做产品决策?<br>我用了一套「AI 思考搭子」方法论</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 · blog-post.html (reported line 590)May include surrounding context.

html
</head>
<body>

<!-- ===== HERO ===== -->
<div class="hero">
  <div class="hero-content">
    <div class="hero-tag">Product Sense Coach</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 · blog-post.html (reported line 590)May include surrounding context.

html
</head>
<body>

<!-- ===== HERO ===== -->
<div class="hero">
  <div class="hero-content">
    <div class="hero-tag">Product Sense Coach</div>

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

This file implements a standalone WeChat publishing utility that is materially unrelated to the declared 'zero-to-launch' ideation and product-validation skill. Capability drift like this is dangerous because it introduces hidden operational behavior and credential-dependent publishing functions that users of the skill would not reasonably expect, increasing the risk of unauthorized content handling or accidental misuse.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The code can authenticate to WeChat APIs, upload media, and create draft articles in a public account, which is an external publishing capability unrelated to the skill's stated purpose. In a mismatched skill context, this becomes dangerous because it could be repurposed to exfiltrate generated content to third-party infrastructure or publish material under a user's account without clear expectation or authorization boundaries.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The contributing guide explicitly asks users to share the full conversation or highlights, but provides no warning to remove secrets, personal data, business-sensitive information, or third-party data. Because this skill is designed for early-stage product ideation, conversations may contain proprietary startup plans, customer details, credentials pasted by mistake, or other sensitive context that could be exposed in public issues.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The bug-report section asks contributors to include conversation context without instructing them to sanitize or minimize that context. In this skill’s domain, bug reports may include raw product ideas, user research notes, target-customer information, or accidentally pasted secrets, creating a realistic risk of oversharing in public trackers.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The promotion content consistently markets a different skill identity ('Product Sense Coach') with a different audience and purpose than the manifested skill ('zero-to-launch'). This can mislead users about what they are installing or invoking, undermine trust, and enable deceptive repackaging or bait-and-switch behavior even if there is no direct code execution risk in this file.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
74% confidence
Finding

The skill explicitly instructs use of external web_search for competitor analysis but does not declare any tool scope, permissions, or allowed-tools boundaries. This creates a capability-governance gap: users and reviewers cannot tell what external access the skill may invoke, and the model may perform network lookups without transparent approval or restriction.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The skill instructions and user-facing examples are entirely in Chinese and present the workflow as the default interaction mode, with no indication that the user can choose another language. This can constitute a language-policy issue because it effectively forces a specific language without explicit user opt-in or a documented region-specific justification.

Content

No source excerpt is available for this finding.