Back to skill

Security audit

Spec Engine

Security checks for vulnerabilities and agentic risk

Overview

The documented spec tool is mostly local, but the package also includes an undocumented news collector that sends search terms to external services and writes untrusted remote content into reports.

Review this skill before installing. The documented spec-generation features are local and mostly purpose-aligned, but avoid running the daily-news scripts with confidential keywords, and treat generated Markdown/HTML reports as untrusted content.

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 (4)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/dashboard.py:42
Finding

Stored HTML Injection in Generated Dashboard

Content
View full analysis
{t}' for t in s['tech_stack'][:3]) issues_text = ', '.join(s['issues'][:3]) if s['issues'] else '✓ 完整' table_rows += f''' {i+1} {s['project_name']} ...[truncated 2144 chars]
Remediation
View remediation
`, ``, and quoted attribute-breaking input. 6. Consider a restrictive Content Security Policy in generated HTML, such as disallowing scripts and remote resource loading, as defense in depth. ]]>

other

Note
Location
scripts/daily_news.py:22
Finding

Undeclared Outbound Network Collection Exceeds the Skill's Documented Scope

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/daily_news.py:161
Finding

Untrusted Remote Content Is Written into Markdown Reports Without Neutralization

Content
View full analysis
Remediation
View remediation

other

Note
Location
scripts/collectors/xiaohongshu.py:143
Finding

Production Collector Substitutes Fabricated Entries When Search Fails

Content
View full analysis
list: """ Generate mock data for testing. Replace with a real API call in production. """ tz = timezone(timedelta(hours=8)) now = datetime.now(tz) mock_items = [ { "title": f"{keyword} 使用体验分享 - 超好用的AI工具", "url": "https://www.xiaohongshu.com/explore/abc123def", "author": "科技小达人", "publish_time": (now - timedelta(hours=2)).strftime("%Y-%m-%d %H:%M"), "description": f"最近发现了{keyword}这个项目,功能真的很强大,适合开发者使用...", "type": "note", }, { "title": f"{keyword} 安装配置教程,小白也能学会", "url": "https://www.xiaohongshu.com/explore/def456ghi", "author": "程序员日记", "publish_time": (now - timedelta(hours=5)).strftime("%Y-%m-%d %H:%M"), "description": f"手把手教你配置{keyword},从零开始搭建AI助手环境...", "type": "note", }, ] return mock_items ``` ```python try: items = _search_via_web(keyword, hours) if not items: items = _generate_mock_data(keyword, hours) error = "搜索引擎未返回结果,使用模拟数据(小红书无公开API)" except Exception as e: items = _generate_mock_data(keyword, hours) error = f"采集失败: {str(e)},使用模拟数据" ``` ### Technical Analysis The production `collect()` path substitutes fixed fictional records whenever the search engine returns no results or an exception occurs. The mock records receive publication times relative to the current time and are included in `items` and `total`, making them structurally indistinguishable from genuine collected entries to downstream consumers. An error string notes that mock data was used, but this does not reliably protect machine consumers that pro ...[truncated 1074 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (49)

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

Presenting a single CLI decomposition tool as a full automation platform is a meaningful description mismatch that can mislead users about safety checks and workflow coverage. While less severe than hidden scraping behavior, it still creates integrity and operational-risk issues because users may overtrust outputs or assumed controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Presenting a single CLI decomposition tool as a full automation platform is a meaningful description mismatch that can mislead users about safety checks and workflow coverage. While less severe than hidden scraping behavior, it still creates integrity and operational-risk issues because users may overtrust outputs or assumed controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Presenting a single CLI decomposition tool as a full automation platform is a meaningful description mismatch that can mislead users about safety checks and workflow coverage. While less severe than hidden scraping behavior, it still creates integrity and operational-risk issues because users may overtrust outputs or assumed controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Presenting a single CLI decomposition tool as a full automation platform is a meaningful description mismatch that can mislead users about safety checks and workflow coverage. While less severe than hidden scraping behavior, it still creates integrity and operational-risk issues because users may overtrust outputs or assumed controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Presenting a single CLI decomposition tool as a full automation platform is a meaningful description mismatch that can mislead users about safety checks and workflow coverage. While less severe than hidden scraping behavior, it still creates integrity and operational-risk issues because users may overtrust outputs or assumed controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Presenting a single CLI decomposition tool as a full automation platform is a meaningful description mismatch that can mislead users about safety checks and workflow coverage. While less severe than hidden scraping behavior, it still creates integrity and operational-risk issues because users may overtrust outputs or assumed controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Presenting a single CLI decomposition tool as a full automation platform is a meaningful description mismatch that can mislead users about safety checks and workflow coverage. While less severe than hidden scraping behavior, it still creates integrity and operational-risk issues because users may overtrust outputs or assumed controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

Presenting a single CLI decomposition tool as a full automation platform is a meaningful description mismatch that can mislead users about safety checks and workflow coverage. While less severe than hidden scraping behavior, it still creates integrity and operational-risk issues because users may overtrust outputs or assumed controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

Presenting a single CLI decomposition tool as a full automation platform is a meaningful description mismatch that can mislead users about safety checks and workflow coverage. While less severe than hidden scraping behavior, it still creates integrity and operational-risk issues because users may overtrust outputs or assumed controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

Presenting a single CLI decomposition tool as a full automation platform is a meaningful description mismatch that can mislead users about safety checks and workflow coverage. While less severe than hidden scraping behavior, it still creates integrity and operational-risk issues because users may overtrust outputs or assumed controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

Presenting a single CLI decomposition tool as a full automation platform is a meaningful description mismatch that can mislead users about safety checks and workflow coverage. While less severe than hidden scraping behavior, it still creates integrity and operational-risk issues because users may overtrust outputs or assumed controls.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

This file implements Xiaohongshu-focused web searching and scraping logic, including outbound requests to DuckDuckGo and parsing third-party content, which does not align with the skill's stated spec-generation and validation functionality. Such hidden or undocumented collection behavior broadens the attack surface, creates unapproved network egress, and may enable covert data gathering or policy-violating behavior within an otherwise unrelated skill.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The manifest describes a tool for generating and validating project specifications, decomposing tasks, providing a dashboard, version comparison, and historical analysis. This script instead orchestrates platform-specific content collectors, aggregates Bilibili/GitHub/ClawHub/Xiaohongshu results, and writes daily news reports in Markdown/JSON, which is a materially different skill behavior rather than an implementation detail.

Content

No source excerpt is available for this finding.

YARA rule 'agent_skill_mcp_tool_poisoning_metadata': MCP/tool metadata poisoning indicators in tool schemas or skill manifests [agent_skills]

High
Category
YARA Match
Confidence
80% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · scripts/daily_report.json (reported line 16)May include surrounding context.

json
026-03-30",
  "generated_at": "2026-03-30 13:12:24",
  "keyword": "OpenClaw",
  "total": 184,
  "sources": [
    {
      "source": "bilibili",
      "items": [
        {
          "title": "漫谈OpenClaw",
          "url": "https://www.bilibili.com/video/BV1gWXaBrE2L",
          "author": "zhugeshensuan",
          "publish_time": "2026-03-30 12:59",
          "views": 0,
          "description": "漫谈OpenClaw"
        },
        {
          "title": "在线指导安装免token的小龙虾,已经装上了,使用的网页api接口,不用配自己的apikey,不用烧token~",
          "url": "https://www.bilibili.com/video/BV1GnXaBCEs8",
          "author": "阿鲲正在飞",
          "publish_time": "2026-03-30 12:56",
          "views": 1,
          "description": "-"
        },
        {
          "title": "【效率神器】全输入框通用语音输入!AI 工具 / VSCode / claude / 养龙虾 都能用|提升开发效率",
          "url": "h

Hidden Instructions

High
Category
Prompt Injection
Confidence
60% confidence
Finding

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

Content

Scanner excerpt · scripts/daily_report.md (reported line 361)May include surrounding context.

md
> OpenClaw Skills 安装指南:四种方法详解 非臀噶疟准巴肯节蚀趁却纬谄浦少 <p>2026年3月23日,人工智能领域传来喜讯,我国在医疗领域的人工智能应用取得了重大突破。以下为您带来今日人工智能早间新闻,带您了解AI赋能医疗的最新进展。</p> <p>一、引言</p> <p>近年来,人工...
16. **[[pull] main from openclaw:main](https://github.com/kiss2u/moltbot/pull/231)** - @pull[bot] - PR - kiss2u/moltbot
   > See [Commits](/kiss2u/moltbot/pull/231/commits) and [Changes](/kiss2u/moltbot/pull/231/files) for more details.  ----- Created by [<img src="https://p...
17. **[AI员工——OpenCode、OpenClaw+Ollama的安装与配置零基础安装部署openClaw并接入飞书企业微信 超详细教程至肝呈擦坏68637](https://github.com/sfrongm/hfrvp/issues/462)** - @sfrongm - issue - sfrongm/hfrvp
   > AI员工——OpenCode、OpenClaw+Ollama的安装与配置 肝医茨衙茨俳酵俜诚股蘸乐皇咳蔽 <p>随着科技的不断发展,自动化办公和智能化管理已成为企业提升效率的关键。openClaw是一款功能强大的自动化工作流平台,可以帮助企业实现自动化办公。而飞书和企业微信作为国内知名的办公协作工...
18. **[ToClaw他来了!融合OpenClaw、网页AI、远程功能,更适合真正的办公落地算法时间复杂度分析和空间复杂度分析:揭秘程序效率的秘密武器颖亟诟疚霖86318](https://github.com/amergbe/4fa8d/issues/449)** - @amergbe - issue - amergbe/4fa8d
   > ToClaw他来了!融合OpenClaw、网页AI、远程功能,更适合真正的办公落地 啪辛负雌谄犊硕诟猜皇壬瘫致怨械 <p>在计算机科学领域,算法是解决问题的核心。然而,如何衡量一个算法的效率呢?这就需要借助算法时间复杂度分析和空间复杂度分析这两把“秘密武器”。本文将深入浅出地介绍这两项分析,帮助读者...

Hidden Instructions

High
Category
Prompt Injection
Confidence
60% confidence
Finding

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

Content

Scanner excerpt · scripts/daily_report.md (reported line 362)May include surrounding context.

md
> OpenClaw Skills 安装指南:四种方法详解 非臀噶疟准巴肯节蚀趁却纬谄浦少 <p>2026年3月23日,人工智能领域传来喜讯,我国在医疗领域的人工智能应用取得了重大突破。以下为您带来今日人工智能早间新闻,带您了解AI赋能医疗的最新进展。</p> <p>一、引言</p> <p>近年来,人工...
16. **[[pull] main from openclaw:main](https://github.com/kiss2u/moltbot/pull/231)** - @pull[bot] - PR - kiss2u/moltbot
   > See [Commits](/kiss2u/moltbot/pull/231/commits) and [Changes](/kiss2u/moltbot/pull/231/files) for more details.  ----- Created by [<img src="https://p...
17. **[AI员工——OpenCode、OpenClaw+Ollama的安装与配置零基础安装部署openClaw并接入飞书企业微信 超详细教程至肝呈擦坏68637](https://github.com/sfrongm/hfrvp/issues/462)** - @sfrongm - issue - sfrongm/hfrvp
   > AI员工——OpenCode、OpenClaw+Ollama的安装与配置 肝医茨衙茨俳酵俜诚股蘸乐皇咳蔽 <p>随着科技的不断发展,自动化办公和智能化管理已成为企业提升效率的关键。openClaw是一款功能强大的自动化工作流平台,可以帮助企业实现自动化办公。而飞书和企业微信作为国内知名的办公协作工...
18. **[ToClaw他来了!融合OpenClaw、网页AI、远程功能,更适合真正的办公落地算法时间复杂度分析和空间复杂度分析:揭秘程序效率的秘密武器颖亟诟疚霖86318](https://github.com/amergbe/4fa8d/issues/449)** - @amergbe - issue - amergbe/4fa8d
   > ToClaw他来了!融合OpenClaw、网页AI、远程功能,更适合真正的办公落地 啪辛负雌谄犊硕诟猜皇壬瘫致怨械 <p>在计算机科学领域,算法是解决问题的核心。然而,如何衡量一个算法的效率呢?这就需要借助算法时间复杂度分析和空间复杂度分析这两把“秘密武器”。本文将深入浅出地介绍这两项分析,帮助读者...

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 · templates/spec-template.md (reported line 5)May include surrounding context.

md
## 1. 项目背景

<!-- 填写指引:描述项目的来源、解决的问题、业务场景等 -->
<!-- 示例:当前用户管理依赖手动 Excel,效率低下且易出错,需要自动化系统 -->

{{BACKGROUND}}

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 · templates/spec-template.md (reported line 5)May include surrounding context.

md
## 1. 项目背景

<!-- 填写指引:描述项目的来源、解决的问题、业务场景等 -->
<!-- 示例:当前用户管理依赖手动 Excel,效率低下且易出错,需要自动化系统 -->

{{BACKGROUND}}

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 · templates/spec-template.md (reported line 26)May include surrounding context.

md
## 4. 技术方案

<!-- 填写指引:描述架构设计、关键技术选型的理由、核心流程 -->
<!-- 示例:采用前后端分离架构,后端使用 FastAPI 提供 RESTful API... -->

{{TECH_SOLUTION}}

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 · templates/spec-template.md (reported line 26)May include surrounding context.

md
## 4. 技术方案

<!-- 填写指引:描述架构设计、关键技术选型的理由、核心流程 -->
<!-- 示例:采用前后端分离架构,后端使用 FastAPI 提供 RESTful API... -->

{{TECH_SOLUTION}}

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 · templates/spec-template.md (reported line 47)May include surrounding context.

md
## 7. API 需求

<!-- 填写指引:列出需要暴露或调用的 API 接口 -->
<!-- 示例:POST /api/users/register - 用户注册接口 -->

{{API_REQUIREMENTS}}

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 · templates/spec-template.md (reported line 47)May include surrounding context.

md
## 7. API 需求

<!-- 填写指引:列出需要暴露或调用的 API 接口 -->
<!-- 示例:POST /api/users/register - 用户注册接口 -->

{{API_REQUIREMENTS}}

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 · templates/spec-template.md (reported line 54)May include surrounding context.

md
## 8. 文件结构

<!-- 填写指引:项目目录结构,可用树形格式 -->
<!-- 示例:
├── src/            # 源代码
├── tests/          # 测试

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 · templates/spec-template.md (reported line 54)May include surrounding context.

md
## 8. 文件结构

<!-- 填写指引:项目目录结构,可用树形格式 -->
<!-- 示例:
├── src/            # 源代码
├── tests/          # 测试

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
90% confidence
Finding

The skill advertises multiple script-based capabilities and the static analyzer detected file, environment, and network-relevant behaviors, but the manifest does not declare any tool scope restrictions such as allowed-tools or permissions. In an agent setting, this increases the risk that the skill can access broader capabilities than users expect, especially given the large gap between the high-level description and the implied operational surface.

Content

No source excerpt is available for this finding.