Back to skill

Security audit

OnboardAI-新员工入职引导系统

Security checks for vulnerabilities and agentic risk

Overview

This is a local onboarding dashboard, but it needs review because it asks for unnecessary shell/write access and handles HR-style data and credential-like content with weak safeguards.

Install only if you are comfortable with a Chinese local demo-style HR dashboard that can store and export onboarding data in the browser. Treat the sample WiFi password and employee records as placeholders, avoid entering real sensitive HR data until escaping/export warnings are improved, and prefer a version that removes Bash and Write permissions.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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 (3)

T09 · Insecure Skill Coding Practices

Warning
Location
assets/onboarding-system.html:753
Finding

DOM-based cross-site scripting through chat input

Content
View full analysis
👤
${text}
`; ``` ### Technical Analysis The `text` value originates from the user-controlled chat input and is interpolated directly into an HTML template. The resulting string is assigned to `innerHTML` without output encoding or sanitization. Because `innerHTML` parses its input as markup, an attacker can supply active HTML, including elements with event-handler attributes. For example, an image element with an error handler would execute JavaScript when the browser parses the generated message. This application is described as a local dashboard, so exploitation generally requires a victim to enter or paste attacker-supplied content. Nevertheless, the code creates a valid arbitrary JavaScript execution primitive within the application's browser origin. ### Attack Path 1. A victim opens the onboarding dashboard and navigates to the chat panel. 2. The victim enters or pastes attacker-controlled HTML into the chat field, such as an element containing an event handler. 3. `sendMessage()` reads the value into `text`. 4. The application interpolates `text` into an HTML string. 5. The string is appended through `msgs.innerHTML`. 6. The browser parses the injected markup and executes its active content. 7. The injected script can access the page DOM and origin-accessible browser storage. ### Impact Assessment Successful exploitation permits arbitrary JavaScript execution in the dashboard's browser context. The injected code can: - Read and modify onboarding information stored in `localStorage`. - Alter checklist, roadmap, task, and satisfaction data. - Manipulate the visible interface or display deceptive content. - Trigger a ...[truncated 259 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
assets/onboarding-system.html:783
Finding

Persistent DOM XSS sinks in rendering of localStorage-backed data

Content
View full analysis
`

${s.name} · ${s.date}

${s.comment}

${s.score} ★
`).join(''); document.getElementById('heatmap-container').innerHTML = data.satisfaction.dimensions.map(d => `
${d.label}
${d.score}
`).join(''); } ``` The same unsafe rendering pattern is used for checklist values, Kanban tasks, roadmap milestones, survey records, and satisfaction dimensions. ### Technical Analysis Data loaded from `localStorage` is parsed but is not validated against a schema, constrained to expected primitive types, or encoded before insertion into `innerHTML`. Fields such as `s.name`, `s.comment`, `d.label`, checklist text, task descriptions, and roadmap text can therefore contain HTML rather than plain text. If an attacker can poison the `onboardai_data` storage entry, the payload remains persis ...[truncated 1932 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:22
Finding

Skill declares unnecessary shell and filesystem-write capabilities

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (11)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 51)May include surrounding context.

md
1. 加载技能后,Agent 会打开 `assets/onboarding-system.html` 交互式仪表盘

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 · assets/onboarding-system.html (reported line 243)May include surrounding context.

html
</head>
<body>

<!-- Header -->
<div class="header">
  <div class="header-left">
    <div class="header-logo">🚀</div>

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The manifest description presents the skill in Chinese only, and the markdown content likewise assumes Chinese as the sole user language. Under the policy, forcing a specific language without user opt-in or documented justification is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger word "onboarding" is extremely broad in workplace contexts and can match routine discussion rather than an intentional request to launch this specific skill. This increases the risk of accidental invocation and unintended tool use, especially given the skill's broad onboarding functionality and permitted filesystem/shell-capable tools.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The phrase "new hire" is ambiguous and commonly used in normal business communication, making accidental activation plausible. In context, this is more concerning because the skill can interact with local files and shell tooling, so unintended activation could lead to actions or file access the user did not mean to authorize.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger phrase "HR系统" is generic and likely to appear in ordinary enterprise conversations about many unrelated HR products or workflows. Overly broad triggers can cause unintended skill activation, which is dangerous because this skill is allowed to use Bash, Read, and Write tools and may be invoked outside the user's intent.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

Multiple descriptions in the UI claim the system 'automatically' sends reminders, creates accounts, pushes schedules, upgrades alerts, generates growth reports, and performs AI sentiment analysis. The implementation contains no scheduling, messaging, provisioning, analytics, or AI integration logic; it only renders static data and local state changes in the browser.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The embedded knowledge base hardcodes a WiFi password and normalizes insecure credential-handling practices by stating initial passwords will be pushed to users. Exposing operational credentials in front-end code makes them trivially recoverable by any user with browser access and can enable unauthorized network access or broader compromise if reused.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest description explicitly emphasizes a '纯前端零依赖,localStorage持久化' onboarding system, which suggests data remains in-browser for app persistence. However, the code creates a downloadable report containing employee records and satisfaction data, extending behavior beyond simple localStorage persistence into local file export.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The export function packages employee records and satisfaction feedback into a downloadable JSON file without any warning, scoping, or access control. In an onboarding context this can lead to accidental disclosure of personal data, especially on shared devices or when users do not realize the report contains identifiable HR information.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The page declares zh-CN as the document language and presents the interface entirely in Chinese, with no visible option to choose another language. Under the stated policy, forcing a specific language without user opt-in is a natural-language policy issue unless it is explicitly justified as region-specific.

Content

No source excerpt is available for this finding.