AI无障碍合规守护者

• Outcome: Pass
• Latest audit: May 29, 2026 at 8:42 PM UTC
• Version:
  1.0.0

Overview

This is a coherent accessibility-audit skill that captures page content for WCAG analysis, but users should avoid submitting sensitive pages unless authorized.

Use this for authorized accessibility checks of public pages, test pages, or redacted HTML. Before scanning login-protected, internal, regulated, or customer-data pages, confirm authorization and understand that the workflow may render the page, execute JavaScript, capture screenshots and DOM content, and send extracted content to AI/reporting components.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The README explicitly describes loading arbitrary user-supplied URLs in a real browser, executing page JavaScript, taking screenshots, and extracting the full DOM, but it does not mention trust boundaries, authentication/session leakage, sensitive-content capture, or restrictions on internal/private targets. In this context, the skill can process authenticated pages, personal data, or intranet resources, so the lack of privacy and data-exposure warnings materially increases the risk of unintended sensitive data collection or SSRF-like browsing to protected locations.

Vague Triggers

Medium

• Confidence: 91% confidence
• Finding: The trigger phrases "网页无障碍问题" and "生成无障碍报告" are broad enough to match ordinary conversation and may cause the skill to activate outside the user's clear intent. Over-broad invocation can unintentionally route URLs, HTML, screenshots, or page data into browser rendering and downstream analysis, increasing privacy and unexpected-action risk.

Missing User Warnings

Medium

• Confidence: 96% confidence
• Finding: The skill states that DOM structures and page content are analyzed by an external AI service but does not clearly warn users that page data, potentially including sensitive internal content, may be sent off-platform. In a skill that renders full pages and extracts DOM trees, this omission can lead to inadvertent disclosure of proprietary, personal, or authenticated content.

Missing User Warnings

Medium

• Confidence: 90% confidence
• Finding: The workflow loads an arbitrary user-supplied URL in a full browser, executes page JavaScript, and captures DOM, screenshots, and links without any indication of consent, scope restriction, or sensitive-data handling. This can expose private page content, authenticated views, embedded tokens, or personal data from internal or non-public pages to downstream components and stored artifacts.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The workflow forwards the extracted DOM to an AI analysis step for WCAG review without warning that page contents may be transmitted to another processing component or external service. If the scanned page contains confidential business data, personal information, or authenticated content, this creates an unannounced data-transfer and privacy risk beyond the user's likely expectations.

VirusTotal

65/65 vendors flagged this skill as clean.