T05 · Unauthorized Access and Privilege Escalation
- Location
SKILL.md:27- Finding
Persistent External Publication of User-Supplied HTML
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This skill can publish user HTML to a long-lived deployed web page and includes unrestricted URL rendering helpers without enough user consent, scoping, or cleanup controls.
Install only if you are comfortable with HTML content being deployed to a potentially reachable long-lived URL. Do not use it for private documents, credentials, internal pages, customer data, or regulated information unless the deployment destination, access controls, expiration, and deletion process are explicitly handled outside the skill.
SKILL.md:27Persistent External Publication of User-Supplied HTML
scripts/generate-pdf.js:1Unpinned Playwright and Chromium Installation Instructions
scripts/generate-pdf.js:12Unrestricted URL Rendering with Chromium Sandbox Disabled
The trigger phrases are broad enough to match common requests like '导出PDF' or '下载PDF' without confirming that the user intends to publish HTML to a public or persistent deployment URL first. In this skill, accidental activation is more dangerous than usual because the workflow explicitly deploys user content and says the resulting link is long-lived, creating a confidentiality and unintended disclosure risk.
The skill description and prescribed user-facing text are written as fixed Chinese-language behavior, including exact phrases for triggering and response templates. There is no indication that the user can choose another language or that the locale restriction is required for a region-specific compliance reason.
The activation conditions only describe when to trigger, not when to refuse or defer, so the skill may activate on ambiguous inputs and proceed with deploying content. Because this skill's prescribed workflow publishes user-provided HTML or URLs to a persistent link before local PDF export, missing guardrails materially increase the chance of exposing sensitive documents or internal pages.
The user-facing workflow omits a clear warning that the content will be deployed to a reachable, persistent URL before PDF export, while even stating that the link is '长期有效'. This is dangerous because users may believe they are performing a simple file conversion, when in fact they are publishing potentially sensitive HTML content or URL-rendered material.
npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.
The page embeds a user-supplied target inside an iframe, which allows the skill to load arbitrary remote web content instead of being limited to a local HTML-to-PDF conversion flow. In an agent setting, this broadens capability into general browsing/network retrieval and can expose users to untrusted pages, tracking, or phishing-style content during the PDF export process.
The script reads a user-controlled url query parameter, decodes it, and assigns it directly to iframe.src, enabling arbitrary navigation to external content. This creates an open content-loading primitive that can be abused to render attacker-controlled pages, perform unexpected network access, or display deceptive content under the skill's UI.
The document sets lang="zh-CN", and the visible UI text is also exclusively in Chinese, which imposes a specific language/locale on all users. The file does not provide any opt-in, alternate locale handling, or justification that this skill is intentionally region-specific.