canghe-markdown-to-html
Warn
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill dynamically loads and executes JavaScript modules from a remote, unverified CDN (
cdn-doocs.oss-cn-shenzhen.aliyuncs.com) to support code syntax highlighting. This is implemented inscripts/md/utils/languages.tsusing theimport()function with computed URLs.- [EXTERNAL_DOWNLOADS]: - The skill downloads files from the
cdn-doocs.oss-cn-shenzhen.aliyuncs.comCDN at runtime. - It downloads remote images from arbitrary URLs found in user-provided Markdown content in
scripts/main.ts.- [COMMAND_EXECUTION]: The skill relies onnpx -y bunto run its internal logic. This command triggers the automatic download and execution of thebunruntime if it is missing from the environment.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted Markdown content and generates HTML without explicit sanitization. - Ingestion points: Reads input Markdown files as specified in
SKILL.mdandscripts/main.ts. - Boundary markers: None identified.
- Capability inventory: Network access for image retrieval and file system writes for HTML and backup creation.
- Sanitization: The skill uses the
markedlibrary but does not filter out or sanitize embedded raw HTML tags.- [DATA_EXFILTRATION]: - The generated HTML output exposes local file system paths through
data-local-pathattributes for every processed image. - Hardcoded local paths from the developer's environment (e.g.,
/Users/jimliu/...) are present in the rendering logic inscripts/md/render.ts, which may leak metadata about the original development environment.
Audit Metadata