The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/export_rich.py executes external system commands using the subprocess.run function.
It calls the pandoc utility to perform document format conversions (e.g., HTML to Markdown, Markdown to HTML/DOCX).
Commands are constructed as lists with fixed arguments, which prevents shell-level command injection.
[DYNAMIC_EXECUTION]: The skill utilizes the playwright library to launch a Chromium browser for rendering HTML content into PDF files.
This involves executing a high-capability browser environment to process content, which is a standard part of its PDF generation functionality.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a content processing pipeline for external documents, creating an attack surface for indirect prompt injection via untrusted inputs.
Ingestion points: Processes .md, .html, and .json files provided via the --input argument in scripts/export_rich.py.
Boundary markers: SKILL.md contains explicit instructions ('先做内容边界') directing the agent to strictly separate internal context (like names, project backgrounds, or system prompts) from the exported content.
Capability inventory: Includes file system read/write access, ZIP archive creation, system command execution for document conversion, and browser-based rendering.
Sanitization: The implementation uses html.escape for media metadata and employs regex-based projections to convert interactive media (video/audio/iframe) into static descriptions for non-HTML formats.

lov-rich-export