The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/render_svg_tiles.py executes system commands (qlmanage and ImageMagick) to perform image conversion. It uses subprocess.run with argument lists (shell=False), which is a secure way to call external processes.
[INDIRECT_PROMPT_INJECTION]: The skill processes content fetched from external WPS and ProcessOn documents. It includes specific hard rules and instructions for the agent to treat this as raw source data rather than instructions to be followed, minimizing the risk of indirect prompt injection.
Ingestion points: Document metadata and definitions are fetched from user-provided URLs in scripts/wps_processon_extract.py.
Boundary markers: SKILL.md contains explicit "Hard Rules" and instructions to avoid editorial rewriting or inference.
Capability inventory: File system writes for archiving and shell execution for image rendering.
Sanitization: Content is normalized and HTML tags are stripped via regex in the clean_text function in scripts/wps_processon_extract.py.
[EXTERNAL_DOWNLOADS]: The skill communicates with official WPS and ProcessOn endpoints (drive.kdocs.cn, wps.processon.com) to retrieve document data for archiving purposes.

wps-doc-scraper