Arabic Word Documents (RTL DOCX)

AI agents write Arabic Word files that look fine in the chat and break the moment someone opens them: paragraphs start at the left margin, the first table column lands on the wrong side, bold Arabic stays regular, PostgreSQL و collapses into PostgreSQLو, and C# turns into #C. None of it shows up in the text, so the agent never notices.

This skill gives your agent a tested Python library (python-docx) that writes every direction flag Word needs, and the rules for when to use each one.

What it handles

• Direction everywhere: paragraphs, runs, tables, sections and document defaults, so text the reader types later is RTL too.
• Mixed Arabic and English: text is split by script automatically. Brackets, version numbers and terms like C#, C++, .NET, +963 stay intact.
• Formatting that actually reaches Arabic: bold, italic and size go into Word's complex-script slots, and template styles are cleaned so headings and the table of contents don't fall back to Calibri.
• Full reports: real headings, a table of contents with page numbers, numbered figure and table captions, page numbers that skip the cover, and tables with fixed column widths and a header row that repeats on every page.
• Existing files: audit_docx() lists what a reader would see wrong; fix_docx() repairs a document made elsewhere; insert_before() adds a chapter without rebuilding the file and losing manual edits.
• Visual check: preview() converts through Microsoft Word (or LibreOffice) and renders every page to PNG so the agent can look before delivering.

Works for Arabic, Persian, Urdu and Hebrew. Free and open source (MIT).