Agent skills for PDF, Word, Excel, and PowerPoint files

Ask an agent for a deliverable and you used to get Markdown or a hand-built HTML file. Agents now have skills: packaged instructions plus the tools to produce a specific kind of output, shared across chat, the workflow builder, and agent workflow runs.

The skills

• PDF: extract text and tables, merge, split, watermark, run OCR on scanned pages, and create new PDFs.
• Word, Excel, PowerPoint: create and read .docx, .xlsx (with formulas and styling), and .pptx files, and convert any of them to PDF.
• Brand guidelines: pull a company's palette, fonts, and logo from its website and apply them consistently.
• Email design: HTML email that holds up in Outlook, dark mode, and Gmail's 102 KB clipping limit.

How it works

• Loaded on demand. Agents see a one-line description of each skill and load the full instructions only when a task needs them, which keeps prompts short and cacheable.
• Ready to run. The sandbox ships with the document toolchain (pypdf, python-docx, openpyxl, python-pptx, pandas, LibreOffice, Tesseract) and 48 fonts preinstalled, so the first PDF in a run doesn't wait on installs.
• Inline previews. Files and images an agent creates render right in the chat.