Your agent can’t can edit Office documents.

The skills

• .pdf

  Read, build, and fill. Extract text and tables, merge and split, create from scratch, OCR scanned pages, and fill forms two ways: native form fields, or drawn annotations validated against a rendered image.

• .docx

  Write and redline. Create documents from JavaScript, edit the raw OOXML, add comments, and produce genuine tracked changes for legal and academic review.

• .pptx

  Design and rearrange. Build decks from HTML with pixel-accurate positioning, reorder and edit existing slides, and render thumbnails so the agent can check its own work.

• .xlsx

  Model and recalculate. Create and edit spreadsheets while preserving formulas, then recalculate through headless LibreOffice and get back every #REF! and #DIV/0! by cell address.

Set up a local checkout

This walkthrough uses macOS with Git and Homebrew installed. For Linux system packages, see the repository dependencies.

1. Clone the repository and enter it. The remaining commands run from this directory, which contains the Brewfile and all four skills.

   git clone https://github.com/appautomaton/document-SKILLs.git
   cd document-SKILLs

2. Install the system tools. The Brewfile provides Poppler, Pandoc, and LibreOffice. It skips uv and Node when they are already on your PATH.

   brew bundle --no-upgrade

   For scanned-PDF OCR, also install tesseract with Homebrew. HTML-to-slide rendering on macOS uses Google Chrome.

3. Install the Node packages used by Word and PowerPoint workflows. PDF and spreadsheet workflows do not need these packages.

   npm --prefix docx install
   npm --prefix pptx install

   Python dependencies resolve when each script runs through uv run.

4. Link the skills into your agent’s skills directory. This example uses Codex. For Claude Code, set skills_dir to "$HOME/.claude/skills" instead. Existing entries are reported and kept.

   skills_dir="$HOME/.codex/skills"
   mkdir -p "$skills_dir"
   for skill in docx pdf pptx xlsx; do
     skill_target="$skills_dir/$skill"
     if [ -e "$skill_target" ] || [ -L "$skill_target" ]; then
       printf 'Already exists: %s\n' "$skill_target"
     else
       ln -s "$PWD/$skill" "$skill_target"
     fi
   done

5. Check the dependencies from the repository root, then ask your agent to use a skill with one of your documents.

   brew bundle check
   npm --prefix docx ls --depth=0
   npm --prefix pptx ls --depth=0

   For example: “Use the docx skill to replace this paragraph with the revised text and return a Word file with tracked changes.”

If you use the plugin installation above, follow the prerequisites in its installed docx and pptx skills. Their Node packages must be installed in the plugin’s skill directories.

How it’s built