Document Processing Skills

English | 中文

A set of improved SKILLs from the Official Anthropic skills for Word, PDF, PowerPoint, and Excel manipulation in Claude Code and Codex.

How to Use

Option 1: Git Subtree

# Replace <skills_dir> with your agent's skills path
git subtree add --prefix=<skills_dir> \
  https://github.com/appautomaton/document-SKILLs.git master --squash

Option 2: Clone and Copy

git clone https://github.com/appautomaton/document-SKILLs.git /tmp/doc-skills
mkdir -p <skills_dir>
cp -r /tmp/doc-skills/{docx,pdf,pptx,xlsx} <skills_dir>/
cp /tmp/doc-skills/requirements.txt <skills_dir>/requirements.txt
rm -rf /tmp/doc-skills

If you already have a requirements.txt in your skills directory, merge instead of overwriting. requirements.txt lists the Python dependencies used by these skills. Recommended: use uv and add a “use uv” requirement to your own agent rules (adapt to your environment/install method).

Dependencies (Install Once)

Note

Use an isolated environment where possible; uv is the recommended installer.

# System packages
sudo apt-get install -y pandoc libreoffice poppler-utils tesseract-ocr

# Optional (PDF CLI workflows)
sudo apt-get install -y qpdf

# Python packages (recommended)
uv venv
source .venv/bin/activate
uv pip install -r <skills_dir>/requirements.txt

# NPM packages
npm install -g docx pptxgenjs playwright sharp react react-dom react-icons
npx playwright install chromium

Codex MCP (Playwright)

This repo is set up to use an MCP Playwright server for browser-backed rendering. Example Codex config ($CODEX_HOME/config.toml or ~/.codex/config.toml):

[mcp_servers.playwright]
command = "npx"
args = ["-y", "@playwright/mcp@latest", "--browser", "chromium", "--headless", "--no-sandbox", "--user-data-dir", "/root/.cache/ms-playwright/mcp-chromium-profile"]
startup_timeout_sec = 60

Important

Adjust --user-data-dir to a writable location for your environment.

Output Organization

Tip

Add outputs/ to your .gitignore to keep generated files out of version control.

Organize all generated files in a dedicated outputs/ directory:

outputs/
└── <document-name>/           # One folder per document project
    ├── final.pptx             # Final output file
    ├── inventory.json         # Intermediate files
    ├── replacements.json
    ├── unpacked/              # OOXML extraction directory
    ├── thumbnails/            # Visual validation images
    └── images/                # Generated assets

Naming Convention

Use descriptive, lowercase, hyphenated names:

• outputs/quarterly-report/
• outputs/client-proposal/
• outputs/budget-2024/

Example Workflow

# Create output directory
mkdir -p outputs/sales-deck/

# Generate presentation
python pptx/scripts/inventory.py template.pptx outputs/sales-deck/inventory.json
python pptx/scripts/replace.py template.pptx outputs/sales-deck/replacements.json outputs/sales-deck/final.pptx

# Validate visually
python pptx/scripts/thumbnail.py outputs/sales-deck/final.pptx outputs/sales-deck/thumbnails/

Source

Based on Anthropic's official skills.