office-files

Setup — run these two lines first

pip install --break-system-packages --quiet pypandoc_binary typst openpyxl python-docx
export PATH="$(python3 -c 'import pypandoc,os;print(os.path.dirname(pypandoc.get_pandoc_path()))'):$PATH"

Takes about 6 seconds; nothing is preinstalled. Both lines are required — the system Python is PEP 668 externally managed, and the wheel ships the pandoc binary inside the package directory rather than on PATH.

Install only what the flow needs, and know which package is which. pypandoc_binary and typst cover the PDF flow and the prose-to-docx render; openpyxl is only for xlsx; python-docx is only for the docx steps that build or edit a reference doc. Installing just the first two is fine — until a later docx or xlsx step in the same session dies with ModuleNotFoundError: No module named 'docx' (or openpyxl). Install all four unless the task certainly never touches docx or xlsx.

export PATH lasts for one shell only, so re-run it in every new shell. Each Bash call starts a fresh shell, so a single setup at the top of a run does not carry forward and later calls fail with pandoc: command not found. Prefix the export line to each command that uses pandoc, or re-export it at the start of every shell.

If the install fails, deliver Markdown or a hosted HTML view instead and say the toolchain was unavailable. Never ship a worse format without saying so.

Pick your flow

Every deliverable is rendered from a source you author. You never write a .docx or a .pdf directly: