Published: Mon 07 July 2025
By Joseph Reagle
In praxis .
tags: pandoc ai
When I ask people to give me feedback, I’d like them to work with
whatever format or app is most convenient to them. I write everything in
markdown, often in Sublime Text and sometimes in Obsidian. Many people
prefer reviewing in Word or GDocs. Using pandoc, I can create most any
file format, but getting others’ annotations back into my markdown
source files has never been easy. This task is now easier with AI.
I use pandoc to generate a Word
docx version, emailed or placed on Google Drive.
The reviewer annotates using Word, Google Docs, OpenOffice,
etc.
I use the docx2md_add_comment.lua
filter to convert the annotated docx file back to markdown.
I Ask Claude Opus 4 Thinking to port the comments from the
feedback file to my source file, which takes 5–10 minutes, with this
prompt:
Someone added html/markdown comments to the file
06-ai-advice-feedback-smith.md
I need you to find Smith’s
comments and port them to the original markdown file
06-ai-advice.md
(keeping them as markdown comments).
To make sure I don’t miss any comments, it’s easy to count the number
of comments or to diff the files.
Round-tripping a reviewers’ granular edits would be more difficult,
especially if I sent them a version of the document with formatting,
citations, and footnotes rendered. It wouldn’t be as bad if I sent them
my source markdown plunked into a docx file, but I don’t think AI is up
to the task to tracking small diffs between a source and rendered
version of a file. For annotations, however, the above workflow works
well!
There are comments .