docxreview

docxreview extracts comments and tracked changes from reviewed Word documents (.docx) and returns them as structured data or formatted Markdown. It is designed for workflows where documents are rendered from Quarto or R Markdown, reviewed in Microsoft Word, and feedback needs to be processed programmatically — for example by passing the structured output to an LLM for automated revision.

Installation

Install the development version from GitHub:

# install.packages("remotes")
remotes::install_github("SchmidtPaul/docxreview")

Quick Start

library(docxreview)

# Get a formatted Markdown summary of all review feedback
extract_review("report_reviewed.docx")

# Save to file (anonymized — replaces reviewer names with "Reviewer 1" etc.)
extract_review("report_reviewed.docx", output_file = "feedback.md",
               anonymize = TRUE)

# Programmatic access as tibbles
comments <- extract_comments("report_reviewed.docx")
changes  <- extract_tracked_changes("report_reviewed.docx")

Workflow

The typical use case is a Quarto/R Markdown review cycle:

flowchart LR
    A[".qmd Source"] -->|"quarto render"| B[".docx v1"]
    B -->|"send to reviewer"| C["Reviewer edits in Word"]
    C -->|"reviewed .docx"| D["docxreview extracts feedback"]
    D -->|"structured Markdown"| E["Author revises source"]
    E -->|"quarto render"| F[".docx v2"]
    B & F -->|"compare_versions"| G["diff .docx"]

Loading

1. Render your Quarto or R Markdown document to .docx.
2. Send the .docx to a reviewer who adds comments and tracked changes in Microsoft Word.
3. Extract the feedback with extract_review() to get a structured Markdown summary of all comments, insertions, and deletions — including paragraph context.
4. Revise your source document based on the extracted feedback.
5. Compare old and new renders with compare_versions() to produce a diff .docx with tracked changes — so the reviewer can see exactly what changed.

LLM-Assisted Workflow

The structured Markdown output from extract_review() can be passed to any LLM to automate source revisions. A prompt like the following works with ChatGPT, Claude, Gemini, or any other LLM that has access to your project files:

"Here is the review feedback extracted from the reviewed DOCX. Please revise report.qmd to address all comments and tracked changes."

Claude Code Integration

docxreview ships with a Claude Code skill file at inst/skills/docxreview/SKILL.md. When used together with btw as an MCP server, Claude Code can extract and apply review feedback in a single step. Register the skill in your project's .claude/settings.json:

{
  "permissions": {
    "allow": []
  },
  "skills": [
    "/path/to/docxreview/skills/docxreview/"
  ]
}

Find the installed path with:

system.file("skills", "docxreview", package = "docxreview")

Functions

Function	Returns	Description
extract_review()	Markdown (character)	Full formatted summary of all comments and tracked changes. Supports group_by_author and anonymize options.
extract_comments()	tibble	Comments with author, date, comment text, commented text, and paragraph context
extract_tracked_changes()	tibble	Insertions and deletions with author, date, changed text, and paragraph context
compare_versions()	file path	Produces a diff .docx with tracked changes between two document versions (requires Word + Python)

See vignette("docxreview") for the full workflow documentation.