word2latex

name	word2latex
description	Convert a Word .docx template (Mitacs, CRSNG, FRQNT, UQAC, partner forms) into a faithful LaTeX source. Use when the user asks to translate a Word gabarit to LaTeX, when a generated .tex must match a .docx visually, or for /word2latex. Drives pandoc + applies the standard patch sequence (Arial, full-grid tables, centered sections, landscape geometry, first-page banner). Delegates the actual patch work to the word-to-latex agent.
allowed-tools	["Read","Write","Edit","Bash"]
permissions	["read","write"]

word2latex Skill

Prerequisites

Before running, verify:

1. pandoc installed: pandoc --version
2. python-docx available (used by docx_inspect.py): python -c "import docx; print('OK')" — if missing, pip install python-docx
3. pdflatex and pdftoppm reachable from the shell (any TeX distribution + poppler)
4. The .docx file is closed in Word (otherwise the ~$ lock file blocks reads)

Mode dispatch

$ARGUMENTS is the path to the .docx file (or a directory containing exactly one .docx).

/word2latex <path-to-docx>

Optional flags appended after the path:

Flag	Effect
--engine=pdflatex (default)	Use helvet to approximate Arial
--engine=xelatex	Use fontspec with native Arial
--no-banner	Skip the first-page TikZ banner block (use for plain documents without a colored header strip)
--legal-landscape	Enable the Legal-landscape macros (14×8.5 in) — needed only if the docx has <w:pgSz w:w="20160" w:h="12240">
--bibliography	Manuscript mode: extract the numbered bibliography to a first-pass .bib, rewrite inline citations to \cite, then validate via /bibclean + /scopus. Opt-in; off for form gabarits.

Workflow

Step 1 — Inspect the .docx

Run the inspector to extract every value the LaTeX preamble needs:

cd "<docx directory>"
python "<workspace>/.claude/skills/word2latex/scripts/docx_inspect.py" <file.docx>

The script prints a structured report with:

• Default font + size
• Line spacing
• Each <w:sectPr> (orientation, page size, margins, header/footer references)
• Heading1 / Heading2 / Title styles (font, size, color, alignment)
• Table border defaults
• Image catalog (path, dimensions, dominant color, transparency stats)
• Header and footer text per section

Save the report — every preamble decision below traces back to it.

Step 2 — Run pandoc

pandoc "<file.docx>" -o "<file>.tex" --extract-media="img_<file>"

This produces a .tex with longtables, hyperlinks, and image references in place but without the visual fidelity adjustments. Do not edit the pandoc output until step 3.

Step 3 — Delegate to the word-to-latex agent

Hand off to the agent with the inspector report as context:

Use the word-to-latex agent to patch <file>.tex against the inspector report above.
Apply the full standard patch sequence in order. The banner uses image<N>.png
(white wordmark) on the left; pages <M>–<P> are landscape <orientation>.

The agent applies the 10-step patch sequence documented in .claude/skills/word2latex/references/preamble_patches.md.

Step 3b — Manuscript bibliography (optional, --bibliography only)

Skip this step for form gabarits. Run it only when the .docx is a manuscript (numbered references + inline citations) and --bibliography was passed.

1. Second pandoc pass to markdown (keeps explicit 1. numbering):

   pandoc "<file.docx>" -t gfm -o "<file>.md"
   

2. Extract the bibliography and rewrite citations:

   python "<workspace>/.claude/skills/word2latex/scripts/manuscript_bib.py" `
       "<file>.tex" --bib-source "<file>.md"
   

   Produces <file>.cited.tex (citations as \cite{...}) and <file>.first-pass.bib.
3. Validate and enrich the bibliography (respects the project reference rules — approved publishers, clickable DOI):

   /bibclean <file>.first-pass.bib
   /scopus validate <file>.first-pass.bib
   

4. The agent (Step 3) then injects the biblatex preamble and replaces the pandoc-rendered reference list with \printbibliography.

Full details: .claude/skills/word2latex/references/manuscript_bibliography.md.

Step 4 — Compile and render

pdflatex -interaction=nonstopmode -output-directory=out "<file>.tex"
pdflatex -interaction=nonstopmode -output-directory=out "<file>.tex"
Remove-Item out\page-*.png -ErrorAction SilentlyContinue
pdftoppm -png -r 120 out\<file>.pdf out\page

Two pdflatex passes are required for \pageref{LastPage} and longtable column widths.

Step 5 — Visual verification

Read at minimum:

• Page 1 — banner, white wordmark on left, title centered, icon on right, no duplicate header
• Each landscape page — text + table on the same page, full grid visible
• The last page — footer reads correctly, no orphan content from a previous section

If any page does not match, return to the agent with the specific page index and the discrepancy.

Output formatting

Final response to the user lists:

1. The .tex file path
2. The page count of the resulting PDF
3. For each landscape page: dimensions in inches
4. Any unresolved overfull \hbox ≥ 5 pt (smaller are absorbed by \hfuzz=3pt)
5. The verification rendering folder (out/page-*.png)

Error handling

Error	Action
pandoc: command not found	Ask user to install pandoc or add it to PATH
python-docx missing	pip install python-docx then retry
~$<file>.docx lock present	Ask user to close the document in Word
! Misplaced \noalign	A \toprule/\midrule was missed by the rule-replacement pass — re-run replace booktabs rules with \hline
Command \endLegalLand already defined	Macro name collides with \begin{LegalLand} / \end{LegalLand} semantics — rename to \Open.../\Close...
paperwidth not available in \newgeometry	Geometry forbids paper-size keys mid-document — set \pdfpagewidth and \paperwidth before the \newgeometry call
Calendar table spills to a second page	Apply row-compaction: replace minipage+enumerate cells with \textbf{N) text}, set \renewcommand{\arraystretch}{0.95}, use \footnotesize
Right header text missing	Header logo image has no width constraint and pushes the right slot off-page — set width=3.5cm,keepaspectratio on the \includegraphics in \fancyhead[L]
Entries : 0 from manuscript_bib.py	The bibliography is not numbered as 1./2. in the source — run the -t gfm markdown pass and pass it via --bib-source
Unresolved cites > 0 after validation	A \cite{refN} survived — the source citation number has no matching numbered reference; fix the source numbering or the reference list

See also

• Agent: .claude/agents/word-to-latex.md
• Patch reference: .claude/skills/word2latex/references/preamble_patches.md
• Manuscript bibliography reference: .claude/skills/word2latex/references/manuscript_bibliography.md
• Inspector script: .claude/skills/word2latex/scripts/docx_inspect.py
• Bibliography module: .claude/skills/word2latex/scripts/manuscript_bib.py