make_latex_model_skill

Overview

This skill provides high-fidelity LaTeX template optimization focused on matching an official Word template at pixel level. It aligns style parameters, title text/numbering/punctuation, and per-line line-breaks, and produces validation reports and automatic repair suggestions. It is designed for Chinese academic templates but works with any LaTeX template requiring strict parity with a Word baseline.

How this skill works

The skill combines deterministic scripts and AI-driven heuristic decisions: scripts extract styles, generate Word-based baseline PDFs, run LaTeX builds, and perform pixel-level PDF comparisons. The AI suggests which style parameters to adjust (font sizes, spacing, kerning, color, numbering formats) and iterates until convergence or a configured stop condition. Outputs include an HTML visual report, suggested edits to @config.tex and safe title-text updates in main.tex, plus optional dry-run previews.

When to use it

• A new annual Word template for NSFC or similar is released
• Existing LaTeX template visibly differs from the Word template
• You must guarantee title texts and numbering exactly match Word
• You require pixel-level verification between Word PDF and LaTeX PDF
• You need automated, traceable optimization with rollback/dry-run support

Best practices

• Start with the recommended pre-check and validation scripts before making changes
• Use moderate optimization by default; escalate to thorough only when necessary
• Keep changes confined to projects/{project}/extraTex/@config.tex and allowed title text in main.tex
• Preserve original comments, conditionals (e.g., \ifwindows) and macro structure when editing
• Run iterative optimization with max iterations and convergence checks; use dry_run for previews

Example use cases

• Align NSFC_Young LaTeX template to the 2025 Word template including title punctuation and numbering
• Automatically tune font sizes, line spacing and margins so each line breaks identically to the Word PDF
• Generate an HTML visual diff report and pixel-change metrics for submission evidence
• Run an AI-assisted iteration loop to converge to changed_ratio < 0.01 between Word and LaTeX PDFs
• Produce a hotfix that only updates section title texts to match a revised Word template