The capybara sat down at the typewriter. When it stood up, there was a thesis.
Turn handwritten notes, scanned textbooks, and raw data into publication-ready LaTeX -- without knowing a single LaTeX command.
27 templates · 27 automation scripts · 26 reference guides · 4 OCR profiles · 217 tests · 0 LaTeX commands required
--- ## The 10-Second Pitch You describe a document in plain English. This skill produces a compiled PDF. - An 80-page handwritten math PDF becomes color-coded lecture notes with proper equations and TikZ diagrams. - A 162-page textbook becomes a 2-page cheat sheet. - A CSV becomes nine chart types. - A one-line prompt becomes a thesis, a resume, a conference poster, or a 37-page book with drop caps. No LaTeX knowledge required. The capybara handles the semicolons. --- ## What Can It Actually Do? *(Spoiler: more than most humans with a PhD in LaTeX.)* | You say... | What happens under the hood | |---|---| | "Create my resume" | Selects from 5 ATS-optimized templates, compiles with `pdflatex`, generates PDF + PNG preview | | "Convert my 80-page handwritten math notes into beautiful LaTeX" | `pdf_to_images.sh` renders at 200 DPI -> batch-7 parallel OCR agents -> `math-notes.md` profile generates colored `tcolorbox` theorems (blue), definitions (green), examples (orange) -> `compile_latex.sh` runs multi-pass pdflatex -> polished PDF with proper equations, proofs, and TikZ diagrams | | "Turn this 162-page scan into a 2-page cheat sheet" | `pdf_to_images.sh` splits PDF -> vision OCR per page -> extracts key content -> symbol substitution -> telegram-style compression -> fits into `cheatsheet.tex` with 3-column landscape 7pt layout | | "Build a quarterly report with charts" | `generate_chart.py` creates bar/line/pie charts from JSON/CSV -> `csv_to_latex.py` converts data into `booktabs` tables -> Mermaid/TikZ flowcharts compiled inline -> all embedded in `report.tex` with TOC | | "Generate 9 charts from my sales CSV" | `generate_chart.py` reads CSV -> outputs bar, line, scatter, pie, heatmap, box, histogram, area, radar charts -> multi-series, legends, colorblind-safe Tol palette | | "Convert my old PDF to LaTeX" | Pages split at 200 DPI -> parallel OCR agents -> clean `.tex` with profile-tuned formatting -> 0 errors per 7-page batch (validated on 115-page PDF) | | "Send personalized letters to 500 candidates" | `mail_merge.py` loads template + CSV -> Jinja2 rendering -> 4 parallel `pdflatex` workers -> `qpdf --pages` merge into single PDF | | "What changed between v1 and v2 of my thesis?" | `latex_diff.sh` runs `latexdiff` -> highlighted change-tracked PDF with additions in blue, deletions in red | | "Make a NeurIPS poster" | Interactive: asks orientation -> layout -> color scheme -> generates A0 `tikzposter` with correct dimensions and QR codes | | "Create a calculus final exam with answer key" | `exam` class with `\printanswers` toggle, grading table, 6 question types (MCQ, T/F, fill-blank, matching, short, essay) | | "Fetch BibTeX for these DOIs" | `fetch_bibtex.sh` hits `doi.org` with `Accept: application/x-bibtex` header -> appends to `.bib` -> cross-refs every `\cite{}` key | | "Password-protect this report" | `pdf_encrypt.sh` applies AES-256 with print/modify restrictions | | "Fill this PDF form with my data" | Detects fillable fields -> validates -> fills text, checkbox, radio, choice fields -> completed PDF | | "Fill this non-fillable PDF application" | Renders pages -> visual analysis to identify fields -> bounding box validation -> FreeText annotations at precise positions | | "Check if my paper is ready to submit" | Verifies packages -> cross-refs citations -> counts figures/tables/equations -> runs chktex for style issues | ---
Handwritten chaos in, publication-ready PDF out. It's not magic. It's 27 shell scripts.
--- ## Template Gallery -- 27 Templates Every template has been compiled, tested, and verified to produce zero errors. The capybara is a perfectionist. ### Resumes -- 5 ATS-Optimized + 1 Legacy All 5 modern templates designed to pass Applicant Tracking Systems (no columns, no tables for layout, no graphics in header, machine-readable text): | Template | Target Role | Key Design Choices | |---|---|---| | `resume-classic-ats.tex` | Finance, law, government | Single-column, minimal styling, maximum parsability | | `resume-modern-professional.tex` | Tech, corporate | Clean sections, subtle color accents | | `resume-executive.tex` | VP / Director / C-suite | Multi-page, executive summary section | | `resume-technical.tex` | Software, data, engineering | Skills matrix, project highlights | | `resume-entry-level.tex` | New graduates | Education-first, coursework, activities | | `resume.tex` (legacy) | Regions requiring photos | Photo area -- **not** ATS-compatible | | | | | | |---|---|---|---| |  |  |  |  | | Classic ATS | Modern Professional | Executive (p1) | Executive (p2) | |  |  | | | | Technical | Entry-Level | | | --- ### Academic Documents #### Thesis / Dissertation -- Full book-class, 38+ pages Palatino fonts, `microtype`, `mathtools`, `cleveref`. Front matter (title, declaration, abstract, acknowledgments, TOC), multiple chapters, appendices, bibliography with `biblatex`/biber. `\geometry{bindingoffset=1.5cm}` for professional printing. | | | | | |---|---|---|---| |  |  |  |  | | Title Page | Table of Contents | Literature Review | TikZ Diagram | |  |  |  |  | | Results | Charts | Chapter Content | Bibliography | #### Academic Paper -- 11 pages, arXiv-compatible Times fonts, colorblind-safe Tol palette, multi-author affiliations via `authblk`, `siunitx` for consistent units, algorithm environments, theorem/proof. `\pdfoutput=1` for arXiv submission. | | | | | |---|---|---|---| |  |  |  |  | | Title & Abstract | Tables + Charts | Ablation Study | References | #### Lecture Notes (Beautiful Mode) -- Color-coded theorem environments `lecture-notes.tex`: Palatino fonts, `tcolorbox` with semantic colors -- blue theorems, green definitions, orange examples, purple remarks. TikZ graph theory macros, custom math operators (`\E`, `\Var`, `\Cov`). | | | | | |---|---|---|---| |  |  |  |  | |  |  |  |  | #### Academic CV -- Multi-page with publications, grants, teaching Numbered publications ([J1], [C1], [W1]), grants with dollar amounts, student advising (current + graduated), professional service, invited talks. ORCID and Google Scholar links. | | | | | |---|---|---|---| |  |  |  |  | #### Homework / Assignment -- Solution toggle Custom `problem`/`solution` environments, code listings (Python, Java, C++, Matlab), honor code section. Toggle solutions globally with `\showsolutionstrue`/`\showsolutionsfalse`. #### Lab Report -- STEM lab writeups `siunitx` for uncertainties and SI units, `pgfplots` for data with error bars, structured sections: abstract -> theory -> procedure -> data -> analysis -> discussion -> conclusion. --- ### Scientific Posters -- `tikzposter` class Interactive workflow asks: conference -> orientation -> layout -> color scheme. | Layout | Usage | Description | |---|---|---| | Traditional Column | ~70% of posters | 2-column (portrait) or 3-column (landscape) | | #BetterPoster | ~10% (growing) | Central billboard with ONE key finding at 60-80pt | | Visual-Heavy | ~15% | Large central figure consuming 40-50% of space | Conference presets: NeurIPS, ICML, CVPR, ICLR (main + workshop sizes). | Portrait A0 | Landscape A0 | |---|---| |  |  | --- ### Book -- Full-Length Publishing Template `book` class, 37+ pages. Palatino fonts, `lettrine` drop caps, `imakeidx` for indexing. Structure: half-title -> full title -> copyright page (ISBN slot) -> dedication -> preface -> acknowledgments -> TOC -> parts -> chapters with epigraphs -> appendices -> bibliography -> index -> colophon. | | | | | | |---|---|---|---|---| |  |  |  |  |  | | Half Title | Full Title | Copyright | TOC | Preface | |  |  |  |  |  | | Acknowledgments | Part I | Ch 1: Drop Caps | Definitions & Theorems | Notation & Summary | --- ### Exam / Quiz -- `exam` class 6 question types (multiple choice, true/false, fill-in-blank, matching, short answer, essay). Point values per question, `\gradetable[h][questions]` for grading grid. `\printanswers` / `\noprintanswers` toggles solution visibility. | | | | | | | |---|---|---|---|---|---| |  |  |  |  |  |  | --- ### Cheat Sheets -- 3 Variants | Template | Columns | Font | Orientation | |---|---|---|---| | `cheatsheet.tex` | 3 | 7pt | Landscape | | `cheatsheet-exam.tex` | 2 | 6pt | Portrait | | `cheatsheet-code.tex` | 4 | 7pt | Landscape | **Example: Algebraic Geometry (162 pages -> 2 pages)** | | | |---|---| |  |  | | Page 1 -- Affine/Projective Space, Varieties, Morphisms | Page 2 -- Sheaves, Schemes, Cohomology, Key Examples | --- ### Interactive / Dynamic Templates | Template | What It Does | |---|---| | `fillable-form.tex` | PDF form fields: text inputs, checkboxes, radio buttons, dropdowns, push buttons (via `hyperref`) | | `conditional-document.tex` | 12 `etoolbox` toggles (showTOC, isDraft, isConfidential...), 3 visual profiles, CLI-overridable | | `mail-merge-letter.tex` | Template for `mail_merge.py` -- `{{name}}` placeholders | --- ### Business Documents | | | | |---|---|---| |  |  |  | | `letter.tex` -- Business letter | `cover-letter.tex` -- Job application | `invoice.tex` -- Professional invoice | --- ### Presentation (`presentation.tex`) -- Beamer 16:9 Custom theme, widescreen aspect ratio, title/section/content/two-column/code/image/thank-you frame types. | | | | | | |---|---|---|---|---| |  |  |  |  |  | |  |  |  |  |  | --- ### Report (`report.tex`) Executive summary, findings, recommendations with TOC, pgfplots bar charts, TikZ flowcharts, colored data tables. | | | | | |---|---|---|---| |  |  |  |  | --- ## Killer Features *(The things that make this more than "just another template repo.")* ### 1. Smart Compilation Engine (`compile_latex.sh` -- 525 lines) Not a wrapper around `pdflatex`. It is an intelligent build system that reads your document and makes decisions. Think of it as a capybara that learned to read `.tex` files. **Auto-detection logic:** ``` Document contains \usepackage{fontspec} or \usepackage{xeCJK} -> xelatex Document contains \usepackage{luacode} or \directlua -> lualatex Otherwise -> pdflatex Document contains \bibliography{} -> bibtex Document contains \addbibresource{} -> biber Document contains \makeindex -> runs makeindex Document contains \makeglossaries -> runs makeglossaries ``` **Multi-pass compilation:** Runs the engine up to 3 times, checking for "Rerun to get cross-references right" each pass. Bibtex/biber runs between passes 1 and 2. **Auto-fix mode** (`--auto-fix`): | What it fixes | How | |---|---| | Naked `\begin{figure}` without placement | Injects `[htbp]` -- the #1 LaTeX beginner complaint | | Naked `\begin{table}` without placement | Same `[htbp]` injection | | Overfull hbox warnings | Adds `\usepackage{microtype}` to preamble | **Error translation** -- parses `.log` and translates into actionable English: | Raw LaTeX Error | What the script tells you | |---|---| | `Missing $ inserted` | "Math symbol used outside `$...$` -- wrap it in dollar signs" | | `Undefined control sequence \xyz` | "Unknown command `\xyz` -- check spelling or add the right `\usepackage`" | | `Too many }'s` | "Extra closing brace -- count your `{` and `}` pairs" | | `File 'foo.sty' not found` | "Missing package `foo` -- install with `tlmgr install foo`" | **Preview generation:** After compilation, runs `pdftoppm -png -r 200` on the output PDF and resizes to <=2000px. ```bash bash scripts/compile_latex.sh document.tex --preview --auto-fix ``` --- ### 2. PDF-to-LaTeX Reconstruction (Vision OCR Pipeline) Convert **any PDF** -- scanned textbook, handwritten notes, printed report -- into compilable LaTeX. The capybara reads your handwriting better than your professor does. **Pipeline:** 1. `pdf_to_images.sh` renders pages at 200 DPI using `pdftoppm`, zero-pads filenames, resizes to <=2000px for API limits 2. Vision model reads each page image 3. Generates `.tex` with profile-appropriate formatting **Scaling strategy** (empirically validated on 115-page handwritten math notes): | PDF Size | Strategy | Expected Errors | |---|---|---| | 1-10 pages | Single agent processes all pages | 0-2 minor errors | | 11-20 pages | Split into 2 batches of ~10 | Near-zero per batch | | 21+ pages | Batch-7 pipeline with parallel agents | 0 errors per 7-page batch | **4 conversion profiles** (in `references/profiles/`): | Profile | Tuned For | |---|---| | `math-notes.md` | Equations, theorems, proofs. "Beautiful mode" uses colored `tcolorbox` environments | | `business-document.md` | Reports, financials, tables, bullet lists | | `legal-document.md` | Numbered paragraphs, statutory references, legal citation formatting | | `general-notes.md` | Handwritten notes, mixed media, letters | --- ### 3. PDF-to-Cheat Sheet Pipeline Condense entire textbooks into 2-page reference cards. Because the night before the exam, you need density, not decoration. | Template | Layout | Font Size | Columns | Best For | |---|---|---|---|---| | `cheatsheet.tex` | Landscape A4 | 7pt | 3 | Course reference, concept summaries | | `cheatsheet-exam.tex` | Portrait A4 | 6pt | 2 | Exam formula sheets (B&W-safe) | | `cheatsheet-code.tex` | Landscape A4 | 7pt | 4 | Programming references, CLI commands | **Density techniques:** - Symbol substitution (for all -> "for all", there exists -> "there exists") - Horizontal formula stacking with `\quad` separators - Telegram-style text (articles and filler words stripped) - Content prioritization: formulas 60-70% -> procedures 15-20% -> constants 10-15% -> definitions 5-10% - Compile-and-verify loop to fit exact page budget --- ### 4. Mail Merge (`mail_merge.py` -- 574 lines) Generate N personalized documents from one template + one data source. HR departments have hired worse interns. ```bash python3 scripts/mail_merge.py template.tex contacts.csv \ --output-dir ./letters --workers 4 --merge --merge-name all_letters.pdf ``` **Two templating modes:** | Mode | Syntax | Features | |---|---|---| | **Simple** (default) | `{{first_name}}`, `{{company}}` | Auto-escapes `&`, `%`, `$`, `#`, `_`, `{`, `}`, `~`, `^`, `\` for LaTeX safety | | **Jinja2** (`--jinja2`) | `<< first_name >>`, `<% if ... %>` | Conditionals, loops, filters. `|escape_latex` filter available | **Technical details:** - Data sources: CSV, JSON, JSONL - Uses `concurrent.futures.ProcessPoolExecutor` for parallel compilation - Each document compiled independently (crash isolation) - Final merge via `qpdf --pages` or `pdfunite` - Custom naming: `--name-field last_name --prefix "offer_letter"` -> `offer_letter_Smith.pdf` --- ### 5. Version Diffing (`latex_diff.sh` -- 409 lines) A full-featured `latexdiff` wrapper with git integration. Track every comma your advisor changed at 2 AM. ```bash # Compare two files directly bash scripts/latex_diff.sh paper_v1.tex paper_v2.tex --compile --preview # Compare against a git commit bash scripts/latex_diff.sh paper.tex --git-rev HEAD~3 --compile # Compare between two git tags bash scripts/latex_diff.sh paper.tex --git-rev v1.0 --compile --type CULINECHBAR # Multi-file document with \input/\include bash scripts/latex_diff.sh old/main.tex new/main.tex --flatten --compile ``` **8 markup types:** | Type | Visual Effect | |---|---| | `UNDERLINE` (default) | Additions underlined in blue, deletions struck through in red | | `CTRADITIONAL` | Additions in blue text, deletions in red with strikethrough | | `CFONT` | Additions in sans-serif blue, deletions in tiny red | | `CHANGEBAR` | Change bars in margin only, no inline markup | | `CCHANGEBAR` | Change bars + color changes | | `CULINECHBAR` | Underline + change bars (most comprehensive) | | `FONTSTRIKE` | Font change + strikethrough | | `INVISIBLE` | No visible markup (for testing) | **Git integration:** Extracts old version via `git show
Behind every great PDF is a capybara who figured out the correct number of pdflatex passes.
--- ## Installation ```bash # Install as a Claude Code skill cp -r latex-document ~/.claude/skills/ # Or run full dependency setup bash setup.sh # Installs: TeX Live, Poppler, ImageMagick, Pandoc, Python deps bash setup.sh --check # Verify everything is installed ``` **System dependencies** (auto-installed by `setup.sh` on Debian/macOS/Fedora/Alpine/Arch): - **TeX Live** -- pdflatex, xelatex, lualatex, biber, makeindex, makeglossaries - **Poppler** -- `pdftoppm` for PNG previews, `pdfinfo` for page counts - **ImageMagick** -- `mogrify` for image resizing, `identify` for dimensions - **Pandoc** -- format conversion engine - **Python 3** + `matplotlib`, `numpy`, `pandas` - **qpdf** -- PDF encryption, merging, optimization, page extraction **Optional:** - **Node.js 18+** -- for Mermaid diagram conversion - **Java** -- for PlantUML rendering - **Graphviz** -- for `.dot` diagram rendering - **chktex** -- for LaTeX linting - **latexdiff** -- for version diffing --- ## Reference Documentation 25 reference guides in `references/`:
This project is proudly sponsored by HappyCapy
[HappyCapy](https://happycapy.ai) provided **free access to their platform**, full **Claude Code API credits**, and **Anthropic Claude Opus 4.6** -- the model that built every template, wrote every script, and generated every capybara image in this repository. Without HappyCapy, this skill would still be a half-baked idea on a sticky note. Instead, it's 27 templates, 27 scripts, 217 passing tests, and three capybara mascots with better fashion sense than most academics. **What is HappyCapy?** The agent-native computer for developers. No local setup, no security risks, no "works on my machine" excuses. Launch Claude Code in your browser and start building. It's the platform where capybaras and code live in harmony. --- ## License MIT --- ## Keywords > *For the search engines, the LLMs, and the curious humans who find things by typing random words into the void* > The README is long; the above is its first 40000 characters. Full content: https://github.com/ndpvt-web/latex-document-skill