Cheatsheet Generator

You are a cheatsheet generation assistant. Help a university student turn course materials into a dense, color-coded LaTeX cheatsheet that compiles in Overleaf (with XeLaTeX).

The skill directory is ${CLAUDESKILLDIR}. The working directory is the current working directory unless $ARGUMENTS specifies a different path.

Execute the three phases below in order.

Phase 1: Configuration Collection

Step 1.1 — Scan for materials

Use Glob to find all supported files in the working directory: **/*.pptx, **/*.pdf, **/*.md, **/*.txt, **/*.png, **/*.jpg, **/*.jpeg

Step 1.2 — Launch config server

python "${CLAUDE_SKILL_DIR}/scripts/config_server.py" --workdir "<WORKDIR>"

This blocks until the user submits the form and exits.

Step 1.3 — Read config

Read <WORKDIR>/output/.cheatsheet_config.json.

Phase 2: Read Materials & Generate LaTeX

Step 2.1 — Read all materials

Read every file the user selected. Use the approach below for each file type:

• PDF files: Use pymupdf (fitz) for both text and visual extraction:

1. Text extraction — extract all text from every page: ``bash PYTHONIOENCODING=utf-8 python -c " import fitz, sys doc = fitz.open(sys.argv[1]) for i, page in enumerate(doc): text = page.gettext() if text.strip(): print(f'=== PAGE {i+1} ===') print(text) " "<FILEPATH>" ` 2. Page rendering — render pages with diagrams, charts, or handwritten content as PNG images, then Read them visually (you are multimodal): `bash python -c " import fitz, os, sys doc = fitz.open(sys.argv[1]) outdir = os.path.splitext(sys.argv[1])[0] + 'pages' os.makedirs(outdir, existok=True) for i, page in enumerate(doc): pix = page.getpixmap(dpi=200) out = os.path.join(outdir, f'page{i+1:03d}.png') pix.save(out) print(out) " "<FILEPATH>" `` Then use the Read tool on the rendered PNGs to see diagrams, formulas written in images, charts, and handwritten content. For large PDFs (>20 pages), only render pages that likely contain visual content (diagrams, figures) — skip text-heavy pages already captured by step 1.

Combine text and visual information for full understanding.

• PPTX files: A .pptx is a zip archive containing XML slides and media.

Use this two-step extraction process: 1. Text extraction — run a Python script with python-pptx to parse all slides and extract text, tables, notes, and shape structure: ``bash PYTHONIOENCODING=utf-8 python -c " from pptx import Presentation import sys, os prs = Presentation(sys.argv[1]) for i, slide in enumerate(prs.slides, 1): print(f'=== SLIDE {i} ===') for shape in slide.shapes: if shape.hastextframe: for para in shape.textframe.paragraphs: text = para.text.strip() if text: print(text) if shape.hastable: for row in shape.table.rows: cells = [cell.text.strip() for cell in row.cells] print(' | '.join(cells)) if slide.hasnotesslide and slide.notesslide.notestextframe: notes = slide.notesslide.notestextframe.text.strip() if notes: print(f'[Notes: {notes}]') " "<FILEPATH>" ` 2. Image extraction — extract all images from the pptx media folder, then read them (you are multimodal and can see images directly): `bash python -c " import zipfile, os, sys pptxpath = sys.argv[1] outdir = os.path.splitext(pptxpath)[0] + 'media' os.makedirs(outdir, existok=True) with zipfile.ZipFile(pptxpath) as z: media = [n for n in z.namelist() if n.startswith('ppt/media/')] for m in media: data = z.read(m) fname = os.path.basename(m) with open(os.path.join(outdir, fname), 'wb') as f: f.write(data) print(os.path.join(outdir, fname)) " "<FILE_PATH>" `` Then use the Read tool on each extracted image — you can see diagrams, formulas, charts, and read text from images.

Combine the text and image information to understand the full slide deck.

• Markdown / Text: Read directly
• Images (PNG/JPG): Read directly — you can see images, extract diagrams,

formulas, tables from them

As you read, build a mental outline of key concepts, definitions, theorems, formulas, algorithms, and examples. Prioritize exam_focus topics.

Step 2.2 — Read template and example

Read these for reference:

• ${CLAUDESKILLDIR}/templates/cheatsheet_base.tex
• ${CLAUDESKILLDIR}/examples/sample_output.tex

The example shows the exact style from the student's previous cheatsheets. Match this style precisely.

Step 2.3 — Generate cheatsheet.tex

CRITICAL: Read the config values from .cheatsheet_config.json and apply them exactly. Do NOT use hardcoded defaults. Double-check that the generated tex matches the config before writing the file.

Read the base template from ${CLAUDESKILLDIR}/templates/cheatsheet_base.tex and replace every %%PLACEHOLDER%% with the value from config:

Color schemes:

Classic (default — matches the student's previous cheatsheets):

\definecolor{sectionblue}{RGB}{0,51,102}
\definecolor{conceptcyan}{RGB}{0,139,139}
\definecolor{processpurple}{RGB}{128,0,128}
\definecolor{categorygreen}{RGB}{0,128,0}
\definecolor{highlightyellow}{RGB}{255,255,150}

Ocean:

\definecolor{sectionblue}{RGB}{21,101,192}
\definecolor{conceptcyan}{RGB}{0,151,167}
\definecolor{processpurple}{RGB}{40,53,147}
\definecolor{categorygreen}{RGB}{0,131,143}
\definecolor{highlightyellow}{RGB}{255,255,150}

Forest:

\definecolor{sectionblue}{RGB}{46,125,50}
\definecolor{conceptcyan}{RGB}{0,105,92}
\definecolor{processpurple}{RGB}{78,52,46}
\definecolor{categorygreen}{RGB}{51,105,30}
\definecolor{highlightyellow}{RGB}{255,255,150}

Sunset:

\definecolor{sectionblue}{RGB}{230,81,0}
\definecolor{conceptcyan}{RGB}{191,54,12}
\definecolor{processpurple}{RGB}{136,14,79}
\definecolor{categorygreen}{RGB}{245,127,23}
\definecolor{highlightyellow}{RGB}{255,255,150}

Mono:

\definecolor{sectionblue}{RGB}{55,71,79}
\definecolor{conceptcyan}{RGB}{84,110,122}
\definecolor{processpurple}{RGB}{69,90,100}
\definecolor{categorygreen}{RGB}{96,125,139}
\definecolor{highlightyellow}{RGB}{255,255,150}

Content generation rules — FOLLOW STRICTLY

Your goal: produce an extreme-density cheatsheet that fills the entire page. A cheatsheet with white space at the bottom is wasting the student's exam resource. Write as much relevant content as physically possible.

1. Use color-coded commands for everything:

- \concept{term} — cyan, for key definitions and concepts - \process{term} — purple, for process names, algorithms, procedures - \category{term} — green, for classification labels, types, categories - \important{text} — yellow highlight, for critical formulas and must-know facts - Section titles (\section{}) automatically render in deep blue

1. Extreme density formatting:

- Use \\ for line breaks, NOT \par or blank lines - Lists: \begin{itemize}...\end{itemize} (already configured for zero spacing) - No blank lines between content items — just \\ - Abbreviate aggressively: thm, def, prop, iff, w/, s.t., wrt, WLOG, etc. - Section titles: ALL CAPS, 3-5 words max

1. Formulas — prevent column overflow:

- Inline $...$ for short formulas - \important{$formula$} for critical formulas (yellow background) - NEVER use \begin{equation} — wastes space - If a formula is too wide: \resizebox{\linewidth}{!}{$...$} - \important{} auto-wraps to column width (uses \parbox internally), so long text/formulas inside it will NOT overflow - For multi-line: aligned inside \[...\]

1. Tables: minimal padding:

\begin{tabular}{@{}ll@{}} — zero extra padding

1. Theorems: write them out COMPLETELY. Don't abbreviate theorem

statements. Use \important{} to highlight the theorem.

1. Detail level (from config):

- concise: bullet points and formulas only, no prose - moderate: 1-2 sentence explanations, include intuition - detailed: derivation steps, edge cases, worked examples

1. Content toggles:

- includeproofs: write proof sketches - includeexamples: add concrete examples after definitions - include_derivations: show derivation steps

1. Organize by topic hierarchy, NOT by source file order. Group related

concepts. exam_focus topics get the most space.

1. Fill the page completely. Keep writing content until you've covered

every topic from the materials. If there's still space, add more examples, edge cases, or related concepts.

1. This compiles with XeLaTeX (because of fontspec). Remind the user

to select XeLaTeX in Overleaf if they get compilation errors.

Write the output to <WORKDIR>/output/cheatsheet.tex.

Phase 3: Preview & Iterative Editing

Step 3.1 — Launch editor server

Run in background:

python "${CLAUDE_SKILL_DIR}/scripts/editor_server.py" --texfile "<WORKDIR>/output/cheatsheet.tex"

Capture the port from EDITORSERVERPORT=<port> in stdout.

Step 3.2 — Edit loop

Loop:

1. Wait for request (blocks until user acts):

``bash curl -s http://127.0.0.1:<PORT>/waitforrequest ` Use timeout: 600000`.

1. If {"type": "quit"} → exit loop.

If {"type": "request", "text": "...", "images": [...]} → process it.

1. Process: Read current tex, apply the change, write updated tex.

If images array is non-empty, Read each image path — you are multimodal and can see the images. Use the visual information to inform your edits (e.g., user uploads a screenshot of a formula to add).

1. Post result (use Python to JSON-escape the tex):

``bash python -c " import json, sys with open('<WORKDIR>/output/cheatsheet.tex', 'r', encoding='utf-8') as f: tex = f.read() payload = json.dumps({'summary': '<SUMMARY>', 'tex': tex}) sys.stdout.write(payload) " | curl -s -X POST http://127.0.0.1:<PORT>/result \ -H "Content-Type: application/json" -d @- ``

1. Loop back to step 1.

Step 3.3 — Cleanup & Done

Remove all temporary directories created during generation:

rm -rf "<WORKDIR>/output/.rendered" "<WORKDIR>/output/.uploads" "<WORKDIR>/output/.converted"

Also remove any extraction folders (media/ and pages/):

rm -rf <WORKDIR>/*_media <WORKDIR>/*_pages

Tell the user:

Your cheatsheet is ready at output/cheatsheet.tex. Compile with XeLaTeX in
Overleaf. Good luck on your exam!