cheatsheet

Output language

Read INTERFACE_LANG from .course-meta (default en). All user-facing prose — chat output, clarification questions, and narrative sections of any MD file you write (including the cheatsheet's prose) — must be in that language. Keep in English regardless: file paths, slash command names, pattern IDs (P1, P2…), YAML keys, LaTeX, code, and the literal section anchors downstream tools regex on.

Load skills/exam-drill/SKILL.md. Read course-index/patterns.md, course-index/coverage.md, course-index/summary.md, and errors/log.md.

Arguments: $ARGUMENTS

Procedure:

1. Collect highest-value items:

   • Top 5 patterns by frequency of appearance (from patterns.md)
   • All formulas boxed in derivations/*.md (final results)
   • User's most-frequent error types (from errors/log.md) — with the correction, not the error
   • ⚪ no-HW sections (and any ⚠weak-flagged §) with one key formula each — never drilled, so reference-grade only

2. Structure the cheatsheet (target: fits on 1 page @ 10pt):

   # <Course name> — Cheatsheet
   
   _Generated <date>. For exam reference only._
   
   ## Core formulas
   <table or compact list of boxed results from derivations/>
   
   ## Pattern quick-ref
   | Pk | Recognition | Move |
   |---|---|---|
   ...top 8 patterns only
   
   ## Traps to remember (from my errors/log)
   - <correction 1>
   - <correction 2>
   ...max 5
   
   ## No-HW formulas (memorize these — ⚪ sections no HW drilled)
   <one formula per ⚪ section, boxed>
   

3. Write to cheatsheet/final.md.

4. If --pdf in arguments:

   • Load skills/pdf/SKILL.md
   • Convert cheatsheet/final.md to cheatsheet/final.pdf using reportlab
   • Use 2-column layout, 9pt font, no margins (for maximum density)
   • Remember: NO Unicode subscripts/superscripts in reportlab — use <sub>/<super> XML tags
   • CJK fonts (required when INTERFACE_LANG=ko). reportlab's built-in Type-1 fonts (Helvetica/Times) cannot render Hangul — every Korean glyph silently comes out blank, so a ko cheatsheet PDF would be empty of prose. Register a CJK-capable TrueType font first and use it for every style (fontName=). reportlab's TTFont only parses TrueType (glyf) outlines — a CFF/OpenType-flavoured .ttc/.otf raises on registration. macOS AppleSDGothicNeo.ttc is CFF on recent macOS, so it reliably throws and the loop falls through to the next candidate; that's expected, not an error. Probe these in order (glyf-loadable Hangul fonts first) and keep the first that registers:

     from reportlab.pdfbase import pdfmetrics
     from reportlab.pdfbase.ttfonts import TTFont
     KFONT, candidates = "Helvetica", [
         "/System/Library/Fonts/Supplemental/AppleGothic.ttf",    # macOS (glyf — loads)
         "/usr/share/fonts/truetype/nanum/NanumGothic.ttf",       # Ubuntu (fonts-nanum)
         "/opt/homebrew/share/fonts/NanumGothic.ttf",             # macOS Homebrew font-nanum
         "/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc",# Noto CJK
         "/System/Library/Fonts/AppleSDGothicNeo.ttc",            # macOS (CFF — usually fails; last resort)
     ]
     for cand in candidates:
         try:
             pdfmetrics.registerFont(TTFont("KFont", cand)); KFONT = "KFont"; break
         except Exception:
             continue
     

     If none is found on a ko course, tell the user (in INTERFACE_LANG) to brew install font-nanum / sudo apt-get install fonts-nanum, or keep the .md and read it directly — don't ship a blank-glyph PDF.
   • Glyph coverage — keep \ell as ℓ, don't downgrade to l. In the LaTeX→text pass, map \ell to the real U+2113 ℓ (the angular-momentum symbol, all over central-force formulas) — not to an ASCII l. CJK fonts almost always carry ℓ (AppleGothic, Nanum and Noto all do), so the glyph renders; silently rewriting \ell→l makes ℓ²/(2μr²) read like a lowercase L and is the more common failure here than a missing glyph. Only fall back to italic l if the registered font truly lacks the glyph — test it (a present glyph maps to a non-zero id; 0 is .notdef):

     from reportlab.pdfbase.pdfmetrics import getFont
     def has_glyph(ch):
         if KFONT == "Helvetica": return False
         return getFont(KFONT).face.charToGlyph.get(ord(ch), 0) != 0
     ELL = "ℓ" if has_glyph("ℓ") else "<i>l</i>"   # use ELL wherever \ell / ℓ appears
     

   • Use pypandoc if available as alternative: pypandoc.convert_file('final.md', 'pdf', outputfile='final.pdf') (its LaTeX engine needs a CJK-aware mainfont, e.g. -V mainfont='Apple SD Gothic Neo', for Korean too)
   • Build with a scratch script outside the course folder. Don't write the reportlab build script into cheatsheet/ — it gets left behind in the user's committed course folder when the post-run cleanup is skipped. Put it in a temp path instead (e.g. SCRIPT=$(mktemp /tmp/paideia-cheatsheet-XXXX.py)), run it from there to emit cheatsheet/final.pdf, then delete it. If you do write one inside cheatsheet/ anyway, remove it after the PDF is verified — cheatsheet/ should end with only final.md and final.pdf.