create-pdf

Guidelines for creating professional, visually distinctive PDF reports via python-pptx's PDFReport helper class and matplotlib.

The design language borrows from modern presentation aesthetics — clean, spacious, intentional — adapted for the long-form reading experience. Light body pages for readability, bold title treatment for impact, and strategic color throughout.

Reference Files

Read these from /tmp/skills/pdf/references/ when you need implementation details.

File	When to load	Contents
`components.md`	Before building the report	Helper method reference, code block implementation, custom component patterns, visual motif ideas
`charts.md`	Before creating any chart	Matplotlib styling, palette integration, chart type selection, chart + context composition
`technical.md`	When troubleshooting or building custom layouts	Page-break safety, write/multi_cell pitfalls, FPDF encoding, common bugs

1. Design Philosophy

Every element earns its place on the page. If it doesn't communicate, clarify, or guide the eye — remove it.

1.1 Core Principles

Content is king. Typography, spacing, and strategic color create hierarchy. Decoration is the enemy. A well-structured report with generous whitespace communicates more authority than one stuffed with borders and ornaments.

Light body, bold accents. Reports are read at length — light backgrounds with dark text for body content. Save bold, dark treatment for the title page and section openers. The title page sets the mood; body pages prioritize readability.

60/30/10 color distribution. 60% neutral (white page background, dark body text), 30% complementary (card/callout fills, section header colors, table header bands), 10% accent (KPI values, chart highlights, key stats, accent bars on callout boxes). If accent color covers more than ~10% of a page's visual area, pull back.

Topic-driven palette. The palette should feel designed for THIS report's subject. If swapping colors into a completely different report would still "work," the choices aren't specific enough. One color dominates (titles, section headers), 1–2 supporting tones for callout fills and tables, one accent for highlights.

Generous whitespace. Let content breathe. When a section feels cramped, split it across two pages rather than shrinking fonts. Space between sections signals confidence, not waste.

Two font weights only. Regular for body, Bold for headings and emphasis. Helvetica covers both. Reserve Italic for quotes and captions only — it's a third voice, use it sparingly.

Flat and clean. No heavy borders, drop shadows, 3D chart effects, or decorative clip art. Subtle background fills, minimal or no borders, and clean lines. Hierarchy comes from size, weight, color, and spatial position — in that order.

Visual motif. Commit to ONE distinctive element and carry it through every page: a colored left-margin bar on callout boxes, accent lines under headings, numbered circle badges for key items, or a consistent card styling. One motif, applied everywhere, creates more design coherence than multiple decorative ideas applied inconsistently.

Layout variety. No two consecutive sections should use the same visual format. Rotate through prose, callout boxes, charts, tables, KPI strips, quotes, and numbered/labeled items. Plan the visual treatment for each section before writing code.

1.2 What Looks Outdated

Outdated	Modern replacement
Heavy borders around every element	Subtle background fills, minimal or no borders
Decorative clip art or icons	Purposeful whitespace
Underlined headings	Size + color contrast + spacing below
Multiple font families (3+)	Max 2: Helvetica for everything, or pair with one serif for titles
Centered body text	Left-aligned body; center only title page and captions
Dense walls of text with no breaks	Whitespace between paragraphs, callout boxes, charts
Rainbow color schemes	Topic-driven palette: 1 dominant + 1–2 supporting
3D chart effects, gradient fills on charts	Flat, clean chart style with palette-harmonized colors
Same layout pattern every section	Rotate through the helper methods
Drop shadows on boxes	Flat fill or very subtle border only

1.3 Accessibility & Contrast

WCAG minimums: Body text (<18pt): 4.5:1 contrast ratio. Headings (≥18pt or ≥14pt bold): 3:1.

Safe on white/light: #000000, #333333, #1B2A4A, #4A4A4A. On dark backgrounds (title page): #FFFFFF, #E0E0E0.

Avoid: light gray text on white, pastel text on light backgrounds, red (#FF0000) on white (only 4:1 — fails for small text). Never use color as the sole information channel.

2. Theme & Color System

2.1 Built-In Themes

Each theme defines title/heading color, accent color, card/callout background, and body text. Apply via pdf.set_theme('name').

Theme	Title	Accent	Card BG	Best for
midnight	#1E2761	#CADCFC	#EDF1FA	Executive, strategy
forest	#2C5F2D	#97BC62	#F0F5E8	Sustainability, nature
coral	#F96167	#2F3C7E	#FFF5F0	Creative, marketing
terracotta	#B85042	#A7BEAE	#F5F0ED	Architecture, culture
ocean	#065A82	#1C7293	#E8F2F7	Technology, research
charcoal	#36454F	#212121	#F2F2F2	Minimal, professional
teal	#028090	#02C39A	#E6F5F2	Healthcare, fintech
berry	#6D2E46	#A26769	#F5EDE8	Fashion, premium
sage	#84B59F	#50808E	#EDF3F0	Wellness, consulting
cherry	#990011	#2F3C7E	#FFF5F5	Impact, bold analysis

2.2 Custom Palettes

For subjects that don't fit a built-in theme, set colors directly:

pdf.COLORS['title'] = (30, 39, 97)
pdf.COLORS['accent'] = (100, 140, 200)
pdf.COLORS['card_bg'] = (240, 244, 248)
pdf.COLORS['body'] = (51, 51, 51)
pdf.COLORS['border'] = (200, 210, 220)

Derive card_bg from the title color at ~5–10% opacity over white. The accent should complement but not match the title color — a different hue or a lighter/brighter version of the same family.

2.3 Chart Palette Integration

Always harmonize charts with the report palette:

pal = pdf.get_chart_palette()
# pal['primary']    — hex, main chart color
# pal['secondary']  — hex, accent chart color
# pal['series']     — list of 6 harmonized hex colors
# pal['title_hex']  — for chart titles
# pal['body_hex']   — for axis labels
# pal['card_bg_hex'] — for chart background fills

Pass pal['series'] to bar colors, line colors, pie slices, etc. Charts that use default matplotlib colors instead of the report palette look disconnected.

3. Script Toolkit

3.1 Loading

import sys, importlib
sys.path.insert(0, '/tmp/skills/pdf/scripts')
from pdf_helpers import PDFReport, THEMES, render_formula

3.2 Initialization

pdf = PDFReport(orientation='P', unit='mm', format='A4')
pdf.set_auto_page_break(auto=True, margin=25)
pdf.set_margins(left=20, top=20, right=20)
pdf.set_theme('ocean')
pdf.alias_nb_pages()

3.3 Helper Method Overview

Method	Purpose
`set_theme(name)`	Apply a named palette
`get_chart_palette(n=6)`	Theme-harmonized hex colors for matplotlib
`ensure_space(min_mm)`	Page break if not enough room
`add_title_page(title, subtitle, author, date, org)`	Dark header-band title page
`add_section(title)`	Section heading with accent underline
`add_subsection(title)`	Smaller sub-heading
`add_body(text)`	Standard body paragraph
`add_callout_box(text, fill_color, min_height)`	Tinted box with left accent bar
`add_quote_strip(text, attribution)`	Centered italic band
`add_kpi_strip(kpis, cols)`	Row of stat cards
`add_indented_rich_text(indent_x, bold_text, regular_text, ...)`	Bold + regular inline at indent
`add_labeled_item(indent_x, indicator_color, label, description)`	Color bar + label/description
`add_numbered_item(indent_x, number, badge_color, label, description)`	Badge + label/description
`add_chart(img_path, caption, width)`	Centered image + sequential caption
`add_table(headers, rows, col_widths, row_height)`	Simple table (clips long text)
`add_table_wrapped(headers, rows, col_widths, line_height, padding)`	Table with multi-line wrapping
`add_formula(img_path, max_w)`	Formula image in tinted strip

For detailed signatures, usage notes, and the add_code_block() standalone function, see references/components.md.

3.4 Font Size Guide

Element	Size	Weight
Report title (title page)	26–32pt	Bold
Section headings	16–20pt	Bold
Sub-headings	13–15pt	Bold
Body text	10–12pt (11pt recommended)	Regular
Captions	8–10pt	Italic
Table text	9–11pt	Regular
KPI values	24–30pt	Bold, in accent color
KPI labels	9–10pt	Regular, muted

Use Helvetica (built into fpdf2). One font family is enough — hierarchy comes from size and weight.

4. Report Planning & Structure

4.1 Before Writing Code

Clarify purpose (informational, analytical, persuasive, status update) and audience (executives want brevity and conclusions first, technical readers want methodology and detail). Outline the sections, decide which visual treatment each gets (prose, callout, chart, table, KPI strip, badges), and plan a target page count.

4.2 Recommended Section Order

Title Page — add_title_page() with dark header band, subtitle, author, date
Executive Summary — Purpose, key findings, conclusions, recommendations. Write LAST, place FIRST.
Introduction / Background — Context, scope, objectives
Methodology (if applicable) — Data gathering approach, kept brief
Findings / Analysis — Core content with charts, organized by theme/chronology/priority
Discussion (if applicable) — Interpret findings, address limitations
Conclusions — Main takeaways
Recommendations (if applicable) — Specific, actionable next steps
Appendix / References (if applicable)

Lead with conclusions (inverted pyramid). One idea per section. Executive summary should stand alone at ~10–15% of total text content.

4.3 Special Report Types

Executive / Status: 3–5 sections, lead with KPI strip, outcomes and next steps. Keep it short.

Analytical / Research: Detailed, includes methodology and discussion, line charts and tables, addresses limitations.

Comparison / Benchmark: Structured around dimensions, uses grouped bar/radar charts, ends with clear recommendation.

Incident / Post-Mortem: Chronological, factual, blameless tone.

5. Writing Style

Clear, direct language. Sentences under 25 words, paragraphs under 75 words. Active voice preferred.

Be specific: "Revenue increased 18% YoY to $4.2M" communicates more than "Revenue increased significantly."

Professional, objective tone — confident but measured. Consistent formatting for numbers, currencies, and percentages. Always provide context for numbers: comparisons, benchmarks, or trends.

6. Chart Sections — Tell the Story, Not Just the Data

<chart_section_composition> A chart without interpretation is a missed opportunity. Every chart section in a report pairs the visualization with a clear takeaway — the chart is evidence for an insight, and the insight should be stated explicitly.

Every chart section includes:

An insight-driven heading — frames what the data reveals. "Revenue Grew 18% as Enterprise Segment Doubled" communicates far more than "Revenue Overview."
Brief context — 1–2 sentences before the chart that set up what the reader should look for.
The chart itself — styled to match the report palette, saved at dpi=150+, with descriptive axis labels.
A "Figure N:" caption — sequentially numbered, placed via add_chart(img_path, caption).
A takeaway paragraph after the chart — what the data means, why it matters, what action it implies. This is where the analysis lives.

The pattern is: heading → context sentence → chart → caption → analysis paragraph. A section with just a heading and a chart leaves the reader to draw their own conclusions, which defeats the purpose of the report. </chart_section_composition>

For chart type selection, matplotlib styling, and palette integration details, see references/charts.md.

7. Layout & Pagination

7.1 Content Density

One key idea per section. Prefer a single clear chart over three small ones. Generous whitespace between sections. When a section feels cramped, split across two pages.

7.2 Page-Break Safety

Composite elements (callout boxes, KPI strips, charts, tables) that draw background fills at absolute Y positions produce rendering artifacts when multi_cell() triggers an auto page break mid-element. All built-in helpers handle this via ensure_space(). When building custom composite elements, call pdf.ensure_space(estimated_height) before drawing.

For detailed technical rules on page breaks, the write/multi_cell overflow problem, code blocks, and FPDF encoding limitations, see references/technical.md.

7.3 Layout Variety

Plan the visual treatment for each section before writing code. Use this as a rotation guide:

Section type	Good visual treatment
Key findings / executive summary	KPI strip + callout boxes
Narrative context	Body prose, maybe a callout for the main point
Data analysis	Chart + takeaway paragraph
Comparisons	Table or side-by-side chart
Steps / process	Numbered items with badges
Pros/cons / categories	Labeled items with colored indicators
Key quotes / takeaways	Quote strip
Code / technical detail	Code block

No two consecutive sections should use the same format. If the last section was a chart, make the next one a callout box or table.

8. QA Checklist

Theme chosen — topic-driven palette applied via set_theme() or custom COLORS
Chart palette used — get_chart_palette() colors passed to all matplotlib charts
Visual motif consistent — same accent treatment on every page
Layout variety — no two consecutive sections use the same visual format
Title page present with dark header band, title, author, date
Executive summary stands alone
Logical section flow with coherent narrative arc
All data/numbers verified for accuracy
Each chart has dpi=150+, palette-harmonized colors, and sequential "Figure N:" caption
Every chart section includes context before and takeaway after the chart
Consistent fonts, colors, and spacing throughout
Tables have themed headers and alternating row shading
Page numbers via alias_nb_pages()
set_auto_page_break enabled
matplotlib.use('Agg') set before pyplot import
plt.close() called after each chart save
No write() + multi_cell(fixed_width) patterns (use add_indented_rich_text)
No manual rect + multi_cell combos without ensure_space() (use built-in helpers)
Code/preformatted text uses add_code_block() — never add_body() for multiline code
WCAG contrast met — body text ≥4.5:1, headings ≥3:1
No outdated patterns — no heavy borders, 3D charts, clip art, rainbow colors
File saved to /tmp/ and generate_download_link called