| name | design |
| description | Create professional visual designs programmatically — cover pages, posters, certificates,
infographic layouts, title pages, invitations, and artistic compositions. Uses reportlab Canvas
and Pillow to draw shapes, typography, and patterns directly in code, with 80+ bundled fonts.
Output: PNG or PDF.
Always use this skill when the user wants a designed visual composition with typography,
layout, and artistic intent — "tạo poster", "thiết kế bìa", "làm certificate", "tạo thiệp",
"thiết kế cover page", "tạo bằng khen", "tạo banner", "design a poster", "make a cover",
"tạo infographic", "tạo hình bìa báo cáo" — even without saying "design".
Do NOT use for data charts (use gen-image) or AI-generated illustrations (use gen-image image mode).
|
| argument-hint | [type: poster|cover|certificate|infographic|invitation|banner|artistic] [style hints] |
| version | 1.1 |
| compatibility | {"requires":["Python >= 3.10","reportlab >= 4.1.0","Pillow (for PNG output)"],"assets":["canvas-fonts/ (80+ TTF files with OFL licenses)"],"tools":["run_in_terminal"]} |
| license | Apache-2.0 (see LICENSE.txt). Fonts under SIL OFL (see canvas-fonts/*.txt). |
Thiết Kế — Programmatic Visual Design
Governance: Read and follow .github/RULE.md — it overrides all instructions below.
References: references/design-principles.md
Bundled fonts: canvas-fonts/ (80+ TTF files with OFL licenses)
This skill creates visual compositions by writing Python code that draws directly onto a canvas
(reportlab Canvas for PDF, Pillow for PNG). The difference from gen-image is fundamental:
gen-image generates data charts and AI images, while design creates designed layouts
where typography, shapes, color, and whitespace work together as a deliberate composition. Think
of it as the difference between a bar chart and a poster — one visualizes data, the other
communicates through design.
All responses to the user are in Vietnamese.
Step 0: State Read-Back (RULE-13)
Call save_state.py read-context design as FIRST action before any processing. Check relevant_artifacts[] for upstream outputs to incorporate.
Bundled Template Scripts
For common design types, use the bundled scripts as a starting point. They handle font
registration, layout structure, and color palettes — you only need to customize content:
SCRIPTS:
gen_poster.py:
purpose: Event poster (A3/A4)
styles: modern | classic | bold
usage: python3 .github/skills/design/scripts/gen_poster.py --title "Event" --style modern --output poster.pdf
input: --title, --subtitle, --body, --footer, or --input data.json
gen_certificate.py:
purpose: Certificate / bằng khen (A4 landscape)
styles: formal | modern | elegant
usage: python3 .github/skills/design/scripts/gen_certificate.py --name "Nguyễn Văn A" --title "Chứng Nhận" --output cert.pdf
input: --name, --title, --course, --date, --issuer, or --input data.json
gen_cover.py:
purpose: Report cover page (A4)
styles: corporate | academic | creative
usage: python3 .github/skills/design/scripts/gen_cover.py --title "Annual Report" --author "Team" --output cover.pdf
input: --title, --subtitle, --author, --date, or --input data.json
For designs that don't fit these templates (infographics, invitations, custom artistic
compositions), write a custom script following Steps 1-4 below. The templates serve as
code references for font registration, color handling, and layout patterns.
When to Use This Skill (vs gen-image)
| User wants | Skill | Why |
|---|
| Bar/line/pie chart from data | gen-image | Data-driven, matplotlib |
| AI-generated illustration | gen-image | Needs diffusers/SD-Turbo |
| Report cover page | design | Typography + layout composition |
| Event poster | design | Artistic design with text hierarchy |
| Certificate / bằng khen | design | Formal layout with decorative elements |
| Infographic layout | design | Structured information design |
| Invitation / thiệp | design | Decorative composition |
| Abstract art piece | design | Creative shapes + typography |
Step 1: Design Philosophy
Before drawing, establish the visual direction. This is not a template — it is an aesthetic
framework that guides every decision (colors, fonts, spacing, shapes).
Purpose & Audience Analysis (do this first)
Design serves a purpose — and the purpose shapes every visual decision. Before choosing
aesthetics, analyze what the design needs to accomplish:
1. Communication goal:
- What is the single most important message? (This becomes the visual focal point)
- What emotion should it evoke? (Authority, excitement, warmth, urgency, celebration)
- What action should the viewer take? (Attend event, feel recognized, read report)
2. Audience:
- Corporate executives → clean, restrained, serif fonts, muted palette
- Young/creative audience → bold colors, modern sans-serif, dynamic composition
- Academic/formal → traditional layout, serif typography, structured grid
- General public → high contrast, large text, immediately clear message
3. Context of use:
- Will it be printed? → CMYK-safe colors, high DPI, check readability at print size
- Digital display? → RGB is fine, consider screen sizes
- Large format (poster)? → Test readability at 3m distance — title should be readable
- Small format (card/badge)? → Simplify, fewer elements, larger relative text
4. Content hierarchy:
Rank every piece of text by importance (1 = most important):
- Rank 1: Main title or name → largest, most prominent
- Rank 2: Key supporting info (date, subtitle) → medium, clear
- Rank 3: Details (organization, location, fine print) → smallest, but still readable
- Don't give everything equal visual weight — that creates visual noise, not design
Report the analysis:
🎨 Phân tích thiết kế:
- Mục tiêu: {communication_goal}
- Cảm xúc: {target_emotion}
- Đối tượng: {audience_type}
- Sử dụng: {usage_context}
- Focal point: {main_message}
Aesthetic Direction
Name the aesthetic (1–2 words): e.g., "Concrete Poetry", "Chromatic Silence", "Metabolist Dreams"
Articulate the philosophy (3–4 paragraphs) covering:
- Space and form — how elements occupy the canvas
- Color and material — palette, texture, visual weight
- Typography — font choices, scale, role of text (accent vs content)
- Composition — rhythm, balance, visual hierarchy
The philosophy should feel like a manifesto for an art movement. Keep it generic enough to
adapt to the specific piece, but specific enough to actually constrain choices. Save it as
tmp/design_philosophy.md for reference.
Philosophy Examples (condensed)
"Concrete Poetry" — Monumental form, bold geometry. Massive color blocks, sculptural
typography, Brutalist spatial divisions. Text as rare, powerful gesture — only essential
words integrated into the visual architecture.
"Chromatic Language" — Color as primary information system. Geometric precision where
color zones create meaning. Minimal typography — small sans-serif labels letting chromatic
fields communicate. Josef Albers meets data visualization.
"Geometric Silence" — Pure order and restraint. Grid-based precision, dramatic negative
space. Swiss formalism meets Brutalist material honesty. Structure communicates, not words.
Step 2: Choose Tools and Canvas
PDF output (vector, crisp at any zoom):
from reportlab.lib.pagesizes import A4, landscape
from reportlab.pdfgen import canvas
from reportlab.lib.units import mm, cm
from reportlab.pdfbase import pdfmetrics
from reportlab.pdfbase.ttfonts import TTFont
PNG output (raster, web-friendly):
from PIL import Image, ImageDraw, ImageFont
Canvas dimensions:
- A4 portrait: 210×297mm (PDF) or 2480×3508px at 300dpi (PNG)
- A4 landscape: 297×210mm or 3508×2480px
- Poster: 1920×2560px (PNG) or custom
- 16:9 slide cover: 1920×1080px (PNG)
- Square social: 1080×1080px (PNG)
Register fonts from canvas-fonts/ directory:
font_dir = ".github/skills/design/canvas-fonts"
pdfmetrics.registerFont(TTFont('WorkSans', f'{font_dir}/WorkSans-Regular.ttf'))
pdfmetrics.registerFont(TTFont('WorkSans-Bold', f'{font_dir}/WorkSans-Bold.ttf'))
Step 3: Create the Composition
Write a Python script that draws the design. The script should be self-contained and saved
to tmp/gen_design.py. Run it via:
python3 tmp/gen_design.py --output output/design.pdf --title "Your Title"
Composition Principles
Typography is art, not just text. Use different fonts at dramatically different scales.
A single huge word can anchor the composition while tiny labels provide context. Browse the
80+ fonts in canvas-fonts/ — each has a distinct personality:
- Serif display: YoungSerif, Gloock, LibreBaskerville, CrimsonPro
- Sans modern: WorkSans, Outfit, InstrumentSans, BricolageGrotesque
- Mono/tech: GeistMono, JetBrainsMono, IBMPlexMono, DMMono, RedHatMono
- Display/art: Boldonse, EricaOne, PoiretOne, NationalPark, Silkscreen, PixelifySans
- Elegant: Italiana, Lora, InstrumentSerif, IBMPlexSerif, Jura
- Handwritten: NothingYouCouldDo, ArsenalSC
Shapes and patterns build rhythm. Repeating geometric elements (circles, lines, grids)
create visual texture. Use them intentionally, not decoratively.
Color palette is limited and intentional. Pick 3–5 colors max. Let one dominate, one
accent, rest neutral. Use hex colors in reportlab: canvas.setFillColor(HexColor('#2D3436')).
Whitespace is a design element. Empty space creates breathing room and draws attention
to what is present. Do not fill every corner.
Nothing overlaps unintentionally. Check that all text and shapes have clear separation.
Every element should be within canvas boundaries with proper margins.
Step 4: Refine
After the first render, take a critical pass:
- Does the composition feel cohesive or cluttered?
- Is the typography hierarchy clear (what reads first, second, third)?
- Are colors harmonious and intentional?
- Is there enough whitespace?
- Does it look like a professional designer made it?
Refine the code — adjust spacing, font sizes, colors, element positions. Do not add more
elements; instead make existing ones work better together.
Step 5: Verify & Report
- Confirm output file exists and size is reasonable (PDF > 5KB, PNG > 50KB)
- Report:
✅ Thiết kế hoàn tất: {path} ({size})
- If called from synthesize pipeline, return
{path, type, width, height} to the caller
Multi-Page Option
When the user requests multiple pages (e.g., a full booklet or card set), create each page
as a variation on the same design philosophy — same palette and typography family but distinct
compositions. Bundle them in a single PDF. Each page should feel like part of a cohesive series.
Examples
Example 1 — Report cover page:
Input: "Thiết kế bìa báo cáo 'Annual Report 2026' cho Team Analytics"
Output: gen_cover.py --title "Annual Report 2026" --author "Team Analytics" --style corporate → cover.pdf (A4, 25 KB)
Example 2 — Event poster:
Input: "Tạo poster cho workshop AI, ngày 15/5, phong cách modern"
Output: gen_poster.py --title "AI Workshop" --subtitle "May 15, 2026" --style modern → poster.pdf (A3, 35 KB)
Example 3 — Certificate:
Input: "Làm bằng khen cho Nguyễn Văn A hoàn thành khóa Data Science"
Output: gen_certificate.py --name "Nguyễn Văn A" --course "Data Science" --style formal → cert.pdf (A4 landscape, 20 KB)
Example 4 — Custom infographic (no template):
Input: "Tạo infographic layout cho 5 bước quy trình tuyển dụng"
Output: Custom script tmp/gen_design.py → infographic.pdf (A4, 45 KB), follows Steps 1-4
Step 6: Artifact Registration (RULE-13)
Call save_state.py register-artifact --step design --path <file> --type draft_output --summary "<text>" for every file created in tmp/ or output/.
What This Skill Does NOT Do
- Does NOT create data-driven charts — use gen-image (matplotlib)
- Does NOT generate AI images from prompts — use gen-image (diffusers/torch)
- Does NOT create interactive or animated content — use gen-html
- Does NOT install dependencies — redirects to setup
- Does NOT copy existing artists' work — all designs are original compositions
