convez.ai
SkillDesign & UIIntermediário
KKenny0

Card Skill

KKenny0/card-skill

Texto em PNG: infográficos, pôsteres e cards visuais em 7 modos

Ver no GitHub
InicialPluginScore · de 0 a 100

Uma skill Claude Code que transforma texto em imagens PNG de alta qualidade (4K) com design profissional. Oferece 7 modos (infográfico, pôster, comics, quadro branco, etc.) e 18+ sistemas de design de marca, ideal para criar conteúdo visual polido sem deixar a interface do Claude.

Instalação

$npx convez-ai KKenny0/card-skill

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
5
Forks
1
Acompanhando
0
Saúde
28%
Último commit
há 4 dias
Versão
v0.22.0
Issues abertas
0
Contribuidores
1

Casos de uso

  • Converter artigos e blog posts em infográficos e cards de leitura
  • Gerar pôsteres, manifestos e headlines com máximo contraste
  • Criar quadros brancos para visualizar argumentos e diagramas de sistemas
  • Produzir conteúdo visual para redes sociais com design de marca consistente
  • Transformar histórias e narrativas em formatos visuais (comics, sketchnotes)

Veredito

Skill sólida que resolve um problema real: converter conteúdo textual em assets visuais de qualidade impressa sem sair do Claude. A análise automática de densidade e tom + 18+ sistemas de design é diferencial. Suporte nativo a chinês e inglês, otimizado para fluxo prático.

README

A tradução para o português deste README ainda está na fila.

card-skill

Content in, PNG out. Mold decides the shape.

A Claude Code skill that renders text into designed images — infographics, posters, comics, sketchnotes, whiteboards, big-text posters, and long-form reading cards — through 30+ brand design systems and 8 content-tone palettes.

Usage

Claude Code

npx skills add KKenny0/card-skill -a claude-code -g -y

Generic agents (Codex, OpenCode, Pi, and other tools that read from ~/.agents/)

npx skills add KKenny0/card-skill -a '*' -g -y

The skill auto-triggers from natural requests, no slash command needed. Optimized for Chinese and English.

Example prompts:

  • English: make this into an infographic / render this as a poster / turn this article into a visual card / make a whiteboard for this argument
  • 中文: 把这段内容做成信息图 / 做成海报 / 渲染成卡片 / 做个白板推演

What it does

Give it an article, a quote, a tweet thread, or a URL. Card analyzes the content's structure, density, and emotional tone, then renders it as a high-resolution PNG (4K width, DPR 2) that looks like printed matter — not a web screenshot.

Seven modes

ModeTemplateBest for
infographStructured data, comparisons, layered ideasDense analytical content
bigSingle statement, maximal contrastHeadlines, manifestos, one-liners
longArticle-length readingEssays, blog posts, long-form text
posterMulti-card seriesChapter-by-chapter, topic-per-card grids
comicNarrative with conflictStories, debates, before/after arcs
sketchnoteWarm narrativePersonal reflections, lessons learned
whiteboardLogical reasoningArguments, system diagrams, decision chains

Design systems

18 brand design systems curated from real DESIGN.md files, plus 8 ljg-card content-tone palettes tuned for Chinese text:

Dark Minimal — linear, vercel Dark Cinematic — spotify Light Minimal — apple, expo, notion Light Editorial — claude, cursor, intercom, replicate, posthog, clay Technical Data — stripe, ibm, opencode.ai, sentry, raycast, together.ai ljg-card tones — 沉思, 锐利, 温暖, 技术, 科研, 创意, 商业, 默认

Each system contributes color palette, shadow philosophy, border-radius style, and whitespace rhythm. Typography is mode-locked — brand systems don't override fonts.

Parameters

ParameterDescriptionDefault
--designSkip candidate preview, go straight to a named design system(enters preview flow)
--dprDevice pixel ratio2
--authorByline textKenny Wu
--photoByline avatar URL or pathassets/avatar.png

Pipeline

  1. Ingest — URL (via WebFetch), pasted text, or file path
  2. Analyze — Extract title, subtitle, density, structure, emotional tone
  3. Match — Pick 3-5 candidate design systems based on mood × theme × density
  4. Confirm — Present candidates in terminal; user picks one or asks for a new batch
  5. Render — Fill the mode's HTML template with mapped design tokens + content
  6. Self-check — Verify element ratios, font sizes, accent limits, no AI tells
  7. Capture — Screenshot via Playwright at 4K width

Output lands in ~/Downloads/.

Project structure

card-skill/
├── SKILL.md                    # Skill definition and execution flow
├── assets/
│   ├── {mode}_template.html    # HTML templates (7 modes)
│   ├── capture.js              # Playwright screenshot script
│   ├── fonts/                  # Xiangcui typeface family
│   ├── avatar.png              # Default byline avatar
│   └── logo.png                # Colophon logo
├── references/
│   ├── taste.md                # Anti-AI aesthetic rules, paper/print baseline
│   ├── design-index.md         # All design systems + CSS variables
│   ├── designs/{name}.md       # Per-brand compact design files
│   └── mode-{name}.md          # Per-mode content theory and layout rules
└── evals/                      # Assertion-based evaluation harness

Dependencies

  • Playwright — headless Chromium for PNG capture (npx playwright install chromium)
  • Fonts — place in assets/fonts/:
    • XiangcuiDengcusong (required) — CJK serif typeface used by all 7 modes
    • XiangcuiDazijiti (optional) — CJK typewriter face used by infograph mode
    • Playwright falls back to system fonts when these are missing.

Credits

Built on two projects:

  • awesome-design-md by VoltAgent — curated DESIGN.md files from popular brand design systems, used as the visual reference library for color palettes, shadow philosophy, border radius, and whitespace rhythm.
  • ljg-card by lijigang — the content-to-PNG visual card skill that inspired the card metaphor, typography rules, content preprocessing pipeline, and taste guidelines.

Gallery

Same text, seven modes — from the philosophical essay Tools and Forgetting.

License

Private skill. Not distributed.

Ver o original no GitHub

Bom para

Relacionados