convez.ai
SkillDesign & UIIntermediário
videnovnebojsa

Excalidraw Agent Skill

videnovnebojsa/excalidraw-agent-skill

Gera diagramas Excalidraw offline com agents isolados e exportação animada

Ver no GitHub
BomPluginScore · de 0 a 100

Skill para Claude Code que gera diagramas Excalidraw estilizados com orquestração multi-agente, renderização offline e exportação SVG animada. Ideal para criar arquiteturas, fluxogramas e diagramas complexos sem sobrecarregar o contexto com JSON.

Instalação

$npx convez-ai videnovnebojsa/excalidraw-agent-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
0
Forks
0
Acompanhando
0
Saúde
100%
Último commit
há 5 meses
Versão
v2.0.0-beta
Issues abertas
0
Contribuidores
1

Casos de uso

  • Gerar diagramas de arquitetura de sistemas e microserviços
  • Criar fluxogramas de login, CI/CD pipelines e mapas de dependências
  • Atualizar e validar diagramas existentes com feedback visual
  • Exportar diagramas como SVG animado para apresentações

Veredito

Ferramenta sólida para geração automatizada de diagramas técnicos dentro do Claude Code. Renderização offline elimina dependências externas, multi-agente permite criação escalável, mas requer instalação local e Node.js 20+. Ideal para equipes que precisam de diagramas consistentes sem sair do fluxo de desenvolvimento.

README

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

excalidraw-agent-skill

CI License: MIT Node.js

A Claude Code skill for generating styled Excalidraw diagrams with isolated diagram-generation agents, an offline renderer, validation, visual critique, and optional animated SVG export.

Use it when you want architecture diagrams, flow charts, CI/CD pipelines, dependency maps, or updates to an existing .excalidraw file without filling your main Claude Code context with thousands of JSON tokens.

Microservices architecture diagram generated by the skill

Generated from one prompt: /excalidraw Draw a system architecture for my Node.js microservices app — semantic colors, layered grouping, and the hand-drawn Excalidraw look, all offline.

Prerequisites

  • Claude Code with local skills support
  • Node.js 20 or newer
  • npm
  • macOS or Linux recommended; the skill documentation uses ~/.claude/... paths

The renderer works offline through @moona3k/excalidraw-export; no browser or Playwright install is required.

Install

git clone https://github.com/videnovnebojsa/excalidraw-agent-skill
cd excalidraw-agent-skill

mkdir -p ~/.claude/skills
cp -r .claude/skills/excalidraw ~/.claude/skills/
cp style-profile.json ~/.claude/skills/excalidraw-style-profile.json

cd ~/.claude/skills/excalidraw
npm install

Installing the skill registers the /excalidraw command. Start a new Claude Code session to pick it up.

First Diagram

Start any Claude Code session and ask:

/excalidraw Draw a login flow: User -> Browser -> API -> Database

The skill writes the final .excalidraw file and rendered PNG to /tmp/diagram-final-<timestamp>.*, then shows the PNG path in the response.

Common Usage

/excalidraw Draw a system architecture for my Node.js microservices app
/excalidraw Draw a login flow and animate it --animate
/excalidraw Draw a CI/CD pipeline --tier=complex
/excalidraw Draw a database replication flow --style=./style-profile.json

To update an existing drawing, include the path in your prompt:

/excalidraw Add an OAuth Provider node between Browser and API.

Existing diagram: .claude/skills/excalidraw/examples/simple-flow.excalidraw

Flags

FlagEffect
--animateExport an animated SVG after diagram approval
--style=<path>Use a style profile for this run instead of the installed default
--tier=simple|standard|complexOverride complexity detection
--fill=hachure|solid|cross-hatch|zigzagOverride shape fill style
--no-validateSkip the visual feedback loop for faster output

Custom Style

The default style profile is installed at:

~/.claude/skills/excalidraw-style-profile.json

Edit that file to change the palette and enforced drawing rules for future diagrams. For one-off runs, pass a profile explicitly:

/excalidraw Draw a customer onboarding flow --style=/path/to/my-style-profile.json

The strongest customization surface is palette.semantic.*, typography.fontFamily, stroke.width, and stroke.roughness; those are checked by validate.js. Some profile fields are advisory or reserved for future work. See customization.md for the exact behavior.

Scripts

Run these from the repository root:

# Render a diagram to PNG
node scripts/render.js .claude/skills/excalidraw/examples/simple-flow.excalidraw /tmp/simple-flow.png

# Validate a diagram against a style profile
node scripts/validate.js .claude/skills/excalidraw/examples/simple-flow.excalidraw style-profile.json

The root scripts forward to the installable skill scripts in .claude/skills/excalidraw/scripts/.

How It Works

User prompt
  -> Design Agent        element inventory and layout sketch
  -> Generation Agent    Excalidraw JSON
  -> Style Review Agent  validate.js plus style-guide.md
  -> Vision Critique     render, inspect, patch, re-render
  -> Confirmation Agent  user-facing summary
  -> Animation Agent     optional animated SVG for --animate

Simple diagrams run inline. Standard diagrams use the full agent pipeline. Complex diagrams are generated in sections and merged before validation and rendering.

Repository Structure

.claude/skills/excalidraw/
├── SKILL.md                 orchestration protocol read by Claude Code
├── package.json             skill runtime dependencies
├── scripts/
│   ├── render.js            offline PNG/SVG renderer
│   ├── validate.js          structure and style checker
│   └── describe.js          diagram fact sheet for visual critique
├── references/
│   ├── generation-guide.md
│   ├── style-guide.md
│   ├── visual-feedback.md
│   ├── animation-guide.md
│   ├── customization.md
│   └── style-extraction.md
├── examples/
└── tests/

Contributing

See CONTRIBUTING.md for local checks, test fixtures, and contribution notes.

Prior Art

License

MIT. See LICENSE.

Ver o original no GitHub

Bom para

Relacionados