convez.ai
SkillProdutividadeIntermediário
grkmaltunbs

Flutter App Bundle para Claude Code

grkmaltunbs/flutter-app-bundle

Desenvolvimento Flutter assistido por IA com planejamento, implementação e testes passo a passo

Ver no GitHub
InicialPluginScore · de 0 a 100

Kit iniciador completo para desenvolvimento Flutter com Claude Code em fluxo human-in-the-loop. Inclui 9 agentes especializados, comandos interativos, templates de requisitos e motor autônomo de builds headless via Agent SDK.

Instalação

$npx convez-ai grkmaltunbs/flutter-app-bundle

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
42%
Último commit
há 1 semana
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Inicializar projetos Flutter com entrevista estruturada de 29 categorias de requisitos
  • Implementar features incrementais com agentes especializados (arquiteto, desenvolvedor, debugger, tester)
  • Executar builds autônomos completos end-to-end sem intervenção humana usando autobuild.py
  • Revisar código, refatorar, gerenciar dependências e preparar releases via comandos slash

Veredito

Skill robusta e bem-documentada que democratiza o desenvolvimento Flutter com Claude Code, reduzindo fricção na comunicação entre humano e IA. Estrutura clara com agentes especializados, mas exige disciplina no fluxo human-in-the-loop para não perder controle do projeto.

README

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

Flutter App Bundle for Claude Code

A starter kit for building Flutter apps with Claude Code. Human-in-the-loop workflow — you direct, Claude Code implements.

What's included

.claude/
├── commands/       Slash commands (/init-app, /step, /qa, /plan-status, etc.)
├── agents/         9 specialist agents (architect, developer, debugger, qa, …)
├── skills/         Official Dart + Flutter skills
└── settings.json   Permissions (allow/ask/deny lists)
templates/          CLAUDE.md, PRODUCT_SPEC, PROJECT_PLAN, HUMAN_SETUP templates
runner/             autobuild.py — headless autonomous build loop (Agent SDK)
docs/
├── requirements-checklist.md  29-category intake coverage engine
├── flutter-rules.md           Primary, authoritative Flutter/Dart rules
├── autobuild.md               Unattended autonomous builds (setup + safety)
└── lessons-learned.md         Known pitfalls and fixes

Quick start

  1. Clone into your project directory:

    mkdir my-app && cd my-app
    git clone https://github.com/grkmaltunbs/flutter-app-bundle.git .
    
  2. Set up MCP (one-time):

    claude mcp add dart -- dart mcp-server
    

    Restart Claude Code. Verify: claude mcp list shows dart: ✓ Connected.

  3. Run init:

    claude
    /init-app
    

    This walks you through: app idea → design intake → a 29-category requirements interview → an assumptions gate (confirm/override every default so nothing is forgotten) → writes PRODUCT_SPEC.md → derives PROJECT_PLAN.md from it → flutter create → setup walkthrough → smoke test.

  4. Build step by step:

    /step           # implement next pending step
    /step auth      # implement a specific step by id
    /plan-status    # see progress
    
  5. Or build the whole plan unattended with the Agent SDK runner:

    pip install -r runner/requirements.txt
    caffeinate -i python3 runner/autobuild.py   # touch .autobuild-stop to halt
    

    It runs /step end-to-end for every pending step — implement, test, verify on the iOS + Android simulators, commit, push — stopping only when done, blocked on a human-only task, or a guardrail/budget trips. See docs/autobuild.md.

Commands

CommandWhat it does
/init-appFull project initialization (idea → design → setup → scaffold)
/step [id]Implement the next (or named) step from the plan
/plan-statusShow progress on all plan steps
/plan-extendAdd, split, or remove plan steps
/feature <desc>Implement a feature outside the plan
/fix <desc>Debug and fix a bug
/refactor <desc>Refactor with tests
/review [files]Code review (read-only)
/test [files]Add or improve tests
/qa [scope]Run/observe the app on iOS + Android simulators (fakes)
/ship [args]Prepare a release
/codegen [args]Run build_runner / gen-l10n
/cleanClean and rebuild
/deps [args]Manage dependencies

Agents

Specialist agents invoked automatically by commands:

  • flutter-architect — Plans features (no code, just design)
  • flutter-developer — Implements features (Bloc + clean architecture)
  • flutter-debugger — Finds and fixes bugs
  • flutter-refactorer — Restructures code (test-driven)
  • flutter-releaser — Builds and prepares releases
  • flutter-reviewer — Reviews code (read-only findings)
  • flutter-tester — Writes unit/bloc/widget/integration tests
  • flutter-qa — Runs the app on iOS + Android simulators, drives every flow, reports runtime errors and overflow (read-only)
  • flutter-ui-designer — Builds polished UI components

Requirements

  • Flutter SDK (stable channel)
  • Claude Code CLI (claude login)
  • Dart MCP: claude mcp add dart -- dart mcp-server
  • Firebase CLI (optional, if using Firebase): npm install -g firebase-tools

Verification approach

No Firebase emulators, no flutter-skill dependency. Every backend sits behind a repository interface with a seeded in-memory fake, wired under a demo flavor (--dart-define=APP_ENV=demo). All verification runs that flavor on real simulators — offline, deterministic, never touching the live project.

Each /step is gated by the flutter-qa agent before it counts as done:

  1. flutter analyze + flutter test — static + unit/bloc/widget tests
  2. The new flow and its dependent flows driven via integration_test on an iOS simulator and an Android emulator
  3. Dart MCP runtime-error sweep — zero unhandled exceptions
  4. Responsive pass — zero render (overflow) errors across the size matrix (iPhone SE → Pro Max → iPad; small/Pixel/tablet Android) at textScale 1.0 & 2.0
  5. xcrun simctl io "<device>" screenshot — visual capture

Run it anytime with /qa.

Architecture

Default stack (customizable during /init-app):

  • State: flutter_bloc + freezed
  • Routing: go_router
  • DI: get_it + injectable (env-scoped: demo | prod)
  • Local DB: drift
  • Backend: Firebase (optional) — behind repository interfaces with fakes
  • Payments: RevenueCat (purchases_flutter); .storekit + fakes for sims
  • Lints: very_good_analysis
  • Testing: bloc_test, mocktail, integration_test (demo flavor on simulators)

Clean architecture with vertical feature slices: domain/ (pure Dart) → data/ (infrastructure + fakes/) → presentation/ (Flutter + Bloc)

Lessons learned

See docs/lessons-learned.md for known pitfalls:

  • Why fakes, not Firebase emulators
  • Firebase bootstrap order
  • iOS cold build times
  • CocoaPods conflicts
  • APNS on the iOS simulator
  • And more

License

MIT

Ver o original no GitHub

Bom para

Relacionados