Claude Craft
TheBeardedBearSAS/claude-craft
Workflow de sprint, QA baseado em navegador e reviewers otimizados para Claude Code
Ver no GitHubFramework abrangente para desenvolvimento assistido por IA que fornece workflow de sprint, agentes multi-stack, testes de aceitação no navegador e otimização de tokens para equipes usando Claude Code. Inclui 31 agentes especializados, 125 comandos e 48 skills organizados em 15 namespaces.
Instalação
npx convez-ai TheBeardedBearSAS/claude-craft1 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
Casos de uso
- Implementar workflow BMAD v6 (Analyze → Plan → Design → Implement → QA) em projetos Claude Code
- Executar testes de aceitação automatizados e gerar testes de regressão a cada correção de bug
- Otimizar consumo de tokens com roteamento de modelo por tarefa (Haiku/Sonnet/Opus) e prompt caching
- Aplicar regras de arquitetura limpa (SOLID/TDD) automaticamente com reviewers especializados por stack
Veredito

Framework maduro e opinião forte para equipes de desenvolvimento que já usam Claude Code em produção. Seu diferencial real — workflow orquestrado, QA no navegador com captura automática de regressão e otimização de tokens — vai além de prompts genéricos. Recomendado para sprints estruturados em stacks múltiplos.
README
A tradução para o português deste README ainda está na fila.
Claude Craft — Sprint workflow, multi-stack reviewers, and browser QA for Claude Code teams
The AI framework where bugs don't come back. Sprint workflow, a continuous build loop, browser-based acceptance testing, and token-optimized multi-stack reviewers for tech leads adopting Claude Code with their team. 11 stacks, 5 languages, BMAD v6.
Why teams pick Claude Craft
Four things Claude Code + an Anthropic cookbook won't give you — they're workflow and orchestration, not prompts:
| Differentiator | What it does |
|---|---|
| 🏆 QA Recette (browser-based) | /qa:recette drives Chrome to run acceptance tests and auto-generates a regression test for every bug you fix — the Golden Rule: a fixed bug never comes back. No other Claude Code framework does this. |
| 🚀 BMAD v6 sprint workflow | Analyze → Plan → Design → Implement → QA, executed by Claude Code directly with quality gates. Quick Flow < 5 min, Standard < 15 min, Enterprise < 30 min. |
| 🔁 Ralph Wiggum loop | /common:ralph-run runs Claude in a continuous loop until your Definition of Done holds — adaptive circuit breaker, native /goal integration, DoD validators. |
| ⚡ RTK token optimization | Forked sub-agents, context: fork on heavy skills, 1h prompt caching, Haiku/Sonnet/Opus routing per task. Target 55-65% token savings vs raw Claude Code. |
Plus 11 stack-specific reviewers (@symfony-reviewer, @react-reviewer, @python-reviewer…) that enforce SOLID/TDD/Clean Architecture so you don't re-type your conventions every sprint.
Why not just Claude Code + a cookbook? Claude already knows React, Symfony, and Flutter patterns — so Claude Craft's value isn't the prompts, it's the workflow and orchestration around them: a sprint lifecycle with gates, a continuous loop with a real DoD, browser-based regression capture, and per-task token routing. Those are the parts a per-stack cookbook doesn't cover.
A comprehensive framework for AI-assisted development with Claude Code. Install standardized rules, agents, and commands for your projects across multiple technology stacks — 31 specialized agents (+39 infra agents on-demand), 125 commands across 15 namespaces, 48 skills, all token-optimized via context: fork and sub-agent model routing.
What's New in v8.15.0
Refonte du design de l'interface web Kanban (v8.15.0) :
- Thème dark-only opiniâtre d'après une maquette Claude Design : accent acid-lime (oklch, distinct des six teintes de statut), typographie Space Grotesk + JetBrains Mono auto-hébergée via
@fontsource(CSPfont-src 'self'). - Panneau Tweaks : code couleur des cartes (statut / priorité / TDD / epic), accent et densité, persistés en
localStorage. - Les 6 vues restylées (board, backlog, sprints, burndown, deps, docs) ; drag&drop, navigation clavier,
<dialog>natifs et a11y préservés. Coloration en helpers JS purs testés.
Interface web Kanban — accès à tous les artefacts BMAD v6 (v8.14.0) :
- Vue Sprints dédiée : navigation listant tous les sprints (points, stories, badges goal/review/retro) ; le détail rend le Goal, les Stories avec leurs Tasks dépliables, et les sections Review et Retro en markdown (
GET /api/sprints,GET /api/sprints/:id). - Docs élargis : les corps markdown des EPIC, US, sprint-goal/board/review/retro sont désormais navigables dans le DocsView.
- Détail US enrichi : la modale du board affiche la liste des tâches associées et la description de la story (
/api/stories/:idrenvoiebody). Tests RED-first, suite complète verte.
Corrections de l'interface web Kanban (v8.13.1) :
- Dashboard Kanban réparé : 4 bugs corrigés en TDD sur le SPA Svelte 5 (
claude-craft kanban) — vue Backlog vide (stories àepic_idinconnu désormais regroupées au lieu d'être perdues), raccourcis clavier inopérants (synchronisation focus + reducer pur), détails de carte inaccessibles (modale au clic/Entrée), et feedback visible (toast) au lieu d'un échec silencieux sur les cartes read-onlysprint-status.yaml.
Fiabilité BMAD + distribution AgentTeams (v8.13.0) :
/gate:*réparé :((var++))sousset -eavortait les scripts de gate/hook après le 1er check (post-incrément = exit 1 quand le compteur vaut 0). 68 occurrences corrigées sur 6 fichiers + tests de non-régression (make test-bmad).- Scripts AgentTeams livrés :
Tools/AgentTeams/lib/*.shdésormais installés automatiquement avec les commandes/team:*(fini le warning « scripts MISSING » ; dashboard coût et--ralph-modedisponibles out-of-the-box). Install/refresh manuel :make install-agentteams TARGET=..
Audit exhaustif + durcissement (v8.12.0) :
- Audit multi-domaines (sécurité, DX, concurrentiel, fiabilité, tokens/modèles, docs, architecture) mené par une équipe d'agents avec devil's advocates et vérification de fraîcheur des 14 stacks -- 80 findings, tous les P0/P1/P2 corrigés.
- Sécurité : CVE FrankenPHP (CVE-2026-45062) et Docker (CVE-2026-33997) patchées ; hook de sécurité distribué corrigé (
exit 2) ; durcissement Kanban (CSRF, COEP) et CI supply chain (digests épinglés, SHA256). - Optimisation tokens : templates
settings.jsondistribués corrigés (IDs de modèles valides +CLAUDE_CODE_FORK_SUBAGENT). - 988 Vitest tests green, mutation testing bloquant sur PRs.
Kanban BMAD v6 integration (v8.11.0):
- Kanban ingère
.bmad/sprint-status.yaml-- le board Kanban lit les sprints directement depuis le fichier YAML BMAD v6 en lecture seule (icône verrou), sans dépendance SaaS. Les projets BMAD v6 n'ont plus le board vide.
Releases 8.8.x → 8.10.x :
- MIT-only strict (v8.8.0) -- Claude Craft est 100 % open-source MIT, aucune licence commerciale ou enterprise. Stratégie open-core abandonnée.
- Parité i18n stricte (v8.8.2) -- la CI bloque désormais si un fichier traduit est à < 80 % de la taille de l'anglais. Dette i18n résorbée (gap 101 → 0).
- Branding the-bearded-bear.com (v8.10.1) -- migration complète des domaines vers
the-bearded-bear.com, normalisation de l'organisation GitHub. - 125 commandes sur 15 namespaces -- namespace
/paperclip:*(8 commandes) ajouté, 48 skills disponibles. - Claude Code 2.1.168 -- version recommandée (Opus 4.8, Dynamic Workflows,
effort: ultracode).
← Versions antérieures (v8.0 → v8.7) : voir le CHANGELOG et .claude/COMPATIBILITY.md.
Install and First Result
# Install to your project (picks your tech stack interactively)
npx @the-bearded-bear/claude-craft
# Or install directly
npx @the-bearded-bear/claude-craft install ~/my-project --tech=react --lang=en
# Zero-prompt install (auto-detects your stack + locale, target < 2 min)
npx @the-bearded-bear/claude-craft install --auto
# Install from a team config URL (Gist or internal endpoint)
npx @the-bearded-bear/claude-craft install --from=https://org.example/cc-team.json
# Add a community skill from npm (claude-craft-skill-* convention)
npx @the-bearded-bear/claude-craft skill add claude-craft-skill-foo
# Open Claude Code and run your first audit
claude
/team:audit
That's it. You get an architecture, security, and quality audit of your project in minutes.
New to Claude Craft? Run /common:getting-started in Claude Code for a 10-minute guided tour that shows you the 3 most valuable commands for YOUR project.
See Quickstart for a step-by-step walkthrough with expected output at each stage.
Why Claude Craft?
Claude Code is powerful on its own. Claude Craft makes it consistent and team-ready:
- Standardized rules -- SOLID, Clean Architecture, TDD enforced across your team, not just suggested
- 31 default agents + 39 infra agents on-demand -- reviewers, architects, coaches that know your stack deeply (70 total potentially installable)
- 125 slash commands across 15 namespaces -- repeatable workflows for audits, code generation, sprint management
- Quality gates -- automated checks at every stage from PRD to deployment
- 5 languages -- English, French, Spanish, German, Portuguese
Supported Technologies
| Stack | Version | Install Command |
|---|---|---|
| Symfony / PHP | 8.1 / PHP 8.4+ | --tech=symfony |
| React | 19.2 + Compiler 1.0 | --tech=react |
| Flutter / Dart | 3.44 / Dart 3.12 | --tech=flutter |
| Python | 3.14+ / FastAPI | --tech=python |
| Angular | 22 | --tech=angular |
| Vue.js | 3.5+ (3.6 beta Vapor) | --tech=vuejs |
| React Native | 0.86 (New Architecture) | --tech=reactnative |
| C# / .NET | 10 LTS / C# 14 | --tech=csharp |
| Laravel | 13.x / PHP 8.3+ (8.5 recommandé) | --tech=laravel |
| PHP | 8.5 | --tech=php |
| Paperclip | 2026.609.0 | --tech=paperclip |
| Docker | 29.5.3 (CVE-2026-33997) | --tech=docker |
| Coolify | v4.1.2 | --tech=coolify |
| Kubernetes | 1.36.1 | --tech=kubernetes |
| OpenTofu | 1.12.1 | --tech=opentofu |
| Ansible | 2.21.0 | --tech=ansible |
| Hcloud | 1.61+ | --tech=hcloud |
| PgBouncer | 1.25.2 (CVE-2026-6664/6665/6666/6667 patched) | --tech=pgbouncer |
| FrankenPHP | 1.12.4 (CVE-2026-45062 patched) | --tech=frankenphp |
See Technologies for full details.
What's Included
| Category | Count | Examples |
|---|---|---|
| Agents | 31 default (+ 39 infra on-demand) | @tdd-coach, @api-designer, @symfony-reviewer, @kubernetes-architect, @hcloud-architect |
| Commands | 133 across 15 namespaces | /workflow:init, /team:audit, /react:generate-component |
| Skills | 55 | Architecture, testing, security best practices |
| Templates | 21 | Code generation patterns, BMAD project templates |
| Checklists | 10 | Commit, feature, release quality gates |
See Agents | Commands | Skills
Workflow Tracks
Claude Craft adapts to your project complexity with three development tracks:
| Track | Setup | Phases | Best For |
|---|---|---|---|
| Quick Flow | < 5 min | Implementation only | Bug fixes, hotfixes |
| Standard | < 15 min | Plan > Design > Implement | New features, refactoring |
| Enterprise | < 30 min | Analyze > Plan > Design > Implement | Platforms, migrations |
/workflow:init # Auto-detects complexity
/workflow:init --quick # Bug fix mode
/workflow:init --enterprise # Full methodology
See BMAD Practical Guide for the complete project management framework.
Key Commands
These are the commands you'll use most:
| Command | What It Does |
|---|---|
/workflow:init | Start a development workflow (auto-detects track) |
/team:audit | Full project audit (architecture, security, quality) |
/common:pre-commit-check | Validate before committing |
/sprint:next-story | Get next story ready for development |
/qa:tdd | Test-Driven Development flow |
/gate:validate-story | Check story meets Definition of Done |
/{tech}:check-architecture | Verify architecture compliance |
/{tech}:generate-* | Generate code following project patterns |
/common:ralph-run "task" | Run Claude in continuous loop until task is done |
/qa:recette | Automated acceptance testing via Chrome |
See CLI Reference for all 125 commands across 15 namespaces.
Installation
Platform: Linux and macOS. Windows is not tested and not officially supported.
NPX (Recommended)
npx @the-bearded-bear/claude-craft install ~/my-project --tech=symfony --lang=en
Clone + Makefile
git clone https://github.com/TheBeardedBearSAS/claude-craft.git
cd claude-craft
make install-symfony TARGET=~/my-project RULES_LANG=en
YAML Configuration (Monorepos)
# claude-projects.yaml
projects:
- name: "my-monorepo"
root: "~/Projects/my-monorepo"
lang: "fr"
modules:
- path: "frontend"
tech: react
- path: "backend"
tech: symfony
make config-install PROJECT=my-monorepo
See Installation Guide | Configuration
Use Claude Craft Without Claude Code
Claude Craft's principles and rules are available as pre-built bundles for other AI surfaces. No installation required — paste and go.
| Surface | Bundle | Doc |
|---|---|---|
| ChatGPT / GPT-5 | bundles/chatgpt/claude-craft-bundle.md | Multi-IDE Guide |
| Claude.ai (Web) / Claude Projects | bundles/claude/claude-craft-bundle.md | Multi-IDE Guide |
| Gemini / Cursor / Windsurf | bundles/gemini/claude-craft-bundle.md | Multi-IDE Guide |
| Codex CLI / other agents | bundles/claude/claude-craft-bundle.md | Multi-IDE Guide |
See bundles/README.md for platform-specific installation instructions and token budgets.
What Gets Installed
your-project/.claude/
CLAUDE.md # Minimal config (~200 tokens, auto-loaded)
INDEX.md # Quick reference summaries
references/ # Full documentation (loaded on-demand via @)
agents/ # AI specialist definitions
commands/ # Slash commands
skills/ # Best practices
checklists/ # Quality gates
templates/ # Code generation patterns
hooks/ # Pre/Post tool execution scripts
mcp/ # MCP server templates
Context usage is optimized: ~3,500 tokens always loaded vs ~70,000 if everything were inline (95% reduction).
Documentation
| Document | Description |
|---|---|
| Quickstart | Get results in 10 minutes |
| Installation | All installation methods |
| Configuration | Project configuration |
| CLI Reference | Full CLI documentation |
| Commands | All 125 commands |
| Agents | All 31 default agents (+ 39 infra on-demand) |
| Skills | Best practices reference |
| Technologies | Stack-specific guides |
| BMAD Guide | Project management framework |
| Hooks | Pre/Post tool execution |
| MCP | Model Context Protocol integration |
| Ecosystem | Curated third-party token/context/review tools |
| Privacy Policy | Data protection and GDPR compliance |
| FAQ | Common questions |
| Troubleshooting | Problem solving |
| Migration v7 | Upgrade from previous versions |
| Skills Publishing | Guide for publishing skills |
| Compatibility | Claude Code version compatibility |
User Guides (Multilingual)
Step-by-step tutorials available in 5 languages:
| Guide | EN | FR | ES | DE | PT |
|---|---|---|---|---|---|
| Getting Started | EN | FR | ES | DE | PT |
| Feature Development | EN | FR | ES | DE | PT |
| Bug Fixing | EN | FR | ES | DE | PT |
All guides | Project Creation | Tools Reference | Troubleshooting | Backlog Management
Project Governance & Sustainability
Claude Craft is maintained by The Bearded CTO, a solo founder with deep involvement in the AFUP and Symfony French ecosystem.
| Item | Status |
|---|---|
| Funding model | Bootstrapped, community-driven — no VC. 100% open-source MIT, sustained by community contributions. |
| Maintenance commitment | Active development since 2026-01. Targeting weekly minor releases, monthly minor versions. |
| Bus factor | Currently 1 (solo maintainer). Co-maintainer search open — looking for one tech lead from the AFUP / Symfony / Flutter / React community. |
| Succession plan | Documented in CHARTER.md. MIT license guarantees indefinite community fork rights if maintainer disappears. |
| Roadmap visibility | GitHub Issues + CHANGELOG.md + recurring audit/YYYY-MM-DD-* reports |
| Decision process | RFC via GitHub Discussions for breaking changes. ADRs in docs/adr/ for architectural choices. |
| Security disclosure | See SECURITY.md. 90-day disclosure timeline, GPG-signed advisories. |
| License | MIT (free, perpetual). Claude Craft is and remains 100% open-source — no commercial, enterprise, or proprietary tier. See CHARTER.md. |
Contributing
Contributions are welcome! Please read our Contributing Guide.
License
This project is licensed under the MIT License - see the LICENSE file for details.
Warranty Disclaimer
THIS SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. See LICENSE for full terms.
Built for Claude Code by Anthropic. Inspired by Clean Architecture and Domain-Driven Design principles.