Supercharge Claude Code with Expert Knowledge A comprehensive framework for AI-assisted development. Install standardized rules, agents, and commands for your projects across multiple technology stacks.
Based on repository activity, growth velocity and community engagement.
34
Growth3/30
Activity10/25
Popularity8/25
Trust13/20
104
Stars
High
Sentiment
Votes
104
README.md
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 — 33 specialized agents (+39 infra agents on-demand), 140 commands across 17 namespaces (234 total), 55 skills, all token-optimized via context: fork and sub-agent model routing.
What's New in v8.18.0
Nouvelle commande /workflow:auto-sprint (2026-06-27, v8.18.0) :
Orchestrateur de sprint de bout en bout : une seule commande joue le rôle Product Owner / Scrum Master et enchaîne start → decompose → validate → implement → PR → CI watch → review → retro → merge. Chaque cérémonie tourne dans un sous-agent au contexte isolé — l'isolation remplace le /clear manuel entre étapes ; la phase d'implémentation assume le rôle conductor inline (réutilise /team:sprint).
PR + CI + merge intégrés (gh) : --auto-merge opt-in (défaut : pause + GO humain) ; échec de gate (validate KO / CI rouge / DoD miss) → auto-fix loop borné (--max-fix-attempts).
Nouveau tutoriel phare : docs/guides/*/10-complete-workflow.md réécrit en walkthrough narré de bout en bout (idée → backlog → sprint → livraison) pour grands débutants — glossaire, modes d'exécution, gates, discipline /clear, exemple TaskFlow + annexe multi-stack. Disponible en 5 langues, lié depuis le README et guides/index.md.
Fraîcheur : compteurs et versions réalignés sur le SSOT (125 core / 219 total commandes, 70 agents, 55 skills, 11 stacks ; Claude Code 2.1.97 min / 2.1.193 rec ; Flutter 3.44, RN 0.85, Symfony 8.1). Suppression de /bmad:init (→ /workflow:init).
Sans doublon : index des guides de migration, en-têtes « narré vs full » (COMMANDS/AGENTS), bannière PLANNED sur MCP-SERVERS, cross-links entre docs concurrentielles, stub docs/SECURITY.md supprimé.
What's New in v8.17.1
Correctifs tests (2026-06-26, v8.17.1) :
tests/scripts/ sous docker sans bash : les 12 fichiers shell-out bash "*.sh" échouaient en /bin/sh: bash: not found sous busybox/alpine. Nouveau target make test-scripts-docker (image bash, node:24) + garde tests/scripts/bash-available.test.mjs.
Audit exhaustif multi-agents + mises à jour 2026-06-24 (v8.17.0) :
Versions infra : Docker 29.6.1, OpenTofu 1.12.2, Ansible 2.21.1, Helm 4.2.2, Node.js 24.x Active LTS.
Sécurité : OWASP Top 10:2025 complet (10/10 catégories dans la règle 11) ; RS256 marqué DEPRECATED ; guidance Argon2id précisée ; hook templates protect-files.json / quality-gate.json corrigés (anti-pattern $TOOL_INPUT → lecture stdin JSON).
Modèles : IDs canoniques dans la table d'effort (Opus 4.8 / Sonnet 5 sorti 2026-06-30, intro $2/$10 / Haiku 4.5). Sonnet 5 remplace 4.6 comme Sonnet courant. Fable 5 / Mythos 5 suspendus (directive export-control US, 2026-06-12) — retirés des recommandations.
CLI : /goal, /usage, /workflows documentés dans CLI-REFERENCE.md.
Optimisation tokens : ENABLE_PROMPT_CACHING_1H et fallbackModel activés dans settings.json distribué.
Trivy action : aquasecurity/trivy-action mis à jour (v8.16.0) — scan CVE supply chain.
Download artifact : actions/download-artifact bumped de 4.2.1 à 8.0.1 (v8.16.0).
CodeQL action : github/codeql-action mis à jour (v8.16.0) — analyse SAST.
Playwright : @playwright/test mis à jour de 1.60.0 à 1.61.0 dans /website (v8.16.0).
Patch de stabilité : correctifs de dernière minute (v8.16.1).
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 (CSP font-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/:id renvoie body). 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 à epic_id inconnu 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-only sprint-status.yaml.
Fiabilité BMAD + distribution AgentTeams (v8.13.0) :
/gate:* réparé : ((var++)) sous set -e avortait 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/*.sh désormais installés automatiquement avec les commandes /team:* (fini le warning « scripts MISSING » ; dashboard coût et --ralph-mode disponibles 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).
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.
126 commandes sur 15 namespaces (220 total avec infra/projet) -- namespace /paperclip:* (8 commandes) ajouté, 55 skills disponibles.
Claude Code 2.1.193 -- version recommandée (Opus 4.8, Artifacts, claude mcp login, /cd, sous-agents imbriqués, effort: ultracode).
# 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
33 default agents + 39 infra agents on-demand -- reviewers, architects, coaches that know your stack deeply (72 total potentially installable)
139 slash commands across 17 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
| 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 126 commands across 15 core namespaces (220 total including infra and project management).
Installation
Platform: Linux and macOS. Windows is not tested and not officially supported.
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 |
| Tutorial: Complete Workflow | End-to-end narrated walkthrough: idea → backlog → sprint → ship |
| Learning Paths | Beginner → Intermediate → Advanced progression |
| Installation | All installation methods |
| Configuration | Project configuration |
| CLI Reference | Full CLI documentation |
| Commands | All 126 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 Guide | Format migration + version-upgrade index (v3→v8) |
| Migration v6→v7 | Upgrade from v6 to v7 |
| Migration v7→v8 | Upgrade from v7 to v8 |
| Agent Teams Guide | Multi-agent team orchestration |
| Cheat Sheet | Quick command reference card |
| 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 |
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. |
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.
Ecosystem Role
Standard MoltPulse indexed agent.
Embed Badge
Show off your Pulse Score in your GitHub README to build trust and rank higher.
