convez.ai
AgenteDesenvolvimento WebAvançado
HarambeeAI

Bayes Ship

HarambeeAI/bayes-ship

Da ideia ao protótipo deployado em um pipeline único

Ver no GitHub
InicialPluginScore · de 0 a 100

Sistema end-to-end de entrega de produtos para Claude Code que leva uma ideia de conceito até prototipo deployado em um pipeline estruturado. Combina engenharia de contexto, pesquisa paralela, geração de PRD, arquitetura, TDD com review em dois estágios e deployment automático.

Instalação

$npx convez-ai HarambeeAI/bayes-ship

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
1
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 7 meses
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Prototipagem rápida de produtos com ciclo completo
  • Desenvolvimento estruturado com TDD obrigatório
  • Pesquisa técnica paralela e arquitetura documentada
  • Deploy automático em Railway/Vercel

Veredito

Ferramenta madura e opinada para ciclo completo de desenvolvimento. Requer disciplina rigorosa (TDD, review em dois estágios, git worktrees) mas entrega automação real do conceito ao deploy. Melhor para times que valorizam estrutura e reproducibilidade.

README

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

Bayes Ship

End-to-end product delivery system for Claude Code. From idea to deployed prototype in one pipeline.

What it does

Bayes Ship takes a product from raw idea to deployed software through a structured pipeline: Discovery Q&A → Parallel Research → PRD Generation → Architecture → TDD Implementation with Two-Stage Review → QA → Deployment.

It combines GSD's context engineering with Superpowers' execution discipline, adds a comprehensive pre-implementation pipeline and domain-specific skills for African markets.

Installation

Option 1: From marketplace (recommended)

/plugin marketplace add HarambeeAI/bayes-ship
/plugin install bayes-ship@bayes-ship-marketplace

Option 2: Local development

git clone https://github.com/HarambeeAI/bayes-ship.git

Then inside Claude Code:

/plugin marketplace add /absolute/path/to/bayes-ship
/plugin install bayes-ship@bayes-ship-marketplace

Quick Start

/bayes-ship:ship-new          # Initialize project
/bayes-ship:ship-discover      # Requirements Q&A
/bayes-ship:ship-research      # Parallel tech research (autonomous)
/bayes-ship:ship-spec          # Generate PRD (you approve)
/bayes-ship:ship-architect     # System architecture (you approve, then choose build mode)
/bayes-ship:ship-auto          # Autonomous build — walk away, come back to a prototype

The Pipeline

PhaseCommandYour InputWhat Happens
Setup/bayes-ship:ship-newName, description, users, platformCreates project scaffold
Discovery/bayes-ship:ship-discover~9 Q&A answers (A/B/C/D)Extracts requirements
Research/bayes-ship:ship-researchNone (autonomous)Parallel agents research tech choices
Spec/bayes-ship:ship-specApprove PRDGenerates and reviews PRD
Architecture/bayes-ship:ship-architectApprove architectureDefines stack, data model, APIs, ADRs
Build/bayes-ship:ship-autoNone (autonomous)Plans, builds, tests, reviews all phases
QA/bayes-ship:ship-testNone (autonomous)Verification, coverage, security scan
Finish/bayes-ship:ship-finish-branchMerge strategyMerge/PR/keep/discard
Deploy/bayes-ship:ship-deployCredentialsRailway + Vercel deployment

Build Modes

After architecture approval, choose how to build:

  • Guided (default): You run /bayes-ship:ship-plan N then /bayes-ship:ship-execute for each phase. Full control.
  • Autonomous (/bayes-ship:ship-auto): Builds everything without stopping. TDD and two-stage review still mandatory. Decisions logged to auto-decisions.md.
  • Semi-autonomous (/bayes-ship:ship-auto --checkpoints): Pauses between phases for a status check.

What Makes It Different

From GSD

  • Full product lifecycle (GSD starts at planning, we start at discovery)
  • Parallel research agents
  • Architecture Decision Records
  • Deployment pipeline
  • Autonomous build mode

From Superpowers

  • Structured command-driven workflow (vs invisible auto-triggering)
  • Domain-specific skills (M-Pesa, BetterAuth, Next.js, Nest.js)
  • File-based state persistence across sessions
  • Research phase with parallel agents
  • Autonomous build mode
  • Deployment pipeline

Adopted from Superpowers

  • Two-stage review (spec compliance → code quality)
  • Strict TDD (RED-GREEN-REFACTOR)
  • Git worktree isolation
  • Systematic 4-phase debugging
  • Verification-before-completion
  • YAGNI enforcement
  • "Human partner" agent framing

Included Skills (44 total)

Core Process (always installed — 9 skills)

test-driven-development · systematic-debugging · git-worktrees · verification-before-completion · two-stage-review · context-engineering · atomic-planning · coding-standards-ecc* · output-skill*

Frontend (optional — 5 skills)

taste-skill* · soft-skill* · minimalist-skill* · redesign-skill* · frontend-patterns*

Backend (optional — 2 skills)

backend-patterns* · api-design*

AI/Agents (optional — 8 skills)

langchain-fundamentals** · langchain-middleware** · langchain-rag** · langchain-dependencies** · langgraph-fundamentals** · langgraph-persistence** · agentic-engineering* · agent-eval*

Quality (optional — 3 skills)

security-scan* · accessibility* · ai-regression-testing*

Auth & Payments (optional — 3 skills, Bayes custom)

betterauth-skill · daraja-mpesa · paystack-skill

DevOps (optional — 2 skills, Bayes custom)

devops-skill · railway-deploy

Pipeline Commands (17 skills)

ship-new · ship-discover · ship-research · ship-spec · ship-architect · ship-worktree · ship-plan · ship-execute · ship-auto · ship-test · ship-review · ship-finish-branch · ship-deploy · ship-status · ship-resume · ship-bugfix · ship-settings

* Adapted from everything-claude-code (ECC) or taste-skill ** Adapted from langchain-skills

Agents

14 specialized agents dispatched as subagents with clean context windows:

discovery-agent · researcher · synthesizer · architect · spec-document-reviewer · planner · implementer · spec-compliance-reviewer · code-quality-reviewer · verifier · deployer · debugger · documenter · tester

Settings

/bayes-ship:ship-settings                    # View all settings
/bayes-ship:ship-settings tdd_mode relaxed   # Change a setting
SettingDefaultOptions
build_modeguidedguided / autonomous / semi-autonomous
tdd_modestrictstrict / relaxed
discover_modeinterviewinterview / assumptions
auto_chainfalsetrue / false
use_worktreestruetrue / false
model_routingautoauto / manual
payment_providerdarajadaraja / paystack / stripe
auth_providerbetterauthbetterauth / nextauth / custom

Utility Commands

/bayes-ship:ship-status    # Progress dashboard
/bayes-ship:ship-resume    # Resume after session restart
/bayes-ship:ship-bugfix    # Systematic bug diagnosis + fix
/bayes-ship:ship-review    # Manual two-stage code review

State Files

All state persists in .claude/state/:

.claude/state/
├── PROJECT.md          # Project basics
├── CONTEXT.md          # Discovery output
├── RESEARCH.md         # Research synthesis
├── PRD.md              # Approved spec
├── ARCHITECTURE.md     # Approved architecture
├── ROADMAP.md          # Phase ordering
├── decisions.md        # Architecture Decision Records
├── progress.json       # Task tracking (survives sessions)
├── settings.json       # User preferences
├── auto-decisions.md   # Autonomous mode decision log
├── REVIEW.md           # QA results
├── research/           # Individual research files
└── plans/              # Execution plans per phase

Requirements

  • Claude Code (latest version with plugin support)
  • Git (for worktrees and version control)
  • Node.js 18+ (for project setup)

License

MIT

Author

Bayes Consulting — bayesconsulting.com Built by Antony Sure

Ver o original no GitHub

Bom para

Relacionados