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

Phase	Command	Your Input	What Happens
Setup	/bayes-ship:ship-new	Name, description, users, platform	Creates project scaffold
Discovery	/bayes-ship:ship-discover	~9 Q&A answers (A/B/C/D)	Extracts requirements
Research	/bayes-ship:ship-research	None (autonomous)	Parallel agents research tech choices
Spec	/bayes-ship:ship-spec	Approve PRD	Generates and reviews PRD
Architecture	/bayes-ship:ship-architect	Approve architecture	Defines stack, data model, APIs, ADRs
Build	/bayes-ship:ship-auto	None (autonomous)	Plans, builds, tests, reviews all phases
QA	/bayes-ship:ship-test	None (autonomous)	Verification, coverage, security scan
Finish	/bayes-ship:ship-finish-branch	Merge strategy	Merge/PR/keep/discard
Deploy	/bayes-ship:ship-deploy	Credentials	Railway + 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

Setting	Default	Options
build_mode	guided	guided / autonomous / semi-autonomous
tdd_mode	strict	strict / relaxed
discover_mode	interview	interview / assumptions
auto_chain	false	true / false
use_worktrees	true	true / false
model_routing	auto	auto / manual
payment_provider	daraja	daraja / paystack / stripe
auth_provider	betterauth	betterauth / 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