The end-to-end product engineering playbook for AI-native shipping.
Idea to shipped product, in weeks not months. Built by CJ Jutba — Product Engineer / Full-Stack / AI-Native.
A reusable, opinionated system for shipping production-ready apps with AI coding agents (Claude Code, Cursor, etc.). It encodes the entire lifecycle — idea validation through portfolio polish — as a copy-paste-able set of artifacts, not a framework.
You drop it into every new project. Your agents read it. Quality is consistent. You ship fast without shipping fragile.
The promise: never start from zero, never re-derive patterns, never let agents drift. Project N+1 always beats project N because Shipkit compounds the lessons.
Most AI-driven projects fail the same way: vague scope, inconsistent patterns, no real verification, mediocre quality at "looks done" stage. Agents are good at generation, mediocre at judgment, and need explicit guardrails.
Shipkit provides those guardrails. It's the difference between using AI and engineering with AI.
| 11-phase playbook | From validation (docs/phases/00) to portfolio polish (docs/phases/10) |
| CLAUDE.md template | The agent constitution — pre-loaded with disciplines, conventions, hard rules |
| 18 reference patterns | Canonical patterns for CRUD, forms, mutations, RBAC, real-time, jobs, i18n, payments, more |
| Subagent prompt library | Plan, build, review, fix templates — including the 6-specialist Council |
| DoD checklists | Module, PR, production-readiness, portfolio readiness |
| Memory / lessons system | Per-project + cross-project compounding wisdom |
| 4 stack presets | Web SaaS, Mobile (RN/Expo), Internal Tool, API-only |
| CI/CD templates | GitHub Actions for CI, E2E, Production Deploy |
| One-command bootstrap | ./scripts/init-project.sh <name> <preset> and you're working |
# Clone Shipkit once
git clone https://github.com/cjjutba/shipkit.git ~/Projects/shipkit
# Bootstrap any new project from it
~/Projects/shipkit/scripts/init-project.sh my-cool-app web-saas
# Open in your editor (Claude Code, Cursor, etc.)
cd ~/Projects/my-cool-appAvailable presets:
| Preset | For |
|---|---|
web-saas (default) |
Public-facing SaaS, marketplaces, B2B/B2C web |
mobile |
React Native + Expo (iOS + Android) |
internal-tool |
Admin panels, ops dashboards |
api-only |
Pure API / backend services |
The init script creates docs/STATE.md pointing at Phase 00. Open docs/phases/00-validation.md and go phase by phase. Each phase tells you exactly what to do, what to hand to agents, and how to know you're ready for the next.
| # | Phase | Owner | Key Output |
|---|---|---|---|
| 00 | Validation & Scoping | Human (alone) | vision.md, locked no-list |
| 01 | Specification | Human + agents | PRD, user stories, data model |
| 02 | Tech Stack & Architecture | Human + agents | tech-stack.md, architecture.md, NFRs |
| 03 | Scaffold & Conventions | Agents | Working repo, CI green, CLAUDE.md filled |
| 04 | Gold-Standard Module | Human-led | One module at highest bar + docs/reference/ populated |
| 05 | Module Backlog | Human | Sequenced backlog with mini-specs |
| 06 | Per-Module Build Loop | Agents orchestrated | Every module shipped at gold-standard quality |
| 07 | Cross-Cutting Hardening | Agents | Perf, a11y, security, UX audits actioned |
| 08 | Production Readiness | Agents | Observability, backups, env hygiene |
| 09 | Deploy & Iterate | Human + agents | Live, monitored, soft-launched |
| 10 | Portfolio Polish | Human + agents | Case study, demos, public artifacts |
Phases 04 and 06 are the highest-leverage. Get them right and the rest is mechanical.
Shipkit codifies these patterns and uses them throughout the playbook:
- Fan-out / Fan-in — parallel research agents (default for planning)
- Specialist Council — 6 parallel reviewers, each with a forced focus area
- Map-Reduce — split work, process in parallel, merge
- Producer-Consumer — one finds, many fix (Phase 07 hardening)
- Scatter-Gather-Decide — research → human gate → execute
- Pipeline — when dependencies are real (rare)
Full reference: docs/ORCHESTRATION.md.
shipkit/
├── README.md ← you are here
├── LICENSE
├── CHANGELOG.md
├── CONTRIBUTING.md
├── docs/
│ ├── START_HERE.md ← entry point for Claude Code in a project
│ ├── PLAYBOOK.md ← 30,000-ft view
│ ├── PRINCIPLES.md ← 15 non-negotiables
│ ├── ORCHESTRATION.md ← multi-agent patterns
│ ├── phases/ ← 00-10, one file per phase
│ ├── reference/ ← 18 gold-standard pattern docs
│ └── stacks/ ← 4 stack presets
├── prompts/
│ ├── plan/ ← research + planning agents
│ ├── execute/ ← build + fix agents
│ ├── review/ ← Specialist Council + lightweight review
│ └── build/ ← parent-level orchestration guides
├── checklists/ ← DoD, PR, production, portfolio
├── memory/
│ ├── lessons.md ← universal lessons (seeded; grows)
│ └── lessons-by-project/ ← per-project snapshots
├── templates/
│ ├── claude-md/ ← CLAUDE.md template (the agent constitution)
│ ├── ci/ ← GitHub Actions workflows
│ ├── env/ ← .env.example
│ ├── readme/ ← project README template
│ └── github/ ← PR + issue templates
└── scripts/
└── init-project.sh ← one-command project bootstrap
Use Shipkit consistently and:
- You never start from zero. Every project inherits the accumulated lessons of every prior project.
- Quality is the default, not the goal. The DoD checklists make "good enough" the floor.
- Agents stay aligned. Every session reads the same CLAUDE.md, gold standard, lessons. No drift.
- You ship fast AND ship sturdy. Phases sized for the weeks-not-months pace without skipping production-readiness.
- Clients see a system, not a hacker. Showing a client your Shipkit-based process is itself a differentiator.
Every project promotes lessons + new patterns + new prompts back into Shipkit. Project N+1 always beats project N.
Project 1 → discovers 3 universal lessons → Shipkit absorbs them
Project 2 → starts with those + discovers 5 more → Shipkit absorbs
Project 5 → builds on the wisdom of 20+ shipped projects
See memory/lessons.md for the running ledger.
SemVer. See CHANGELOG.md.
- MAJOR: breaking changes to the template structure (rare)
- MINOR: new patterns, new phases, new prompts
- PATCH: polish, typos, doc improvements
- The Anthropic Claude Code team for the agent model that makes this workable
- Marty Cagan, Lenny Rachitsky for product-engineering thinking
- The pragmatic-tech-stack consensus (React, TS, Postgres, Tailwind, Stripe) that lets us focus on product
- Every "what did I do wrong this time" reflection that became a Shipkit lesson
MIT. See LICENSE.
Built by CJ Jutba.
If Shipkit helps you ship, star it ⭐ and tell someone about it.