create-01x-project

Scaffold the Claude Code multi-agent build system into any project. One command, two questions.

What It Does

Drops 14 pre-wired Claude Code agents into your project — an orchestrator, four planning agents, a review agent, an architect agent, a TDD build loop, a cache health monitor, and a PR review agent that fixes bot comments automatically. Stack agnostic. Works for any language, framework, or toolchain.

npx create-01x-project

Fill in agent_docs/product-seed.md, open Claude Code, and type Run the orchestrator agent.

Don't have a seed yet? The seed is the single source of truth the agents build from. Generate one before scaffolding — three ways depending on which AI you use:

| Your AI | How to generate the seed | |---|---| | ChatGPT | Product Builder GPT — full ideation workspace with web search | | Claude | Install product-seed.skill — triggers automatically in any conversation | | Any AI | Copy-paste product-seed-prompt.md at the start of a new chat |

Usage

mkdir my-app && cd my-app
npx create-01x-project

  ╔══════════════════════════════════════════╗
  ║   create-01x-project  v1.4.0             ║
  ║   Claude Code agent system scaffolder    ║
  ╚══════════════════════════════════════════╝

  Project name? › my-app

  The following files will be created:

  my-app/
  ├── CLAUDE.md
  ├── README.md
  ├── .gitignore
  ├── agent_docs/
  │   ├── product-seed.md  ← fill this after ideation
  │   └── build/
  └── .claude/
      ├── agents/  ← 14 agents
      │   ├── orchestrator.md  ← invoke this
      │   ├── system-design-agent.md
      │   ├── milestone-agent.md
      │   ├── user-stories-agent.md
      │   ├── design-spec-agent.md
      │   ├── product-brief-agent.md
      │   ├── review-agent.md
      │   ├── architect-agent.md
      │   ├── build-agent.md
      │   ├── test-agent.md
      │   ├── build-review-agent.md
      │   ├── cache-health-agent.md
      │   ├── pr-review-agent.md
      │   └── ui-ux-review-agent.md
      └── commands/
          └── fix-pr-review.md

  Initialise a git repo? › Yes

  ✔ Done!

  Next steps:

  1.  Fill in agent_docs/product-seed.md
  2.  Open this folder in VSCode
  3.  Open Claude Code and type:

       Run the orchestrator agent.

Two questions. A tree. Done.

What Gets Created

your-project/
├── CLAUDE.md                           ← project operating manual
├── doctor.sh                           ← run before starting (checks PinchTab, gh, etc.) (auto-loaded by Claude Code)
├── README.md                           ← how-to guide for this project
├── .gitignore
├── agent_docs/
│   ├── product-seed.md                 ← fill this after ideation
│   └── build/                          ← agents write state here during builds
└── .claude/
    ├── agents/
    │   ├── orchestrator.md             ← the only agent you ever invoke manually
    │   ├── architect-agent.md          ← Phase 0: scaffold + install
    │   ├── system-design-agent.md      ← Phase 1: technical blueprint
    │   ├── milestone-agent.md          ← Phase 1: delivery plan
    │   ├── user-stories-agent.md       ← Phase 1: stories + edge cases
    │   ├── design-spec-agent.md        ← Phase 1: design system + token spec
    │   ├── product-brief-agent.md      ← Phase 1: product positioning
    │   ├── review-agent.md             ← Phase 2: cross-doc alignment check
    │   ├── build-agent.md              ← Phase 3: TDD implementation
    │   ├── test-agent.md               ← Phase 3: test runner + reporter
    │   ├── build-review-agent.md       ← Phase 3: code review + fix notes
    │   ├── ui-ux-review-agent.md       ← Phase 3 gate: validates UI against design-spec
    │   ├── cache-health-agent.md       ← utility: diagnose slow/expensive sessions
    │   └── pr-review-agent.md          ← Phase 4: fix PR bot comments automatically
    └── commands/
        └── fix-pr-review.md           ← manual trigger: /fix-pr-review

The Workflow After Scaffolding

Step 1 — Fill the seed

agent_docs/product-seed.md is the only file you write manually. Paste in what you built during your ideation session — problem, target user, features, constraints, out of scope. The agents read this and produce everything else.

Step 2 — Run the orchestrator

Open the project in VSCode, open Claude Code, type:

Run the orchestrator agent.

Claude Code finds .claude/agents/orchestrator.md automatically from your open workspace.

Step 3 — Approve the gates

The orchestrator runs four planning agents in parallel, then a review agent. It stops and waits at two human gates before writing any code:

✅ PLANNING COMPLETE — GATE 1
5 docs approved. Ready to scaffold.
Type: proceed with scaffold

✅ SCAFFOLD COMPLETE — GATE 2
Type: proceed with milestone 1

Step 4 — Build

The build loop runs story by story — build → test → review → fix — committing as it goes. At the end of each milestone, the orchestrator opens a PR and automatically spawns the pr-review-agent to handle any bot review comments before showing the next gate.

Your total keyboard input for a complete build:

Run the orchestrator agent.
proceed with scaffold
proceed with milestone 1
proceed with milestone 2

Stack Agnostic by Design

The architect agent reads agent_docs/system-design.md — produced by the planning phase — and sets up whatever stack is defined there. It does not assume Next.js, Node, Python, Go, or anything else. If you spec Rust and SQLite in your system design, that's what it installs.

The CLAUDE.md that gets created has placeholder test commands that the architect agent fills in after scaffold, based on what it actually installed.

End-to-End Example — PerishNote

Here's a complete walkthrough using a real product idea: a shared, password-protected checklist that self-destructs after 24 hours. No signups.

1. Scaffold the project

mkdir perishnote && cd perishnote
npx create-01x-project

  Project name? › perishnote

  The following files will be created:

  perishnote/
  ├── CLAUDE.md
  ├── doctor.sh
  ├── README.md
  ├── .gitignore
  ├── agent_docs/
  │   ├── product-seed.md  ← fill this after ideation
  │   └── build/
  └── .claude/
      ├── agents/  ← 14 agents
      │   ├── orchestrator.md  ← invoke this
      │   └── ...
      └── commands/
          └── fix-pr-review.md

  Initialise a git repo? › Yes

  ✔ Done!

2. Fill in the product seed

Open agent_docs/product-seed.md and fill it in:

# Product Seed — PerishNote

## Problem Statement
Shared checklists built on-the-fly have no lightweight, zero-auth home.
Existing tools require accounts, persist forever, and feel too heavy
for transient coordination like a grocery run or a packing list.

## Target User
Anyone who needs to share a quick checklist with a specific person —
no technical setup, no account, no friction.

## Core Value Proposition
A URL-shareable, password-protected checklist that auto-deletes
after 24 hours. Zero signup required.

## Key Features
- Create a list, set a password, get a unique shareable URL
- Checkbox-style items with text labels
- Real-time sync for anyone with the URL + password
- Auto-deletion at 24h from creation

## Tech Preferences
Next.js, Redis for TTL-based auto-expiry, deploy to Vercel

## Constraints
- Must work on mobile browser
- No accounts, no email, no cookies

## Out of Scope
- User history or saved lists
- File attachments
- Notifications

3. Run the orchestrator

Open VSCode with the project. Open Claude Code. Type:

Run the orchestrator agent.

Phase 1 — four planning agents run in parallel (you watch, you don't type):

⠸ system-design-agent    writing agent_docs/system-design.md...
⠸ milestone-agent        writing agent_docs/milestones.md...
⠸ user-stories-agent     writing agent_docs/user-stories.md...
⠸ product-brief-agent    writing agent_docs/product-brief.md...

✓ All 4 planning docs complete.

Phase 2 — review:

⠸ review-agent    cross-checking all 4 docs...

Verdict: APPROVED

═══════════════════════════════════════
✅ PLANNING COMPLETE — GATE 1
5 docs approved. Ready to scaffold.
Type: proceed with scaffold
═══════════════════════════════════════

4. Approve Gate 1

proceed with scaffold

⠸ architect-agent    scaffolding repo...
  → next-app initialised
  → packages installed (redis, nanoid, zod, bcrypt)
  → .env.example written
  → CLAUDE.md test commands updated
  → dev server: OK ✓

═══════════════════════════════════════
✅ SCAFFOLD COMPLETE — GATE 2
Type: proceed with milestone 1
═══════════════════════════════════════

5. Approve Gate 2

proceed with milestone 1

Story STORY-101: Create a list and get a shareable URL
  ✓ PASS → committed: [STORY-101] Create list endpoint

Story STORY-102: Access a list with password
  ✗ NEEDS FIX (cycle 1/3): bcrypt compare called before await
  ✓ PASS → committed: [STORY-102] Password-protected list access

Story STORY-103: Real-time checkbox sync
  ✓ PASS → committed: [STORY-103] Real-time sync via SSE

⠸ Opening PR for Milestone 1...
  → gh pr create #4 — "Milestone 1: Core list creation + access"

⠸ pr-review-agent    polling for bot comments...
  → CodeRabbit reviewed (14s)
  → 2 actionable comments found

  Fixing:
    ✓ lib/list.ts:42   [CodeRabbit] missing null check on expiry — fixed in a3f1c2
    ✓ lib/auth.ts:17   [CodeRabbit] bcrypt rounds hardcoded — moved to env var — fixed in a3f1c2
  Replied + resolved both threads.

  Tests: PASS  |  Type-check: clean

═══════════════════════════════════════
✅ MILESTONE 1 COMPLETE — GATE 3
3 stories completed, PR reviewed and clean.
Type: proceed with milestone 2
═══════════════════════════════════════

What you typed across the entire build

Run the orchestrator agent.
proceed with scaffold
proceed with milestone 1
proceed with milestone 2

The PR Review Agent

After each milestone, the orchestrator opens a PR and spawns pr-review-agent automatically. It:

• Polls for comments from Entelligence, CodeRabbit, Codex, or human reviewers
• Fixes only actionable issues — skips praise, walkthroughs, and outdated threads
• Replies to each fixed thread with the commit SHA: "Fixed in a3f1c2 — moved bcrypt rounds to env var"
• Resolves the conversation thread via GitHub GraphQL API
• Verifies tests and type-check pass before pushing
• Runs up to 3 cycles (each push triggers a re-review from the bots)

If a fix can't be resolved after 3 cycles, it writes to agent_docs/build/blocked.md and waits for you.

Requires: gh CLI authenticated + at least one PR review bot configured on the repo. Manual trigger: /fix-pr-review or Run the pr-review-agent.

Package Structure

create-01x-project/
├── bin/
│   └── create.js                  ← CLI entry point
├── templates/
│   └── .claude/
│       ├── agents/                ← all 12 agent .md files
│       └── commands/
│           └── fix-pr-review.md   ← manual PR review trigger
├── package.json
└── README.md

No lib/ folder. No stack definitions. No generators. The tool has one job — copy the agents and commands, create the folder structure, and write CLAUDE.md, README.md, and product-seed.md. Everything stack-specific is handled by the agents themselves at runtime.

The Full Circle

Ideate  → ChatGPT GPT   ──┐
          Claude Skill  ──┼──▶  product-seed.md  ──▶  npx create-01x-project  ──▶  ship
          Any AI Prompt ──┘

You think through your idea in whichever AI you prefer. When it's ready, the seed goes into agent_docs/product-seed.md. The build system takes it from there — five planning agents, a review gate, scaffold, and a story-by-story build loop that commits as it goes.

→ product-seed-skill repo — all three seed generation formats in one place.

About 01x

This tool is part of 01x — a paid builder environment where you join a cohort, explore an idea, use AI to accelerate, and ship an MVP. If you want to build real products (not just learn concepts) and are serious about using AI as a force multiplier, check it out. Applications are reviewed manually and cohort sizes are kept small by design.

From zero to one to scale.