🧠 Why create-runway?

AI agents are powerful execution engines, but they suffer from four primary failure modes:

1. Silent Assumptions: Guessing requirements instead of asking.
2. Collateral Damage: Modifying adjacent, working code unnecessarily.
3. Context Bloat: Drowning in outdated context and hallucinating.
4. Poor Handovers: Losing the thread when you switch computers, developers, or AI tools.

create-runway solves this by generating an institutional memory architecture. Inspired by the public workflows of Boris Cherny (creator of Claude Code) and Andrej Karpathy, it enforces strict behavioral boundaries, verification loops, and session state tracking across your repository.

🚀 Getting Started

Bootstrap a brand new AI-ready project, or easily inject the AI intelligence into an existing repository.

npx create-runway@latest init

The interactive terminal wizard will ask you what you want to do. If you select an existing directory, create-runway intelligently skips project scaffolding and simply injects the AI rules instantly.

⚡️ Headless Initialization (For CI/CD)

# Skip prompts and use defaults
npx create-runway init --yes

# Bootstrap instantly with a community template
npx create-runway init --template frontend-react

# Enforce your company's AI standards via a preset JSON
npx create-runway init --preset ./team-preset.json

🛠 Features & CLI Commands

create-runway is more than a one-time generator. It is a persistent toolkit for managing your AI agents.

🔄 npx create-runway sync

Changed your framework or token strategy in .enter/config.json? Run sync to regenerate all AGENTS.md and .cursorrules files. It shows a precise diff of what will change before overwriting.

🤝 npx create-runway handover

Don't just close your laptop. Run handover at the end of your session. It uses an LLM to read your recent changes, tests, and file tree, then generates a comprehensive handover.md file. The next morning (or the next AI agent) will pick up exactly where you left off.

📊 npx create-runway audit

LLMs lose precision (hallucinate) when their context window gets stuffed with stale data. The audit command analyzes your rules files using @anthropic-ai/tokenizer and js-tiktoken to give you a live health report of your context files.

💡 Pro-tip: Run with --watch for a live dashboard while you prune your mistake cache!

➕ npx create-runway add <agent>

Teammate just downloaded Windsurf but you only scaffolded for Cursor? No problem.

npx create-runway add windsurf

Instantly drops the optimized .windsurfrules into your project without touching the rest of your setup.

🎓 npx create-runway learn

New to agentic coding? Run the interactive learn walkthrough directly in your terminal to understand how to leverage Plan Mode, Subagents, and the Mistake Cache.

📂 What gets generated?

When you initialize create-runway, you get the ultimate AI ruleset injected into your workspace:

| File | Purpose | |------|---------| | 🏛 AGENTS.md | The central source of truth for your architecture. It tells the AI what framework it is operating in and lists the core rules. | | 📜 instructions.md | Strict execution boundaries. Enforces surgical edits, verification loops (testing before saving), and forbids silent assumptions. | | 🧠 feedback.md | The Mistake Cache. When the AI messes up, log it here so it never repeats the same error. Ruthlessly prune this over time. | | 🤝 handover.md | Session tracking file updated dynamically by the handover command. | | 🤖 Agent Wrappers | Thin wrappers (ANTIGRAVITY.md, CLAUDE.md, .cursorrules, .windsurfrules) that point the specific AI tool toward the centralized AGENTS.md logic. | | ⚙️ .enter/config.json | The central configuration that powers the create-runway CLI. |

🔒 Enterprise-Grade Security

We take API Key security seriously. create-runway will never store API keys in your project directory (not even in .gitignore files). Instead, your AI provider keys (used for the handover command) are safely saved in a strict offline hardware enclave at ~/.runway/credentials.json with OS-level 0o600 permissions. Only you can read them.

🤝 Philosophy

"Treat AI coding as a parallel execution engine, not a chatbot. Front-load thinking, build persistent knowledge, automate inner loops, and give the AI a way to verify its own work."

Built with ❤️ for the future of compounding engineering.