Draft

The Mission

AI assistants need better workflows. We're building them together.

Draft is a community-driven framework that teaches AI assistants to ask questions, create specs, and verify implementations before jumping to code.

Help us define best practices for agentic coding.

The Workflow

Spec-driven development workflow encoded as reusable skills:

• /spec - Clarify requirements, check scope, create spec
• /implement - Assess blast radius, build as vertical slice
• /verify-spec - Verify implementation matches spec
• /refine - Refine existing spec at any point in time

Installation

Quick Install

curl -fsSL https://raw.githubusercontent.com/heiko-braun/draft/main/install.sh | bash

Installs to ~/.local/bin and configures PATH for bash and zsh.

Other Installation Methods

Using Go:

go install github.com/heiko-braun/draft/cmd/draft@latest

Manual Download:

Download pre-built binaries from releases:

• macOS (Intel): draft-darwin-amd64
• macOS (ARM): draft-darwin-arm64
• Linux: draft-linux-amd64
• Windows: draft-windows-amd64.exe

Quick Start

1. Initialize Your Project

draft init              # Claude Code & Cursor
draft init --agent claude   # Claude Code only
draft init --agent cursor   # Cursor only

2. Use the Skills

/spec Add user authentication
/implement authentication
/verify-spec authentication
/refine authentication

Review & Discuss

Review specs collaboratively with inline comments, threaded discussions, and approvals:

draft review

Opens a browser UI with rendered documents, inline text annotation, comment threads, and publish/sync workflow — all stored on a virtual git branch with no external dependencies.

Contribute

The workflows are defined in markdown files at .claude/commands/ and .cursor/skills/. Test them, shape the details, share learnings.

View on GitHub →