A workflow specification for AI coding assistants. Define how AI should approach development — from requirements to implementation — through structured phases and clear deliverables.

→ workflow not tooling
→ iterative not waterfall
→ discuss then implement
→ built for brownfield not just greenfield

Smooth is a workflow standard, not a framework. It defines a set of phases (product → technical → tasks → apply → archive) that guide AI assistants through structured development. Each phase has a clear stance, deliverable, and set of guardrails.

After initialization, your AI assistant gets slash commands that enforce this workflow. No runtime dependencies, no CLI needed after init — just markdown templates that shape how AI thinks and works.

You: /smooth:product add-dark-mode
AI:  Let's think about what dark mode means for this project...
     [discussion happens, product.md gets written incrementally]
     [open questions listed for technical validation]

You: /smooth:technical
AI:  Let me resolve the open questions first...
     [reads code, validates feasibility, updates product.md]
     Now let's design the architecture...
     [designs solution, writes technical.md with diagrams]

You: /smooth:tasks
AI:  Let's break this into implementable chunks...
     [creates ordered task list in tasks.md]

You: /smooth:apply
AI:  Implementing task 1/5: Add theme context provider...
     ✓ Task complete
     Implementing task 2/5: Create toggle component...
     ✓ Task complete
     ...

You: /smooth:archive
AI:  Archived to smooth/archive/2025-01-23-add-dark-mode/

npx @pureforge/smooth init

Now use these slash commands in your AI coding assistant:

Each phase is a stance with a deliverable — you're having a conversation with your AI assistant while it simultaneously captures decisions into structured documents.

• Product → Define what to build, list open questions (product.md)
• Technical → Resolve open questions via code investigation, then design architecture (technical.md)
• Tasks → Break into implementable chunks (tasks.md)
• Apply → Implement task by task
• Verify → Validate against acceptance criteria (verify.md)
• Archive → Finalize and archive

You don't have to follow the phases linearly. The AI suggests the next step based on change complexity — simple changes can skip technical design, internal refactors can skip verification. The workflow is fluid.

After smooth init, your project gets:

your-project/
├── .claude/
│   ├── commands/smooth/     # Slash commands
│   │   ├── product.md
│   │   ├── technical.md
│   │   ├── tasks.md
│   │   ├── apply.md
│   │   ├── verify.md
│   │   └── archive.md
│   └── skills/              # Agent skills
│       ├── smooth-product/SKILL.md
│       ├── smooth-technical/SKILL.md
│       ├── smooth-tasks/SKILL.md
│       ├── smooth-apply/SKILL.md
│       ├── smooth-verify/SKILL.md
│       └── smooth-archive/SKILL.md
└── smooth/                  # Change artifacts live here
    ├── <change-name>/
    │   ├── product.md
    │   ├── technical.md
    │   ├── tasks.md
    │   └── verify.md
    └── archive/             # Completed changes

vs. Kiro specs — Kiro couples specs to its IDE. Smooth is tool-agnostic — works with any AI assistant that supports slash commands or skills.

vs. nothing — AI coding without workflow structure means vague prompts and unpredictable results. Smooth brings process without ceremony.

MIT