GitHub - ambystechcom/AmbyKit: Spec-Driven Development framework that helps AI coding assistants build better software — author your specs, user stories, UI design, plans and tasks once, and emit native config for Claude Code, OpenCode, GitHub Copilot, Cursor, Antigravity and more.

Spec-Driven Development that specs your UI too. Most SDD frameworks take you from user stories straight to a technical plan and leave the interface to the model's imagination. AmbyKit adds a first-class design phase — ui.md plus a real design-tokens.json — between the what and the how, then emits native commands and rules for every AI coding assistant your team uses.

Quick start

npx @ambystech/ambykit init     # scaffold .amby/ and pick your assistants
# — or install it —
npm install -g @ambystech/ambykit
ambykit init

Node ≥ 20. No Python toolchain, no separate installer — it's an npm package.

Already have a project? init is non-destructive — an existing CLAUDE.md/AGENTS.md is preserved (and backed up); AmbyKit only adds/updates its own ### AmbyKit usage section. Re-run ambykit sync any time to keep it current, or ambykit restore to roll a file back.

Then, inside your AI assistant, walk the workflow:

/amby.constitution   # one-time: set your project's guiding principles
/amby.specify        # describe a feature → spec.md (user stories + EARS requirements)
/amby.clarify        # resolve open questions
/amby.revise         # continue an existing spec — add and refine in place
/amby.design         # UI spec + design-tokens.json
/amby.plan           # technical plan
/amby.tasks          # ordered, dependency-aware task list
/amby.implement      # build it
/amby.converge       # verify the code against the spec; append gap tasks, repeat until converged
/amby.review         # have another role (QA, Architect, PM…) critique an artifact, findings by ID

Track progress from the terminal:

ambykit dashboard
ambykit dashboard 001:US-3   # story ids restart per feature — qualify with the feature ref

📖 Full documentation: ambykit.ambystech.io

Why

AI coding assistants build better software when they start from good requirements. AmbyKit gives you a rigorous, tech-agnostic SDD workflow and a single source of truth that compiles to each tool's native format.

• UI is a first-class artifact. /amby.design produces a UI spec plus design tokens — the part every other spec framework skips. Your assistant builds the interface you described, not the one it guessed.
• WHAT before HOW. spec.md captures user stories + testable requirements with no tech decisions; plan.md captures the technical approach separately.
• Parallel work is built in. ambykit worktree gives each feature its own git worktree, so several features — or several assistants — run side by side without checkout switching.
• Author once, emit per tool. One neutral source → Claude Code, OpenCode, GitHub Copilot (VS Code + CLI), Cursor (+ CLI), Antigravity (IDE + CLI), Codex CLI.

How AmbyKit compares

The honest version. GitHub Spec-Kit is the big one in this category and it is genuinely good — if you want the widest assistant coverage and the largest community, use it. AmbyKit is the choice when the UI matters and you live in the Node ecosystem.

Why not just Spec-Kit? If breadth of assistant support is your constraint, use Spec-Kit — it covers far more tools. Pick AmbyKit if you want the design phase, per-feature worktrees, and an install that's one npx away instead of a Python toolchain.

Comparison current as of AmbyKit v1.1.0; competitors move fast — corrections welcome via issue or PR.

The workflow

Requirements use user stories (US-#) + EARS functional requirements (FR-###) + Given/When/Then acceptance criteria. Stories carry priority and depends-on/blocked-by so work can be ordered and blocked; ambykit dashboard reports progress across the story/task graph.

Supported tools

Claude Code (CLI + VS Code), GitHub Copilot (VS Code + CLI), OpenCode, Cursor (+ CLI), Antigravity (IDE + CLI), Codex CLI. See docs/tool-compatibility.md.

Missing yours? Open a request — a new target is a thin BaseEmitter subclass.

CLI

See docs/cli-reference.md.

Contributing

AmbyKit is built with AmbyKit — see CONTRIBUTING.md, AGENTS.md, and docs/contributing.md. Changes are logged in CHANGELOG.md. MIT licensed.