Spec Kit

GitHub Spec Kit — Spec-Driven Development

"Stop vibe coding. Define intent first. Let AI execute with precision." Repo: https://github.com/github/spec-kit

When Claude Should Use This Skill

Trigger automatically when user:

• Starts a new project and wants AI to help build it
• Says "let's build X" without any existing spec
• Asks to "plan before coding" or "structure this project"
• Shares a GitHub URL to github/spec-kit
• Mentions "constitution", "spec.md", "speckit", "/speckit.*"

Installation (One-Time)

# Install CLI permanently
uv tool install specify-cli --from git+https://github.com/github/spec-kit.git

# Upgrade
uv tool install specify-cli --force --from git+https://github.com/github/spec-kit.git

# Verify tools
specify check

Quickstart for Claude Code

# New project
specify init my-project --ai claude

# Existing project (init in-place)
cd my-project
specify init . --ai claude --here

# New feature on existing project
bash .specify/scripts/create-new-feature.sh my-feature-name

8-Command Workflow (Always in this order)

1. /speckit.constitution   → Set immutable principles & tech stack
2. /speckit.specify        → Define WHAT to build (user stories, no tech)
3. /speckit.clarify        → Resolve any ambiguities
4. /speckit.plan           → Define HOW (architecture, components, dependencies)
5. /speckit.analyze        → Validate consistency across all artifacts
6. /speckit.tasks          → Break into ordered, parallelizable tasks
7. /speckit.implement      → Execute tasks (test-driven)
8. /speckit.checklist      → Quality validation

Project File Structure

.specify/
├── memory/
│   └── constitution.md          ← IMMUTABLE: tech stack, principles, conventions
├── specs/
│   └── <FEATURE_ID>/
│       ├── spec.md              ← WHAT & WHY (no implementation detail)
│       ├── plan.md              ← HOW (architecture, components)
│       ├── tasks.md             ← ordered tasks with [P] parallel markers
│       ├── data-model.md        ← entities & schema
│       ├── research.md          ← pre-implementation research
│       └── contracts/
│           ├── api-spec.json    ← REST/OpenAPI contracts
│           └── signalr-spec.md  ← realtime contracts
└── scripts/
    ├── create-new-feature.sh
    ├── setup-plan.sh
    └── update-claude-md.sh

File Templates

constitution.md

# Project Constitution

## Non-Negotiable Principles
- No class components — functional only
- All async operations must have error handling
- Tests required for all business logic

## Tech Stack
- Framework: React + Vite + TypeScript
- Styling: Tailwind CSS
- State: useState / Context API (no Redux)
- HTTP: Axios
- Testing: Vitest + Testing Library

## Code Conventions
- Files: PascalCase for components, camelCase for utils
- Max function length: 50 lines
- No magic numbers — use named constants

spec.md

# Feature: <NAME>

## Problem Statement
<What problem does this solve? For whom?>

## User Stories (P1 = must-have, P2 = should-have, P3 = nice-to-have)
- [P1] As a <user>, I want to <action> so that <value>
  - Independently testable: ✓
  - MVP slice: ✓

## Functional Requirements
- [] REQ-01: <specific capability>
- [] REQ-02: <specific capability> ← NEEDS CLARIFICATION: <question>

## Success Criteria (measurable, tech-agnostic)
- [] <metric>: <target value>

plan.md

# Implementation Plan: <FEATURE>

## Technical Decisions
| Dimension      | Decision        | Status              |
|---------------|-----------------|---------------------|
| Language       | TypeScript      | ✓ confirmed         |
| Framework      | React + Vite    | ✓ confirmed         |
| Storage        | PostgreSQL      | NEEDS CLARIFICATION |
| Performance    | <200ms p95      | ✓ confirmed         |

## Constitution Check
- [] Follows naming conventions
- [] Uses only approved dependencies
- [] Passes security review

## Component Architecture
<file tree with component responsibilities>

tasks.md

# Tasks: <FEATURE>

## Phase 1 — Init
- [] T001 Scaffold project

## Phase 2 — Foundation (BLOCKS all subsequent)
- [] T002 Set up DB schema
- [] T003 [P] Auth middleware     ← [P] = parallelizable
- [] T004 [P] API routes

## Phase 3 — P1 Story: <name>
- [] T010 [P1] Build <Component> — src/components/Foo.tsx
- [] T011 [P1] Tests — src/tests/Foo.test.ts

## Strategy
- [] Commit after each task
- [] Validate P1 before starting P2

Claude Code Execution Flow

When user says "let's build X", Claude should:

1. Run: specify init . --ai claude --here
2. Open .specify/memory/constitution.md → ask user to confirm/edit tech stack
3. Run /speckit.specify → fill spec.md with user's requirements
4. Run /speckit.clarify → ask 3-5 clarifying questions
5. Run /speckit.plan → generate plan.md
6. Run /speckit.analyze → check for gaps
7. Run /speckit.tasks → generate tasks.md
8. Run /speckit.implement → start building task by task
9. Commit after each completed task

Key Principles

Supported AI Agents (24+)

claude · gemini · copilot · cursor-agent · windsurf · kiro-cli · codex qwen · roo · kilo · opencode · amp · shai · tabnine · kimi ibm-bob · jules · codebuddy · vibe · auggie · antigravity · trae · qoder · generic