Spec Kit — Spec-Driven Development with Coding Agents
by GitHub
GitHub's method for making the spec, not the code, what your agent works from
Overview
Spec Kit ships as the specify CLI (installed via uv or PyPI) plus a documentation site organised into Getting Started, Reference, Concepts, Community and Development. The workflow is a set of slash commands your coding agent runs: /speckit.constitution establishes the project's governing principles, /speckit.specify turns a feature description into a numbered spec.md on its own branch, /speckit.clarify resolves anything marked [NEEDS CLARIFICATION], /speckit.plan produces plan.md alongside data-model.md, contracts/ and research.md, /speckit.tasks derives a parallelizable tasks.md, /speckit.implement builds it, /speckit.analyze cross-checks artifacts for consistency, and /speckit.converge assesses an existing codebase against its specs for brownfield adoption. The accompanying spec-driven.md is a genuine methodology essay arguing that specifications, not code, should be the source of truth, and it encodes nine constitutional articles — library-first features, CLI-exposed interfaces, tests written and confirmed failing before implementation, a three-project maximum, no speculative abstraction, and integration-first testing against real services. The repository is MIT-licensed with 125k stars, 1,681 commits and support for 35+ coding agents, plus 138 community extensions, 25 presets and alternative processes (AIDE, Canon, Product Forge, MAQA). Enterprise air-gapped and offline installs are documented. Practitioners are not uncritical: Birgitta Bockeler's independent evaluation on martinfowler.com found it the most customizable of the SDD tools but verbose, and noted agents still ignored parts of the constitution.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Guide
- Cost
- Free
- Duration
- ~1 hour to read the SDD methodology; ~2-3 hours for a first end-to-end feature run
- Provider
- GitHub
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Run the full spec-driven loop: constitution, specify, clarify, plan, tasks, implement, analyze
- ✓Write a project constitution that encodes non-negotiable engineering principles for every generation
- ✓Turn a feature description into a structured spec.md with explicit [NEEDS CLARIFICATION] markers
- ✓Generate plan.md, data-model.md, contracts/ and research.md from a spec before any code is written
- ✓Decompose an implementation plan into parallelizable tasks a coding agent can actually execute
- ✓Apply the same workflow across 35+ coding agents including Copilot, Claude Code, Cursor and Gemini CLI
- ✓Assess an existing codebase against its specifications with /speckit.converge for brownfield work
- ✓Install and pin the specify CLI via uv or PyPI, including air-gapped enterprise setups
Highlights
- •125k GitHub stars and MIT-licensed — the reference implementation of spec-driven development, not a vendor pitch
- •spec-driven.md is a real methodology essay, arguing specs are the source of truth and code is the generated artifact
- •Agent-agnostic by design: 35+ CLI and IDE assistants supported, so the method outlives whichever tool you use today
- •138 community extensions and 25 presets, plus alternative processes (AIDE, Canon, Product Forge, MAQA) for teams whose workflow differs
- •Honest external assessment exists: martinfowler.com found it the most customizable SDD tool but criticised markdown verbosity and agent non-compliance
Who It’s For
Best For
- ✓Tech leads standardising how their team drives AI coding agents
- ✓Developers whose agent output is too large or too unreviewable to trust
- ✓Teams retrofitting specifications onto an existing brownfield codebase
- ✓Engineers comparing spec-driven approaches such as Kiro, Tessl and Spec Kit
Prerequisites
- •An AI coding agent already installed — Copilot, Claude Code, Cursor, Gemini CLI or similar
- •Comfort with Git branches and a terminal, since each spec creates its own branch
- •uv or Python available to install the specify CLI
FAQ
What is Spec Kit — Spec-Driven Development with Coding Agents?
GitHub's official toolkit and methodology for Spec-Driven Development, the practice of writing an executable specification before letting a coding agent generate anything. It is written for developers and tech leads who already use Copilot, Claude Code, Cursor or Gemini CLI and find that ad-hoc prompting produces unreviewable output. After working through it you can run a repeatable constitution to spec to plan to tasks to implement loop on both greenfield and existing codebases.
Is Spec Kit — Spec-Driven Development with Coding Agents free?
Spec Kit — Spec-Driven Development with Coding Agents is free to access.
What level is Spec Kit — Spec-Driven Development with Coding Agents for?
Spec Kit — Spec-Driven Development with Coding Agents is aimed at a intermediate audience. Recommended background: An AI coding agent already installed — Copilot, Claude Code, Cursor, Gemini CLI or similar, Comfort with Git branches and a terminal, since each spec creates its own branch, uv or Python available to install the specify CLI.
How long does Spec Kit — Spec-Driven Development with Coding Agents take?
Expect roughly ~1 hour to read the SDD methodology; ~2-3 hours for a first end-to-end feature run. Most learners work through it at their own pace.
What will I learn from Spec Kit — Spec-Driven Development with Coding Agents?
You'll learn: Run the full spec-driven loop: constitution, specify, clarify, plan, tasks, implement, analyze; Write a project constitution that encodes non-negotiable engineering principles for every generation; Turn a feature description into a structured spec.md with explicit [NEEDS CLARIFICATION] markers; Generate plan.md, data-model.md, contracts/ and research.md from a spec before any code is written; Decompose an implementation plan into parallelizable tasks a coding agent can actually execute; Apply the same workflow across 35+ coding agents including Copilot, Claude Code, Cursor and Gemini CLI; Assess an existing codebase against its specifications with /speckit.converge for brownfield work; Install and pin the specify CLI via uv or PyPI, including air-gapped enterprise setups.
Topics
Sources
This page was written from 4 sources, 3 on domains other than github.github.com.