Spec Kit is GitHub's free open source tool for spec-driven development. It does not make the agent smarter. It makes the agent know what to build, in what order and how the result will be checked. That is really the whole difference between code that looks right and code that is right.
What Spec Kit does
Spec Kit adds templates, scripts and ready-made commands to your project. The commands run in the chat with your agent and work with Claude Code, GitHub Copilot, Cursor and several others. Each feature gets its own folder with the spec, the plan and the tasks as plain markdown files.
That may sound like a detail, but it is the most important part. The requirements live in the repo next to the code and are reviewed in the same pull request. If someone wants to know in six months why the code looks the way it does, they can read it.
What you need
You need Git, Python 3.11 or later, the uv package manager and an AI agent. If you already have the first three, installation takes a couple of minutes.
Install in five minutes
First install uv following the instructions at docs.astral.sh/uv. Then install Spec Kit with uv tool install specify-cli and create a project with specify init my-project --integration claude.
If you use another agent, replace claude with copilot for example. The command adds everything Spec Kit needs to the project. You install in the terminal and write everything else in the chat with the agent.
Six commands from idea to code
The workflow is six commands. Review the result after each step before moving on, because a mistake that follows you into the next step gets more expensive with every step.
/speckit-constitutionwrites the project's rules, once per project./speckit-specifyturns your description into a spec with user stories and acceptance criteria./speckit-clarifyfinds what is unclear and asks questions before anything is built./speckit-planmakes the technical decisions and checks them against the rules./speckit-tasksbreaks the plan into small steps with the test before the code./speckit-implementlets the agent build step by step.
The most important minute in the whole workflow is when you read the spec. If a requirement is missing there, it will be missing in the code too.
How we use it
Activator.se was rebuilt from Umbraco to Payload and Next.js with Spec Kit. The repo holds just over 60 specs, one for each feature. On top of that there are more than 2,000 automated tests. Every change goes from spec to tests to two reviewing agents before a person approves the deployment.
The interesting part is what happens to the questions. They come up in the spec, where they cost a sentence to answer, instead of as bugs in production.
Three tips for your first week
Start with a small feature with clear requirements, not with the most critical system. Keep the constitution short, five principles is enough. When a requirement changes, change the spec and not just the chat. Otherwise the code and the requirements drift apart.
If you want the whole workflow with examples and ready-made files to copy, it is in the guide Build it right with AI agents.
