Spec Kit
A local CLI turns requirements into plans and tasks before an AI coding agent implements them.
What it is
Spec Kit is a spec-driven development toolkit that works alongside an AI coding agent. You first spell out project principles and functional requirements; it then works the requirements up step by step into a technical plan and executable tasks, and finally hands them to the agent to code, checking whether the implementation matches the agreements made earlier.
A typical run is not a matter of typing one prompt and waiting for code generation, but of completing specification, plan, tasks, and implementation in order. Compared with the usual one-shot prompt, this approach adds inspectable, editable intermediate artifacts: when requirements change or the implementation drifts, you can go back and adjust the corresponding stage instead of repeatedly correcting things with follow-up prompts.
How it runs
A project is initialized through the Specify CLI. Before running it you need to install uv and prepare a supported AI coding agent. The CLI can be installed from PyPI:
uv tool install specify-cli
specify init my-project --integration copilot
cd my-projectAfter initialization, the commands or skills used by the corresponding agent are written into the project. Then you run the specification, plan, tasks, and implementation commands in order inside the agent. If the current environment has no interactive terminal, add --non-interactive after specify init.
Core mechanisms
- Project constitution: long-term principles such as code quality, testing, experience, and performance are recorded first, and the specifications, plans, tasks, and implementations generated afterwards must comply with these constraints.
- Staged artifacts: use
/speckit.specifyto organize requirements, then/speckit.planto draw up the technical approach, then/speckit.tasksto break it into tasks, and finally run/speckit.implementto execute. - Convergence check:
/speckit.convergechecks the code against the specification, plan, and tasks, and appends unfinished work back to the task list for later handling. - Layered customization: project overrides, Presets, Extensions, and core templates are resolved by priority. Bundles can combine extensions, presets, and workflows into a versioned team configuration.
Who it is for
It suits teams that want an AI agent to develop new features according to explicit requirements, and also developers who are reworking an existing codebase and need to keep specifications and implementation corresponding to each other. If an organization wants to standardize requirement formats, technical plans, and the way tasks are broken down, Presets and Bundles can lock those conventions in.
Boundaries
Spec Kit does not replace a coding agent; implementation quality still depends on the chosen model, the agent integration, and the input specifications. It runs through a local CLI, and installation depends on uv. Community extensions are maintained by their respective authors, and the repository recommends reviewing the source before installing. The project uses the MIT license.