Skip to content

Getting Started

Prerequisites

ToolRequiredPurpose
Python 3.11+YesRuntime for the CLI
uvYesPython package manager
GitYesVersion control
Claude CodeRecommendedDefault AI assistant

Other AI agents are also supported.

Installation

Install once, use everywhere:

bash
uv tool install maistik-specify-cli --from git+ssh://git@gitlab.com/maistik-studio/tools/spec-kit.git

This adds specify to your PATH. Manage it with standard uv tool commands:

bash
uv tool list        # See installed tools
uv tool uninstall maistik-specify-cli  # Remove

One-Time Usage

Run without installing:

bash
uvx --from git+ssh://git@gitlab.com/maistik-studio/tools/spec-kit.git specify init my-project --ai claude

Upgrade

bash
uv tool install maistik-specify-cli --force --from git+ssh://git@gitlab.com/maistik-studio/tools/spec-kit.git

Verify

bash
specify check

This checks for installed tools (Git, Claude Code, etc.) and reports any missing dependencies.

Initialize a Project

New Project

bash
specify init my-project --ai claude

This creates a my-project/ directory with the full Spec Kit structure.

Existing Project

bash
cd my-existing-project
specify init . --ai claude
# or
specify init --here --ai claude

Use --force to skip confirmation when the directory is not empty:

bash
specify init . --force --ai claude

Self-Hosted GitLab

If you use a self-hosted GitLab instance:

bash
# Set your GitLab token
export GL_TOKEN=glpat-xxxxxxxxxxxx

# Point to your instance
export SPECKIT_GITLAB_HOST=gitlab.mycompany.com
export SPECKIT_GITLAB_PROJECT_ID=group%2Fspec-kit

specify init my-project --ai claude

Your First Spec

Open the project in Claude Code and run the commands in order:

1. Set principles

text
/speckit.constitution Code quality, testing standards, performance requirements

This creates the project constitution — the governing principles that guide all decisions.

2. Write a specification

text
/speckit.specify Build a photo album app with drag-and-drop, tile previews, and date grouping

Describe what and why, not how. The AI creates a structured spec with requirements, user stories, and acceptance criteria.

3. Create the plan

text
/speckit.plan Use Vite with vanilla JS, SQLite for local storage

Now specify your tech stack. The AI generates an implementation plan with architecture, data model, and technical details.

4. Break into tasks

text
/speckit.tasks

Generates an ordered, dependency-aware task breakdown with parallel execution markers.

5. Build it

text
/speckit.implement

The AI executes all tasks in order, following the plan and respecting the constitution.

What's Next?

Released under the MIT License.