Spec Kit
GitHub's official open-source spec-driven development toolkit, connecting requirements, clarification, planning, tasks, implementation, and convergence checks through reviewable prompt commands.
Project overview
Spec Kit is GitHub's official open-source toolkit for spec-driven development. It turns requirements, clarification, technical planning, task decomposition, consistency analysis, implementation, and convergence checks into project-local prompt commands, while Specify CLI installs the templates for different AI coding agents. It is not an automated proof that software is correct, and it does not narrow the host agent's file, command, or network permissions. It works best as a reviewable development-process scaffold with human judgment retained at every quality gate.
Core capabilities
Requirements and clarification
/speckit.specify creates or updates a feature specification from natural language, while /speckit.clarify asks up to five targeted questions and writes the answers back into the spec.
Planning and task decomposition
/speckit.plan turns technology and constraints into design artifacts, and /speckit.tasks generates dependency-ordered work organized by user story.
Consistency analysis and implementation
/speckit.analyze performs a read-only cross-check of spec, plan, and tasks; /speckit.implement executes task phases and treats checklist state as a gate before proceeding.
Convergence loop
After implementation, /speckit.converge compares the codebase with specification artifacts and appends tasks only when it finds remaining gaps, supporting repeated implement-and-check cycles.
Multi-agent integrations
Specify CLI generates files for command-based and skill-based hosts and offers a generic integration for unlisted agents with a custom commands directory.
Access and usage
Install v1.0.4 into uv's isolated tool environment, then initialize a new test directory with an explicit installed agent integration. Review generated files before applying it to a real repository.
Let an AI Agent install it
Send this prompt to Codex, Claude Code, or another AI agent that can work with your local environment.
Help me safely install and evaluate GitHub Spec Kit v1.0.4.
Project: https://github.com/github/spec-kit
First read README.md, docs/installation.md, docs/reference/core.md, docs/reference/integrations.md, SECURITY.md, and LICENSE. Confirm the Python version, uv, the target agent integration key, target directory, and version-control state. Use the pinned command uv tool install specify-cli --from git+https://github.com/github/spec-kit.git@v1.0.4; do not track main. Run specify version, then initialize only a new test directory. Do not use --force in an existing non-empty directory. List the .specify, specs, and agent-specific files that will be created, and explain that prompts inherit the host agent's file, command, and network permissions. Do not configure credentials, install community extensions, execute implementation tasks, or overwrite existing files; pause for confirmation first. Report commands run, file changes, verification results, and unverified items.- Python 3.11 or newer
- uv (recommended) or pipx
- An installed and callable supported AI coding agent
- Network access to GitHub
- Write access to the target project and recoverable version-control state
uv tool install specify-cli --from git+https://github.com/github/spec-kit.git@v1.0.4- 1Pin a release and verify prerequisites
Confirm Python 3.11+, uv, and the target agent, and choose the v1.0.4 release instead of deploying the continuously changing main branch.
- 2Initialize a new directory
Run specify init my-project --integration <key>, inspect .specify, specs, and host command or skill files, and avoid --force in existing repositories.
- 3Establish principles and a specification
Use /speckit.constitution for non-negotiable engineering principles, then /speckit.specify for goals, user stories, and acceptance boundaries.
- 4Resolve ambiguity before planning
Run /speckit.clarify and /speckit.checklist for consequential work, and have a person review questions and checklist decisions instead of letting the implementation agent self-approve.
- 5Plan, decompose, and analyze
Run /speckit.plan, /speckit.tasks, and read-only /speckit.analyze. When conflicts appear, edit the upstream artifact that owns the issue.
- 6Implement in stages and converge
Scope /speckit.implement to manageable phases, run the project's own tests and review, then use /speckit.converge to find gaps. Repetition does not replace final acceptance.
Run specify version first, then execute specify init my-project --integration <key> in a new directory. Confirm that .specify/, specs/, and the agent-specific directory were created. Run only /speckit.specify on a low-risk example and review the file changes manually; this does not verify later implementation quality.
- README also documents PyPI, pipx, and one-shot uvx methods; pinned release tags are preferable for team deployments
- Use --non-interactive with an explicit --integration or --ignore-agent-tools in CI and non-PTY environments
- Git is optional for core initialization but is needed by the git extension and normal branch-based workflows
- Inspect local changes before upgrading or switching integrations; --force may overwrite managed files
Use cases
Delivering a new feature
Turn an ambiguous product idea into user stories, clarification answers, a technical plan, and executable tasks before an agent jumps directly into code.
Team engineering policies
Use the constitution and presets to carry testing, security, architecture, or compliance principles into each feature's planning and review path.
Moving workflows across agents
Reuse core specification artifacts across supported hosts while the CLI generates each agent's command or skill layout.
Staged implementation for large work
Limit implement to selected task phases, validate each stage independently, and use analyze and converge to find cross-artifact gaps.
Assessment
Spec Kit's value is not a magical single prompt but a visible, reversible, and repeatedly checkable chain from requirements to implementation, with explicit integration support for many agents. The public repository has an MIT license, SECURITY policy, stable releases, tests, and active maintenance, and its installation docs clearly cover Python, uv, non-interactive use, and overwrite controls. The main risks sit at two layers: initialization and agent execution write project files, while third-party catalog content can introduce remote downloads. The artifacts are still model-generated, so checklists and convergence commands do not replace human acceptance. This catalog review verified source and documentation only; it did not claim installation or prompt execution.
Why it may be useful
- Requirements, planning, tasks, analysis, implementation, and convergence form a complete reviewable workflow
- More than 30 host integrations plus a generic adaptation path
- Stable releases, MIT licensing, coordinated disclosure, tests, and recent fixes are publicly verifiable
- Documentation covers unattended runs, overwrite behavior, extension provenance, authentication, and offline use
What to know first
- This review did not install the CLI or run core commands in any agent
- Host compatibility changes with each agent version and invocation model and needs target-environment verification
- --force, integration switching, and upgrades can overwrite managed files
- Community extensions, presets, and workflows require separate source and permission review
README
Spec Kit
Overview
GitHub's official open-source spec-driven development toolkit, connecting requirements, clarification, planning, tasks, implementation, and convergence checks through reviewable prompt commands. GitHub's official spec-driven development toolkit, providing requirements, planning, task, implementation, and convergence prompts for more than 30 AI coding agents.
Getting started
- Install v1.0.4 into uv's isolated tool environment, then initialize a new test directory with an explicit installed agent integration. Review generated files before applying it to a real repository.
- The official sequence begins with /speckit.constitution for project principles, then /speckit.specify for requirements, /speckit.plan for technical design, /speckit.tasks for dependency-ordered work, and /speckit.implement for execution. Clarify, checklist, and analyze are risk-sensitive quality gates, while converge checks for implementation gaps.
- The installation guide recommends uv and requires Python 3.11 or newer. A pinned install can use the GitHub v1.0.4 tag. Initialization should select a host explicitly with --integration; unattended runs should add --non-interactive, and --force should be limited to reviewed cases where merging or overwriting files is acceptable.
- Run specify version first, then execute specify init my-project --integration <key> in a new directory. Confirm that .specify/, specs/, and the agent-specific directory were created. Run only /speckit.specify on a low-risk example and review the file changes manually; this does not verify later implementation quality.
Configuration
uv tool install specify-cli --from git+https://github.com/github/spec-kit.git@v1.0.4