Spec Kit: AI Development Toolkit for Coding Agents

Understanding Spec-Driven Development with Spec Kit

Spec Kit is GitHub's implementation of Spec-Driven Development, where a written specification, not the code, is the source of truth, and the implementation gets regenerated from it instead of hand-maintained after the fact. It installs the Specify CLI into a project and gives your AI coding agent a fixed sequence of slash commands, from constitution through implement, that turn a spec into working code step by step.

Shifting from Code-First to Spec-First Development

Most teams treat a specification as a document you write once, get sign-off on, and then quietly ignore once coding starts, so the code becomes the real truth and the spec goes stale. Spec Kit's premise is that this order is backwards. Instead of a spec that only guides a human, it treats the spec as the thing that actually gets executed, with an AI coding agent regenerating the implementation from it rather than a developer translating it once and moving on. That distinction matters most when a project's requirements or tech stack shift, since the spec gets updated and the code regenerated instead of drifting out of sync with a document nobody rereads.

Core Commands and Extensibility

• ✓Seven core slash commands drive the workflow: /speckit.constitution, /speckit.specify, /speckit.plan, /speckit.tasks, /speckit.taskstoissues, /speckit.implement, and /speckit.converge, each mapped to an agent skill of the same name.
• ✓Three optional commands add quality gates. /speckit.clarify flags underspecified areas before planning, /speckit.analyze checks cross-artifact consistency after tasks are generated, and /speckit.checklist generates checklists the README calls 'unit tests for English.'
• ✓Extensions add new commands and templates for things the core workflow doesn't cover, installed with specify extension add and searchable with specify extension search.
• ✓Presets change how existing commands behave (spec format, terminology, organizational standards) without adding new capabilities, installed with specify preset add.
• ✓Bundles package a role-based set of extensions, presets, and workflows behind one bundle.yml manifest, installable in one step with specify bundle install.
• ✓Project-local overrides in .specify/templates/overrides/ sit above presets and extensions in priority, so a single project can patch a template without building a full preset.
• ✓Per the README, Spec Kit integrates with 30+ AI coding agents, including CLI tools and IDE-based assistants; specify integration list shows what your installed version supports.

Getting Started with Spec Kit

Spec Kit needs uv and Python 3.11+ before you touch the CLI. Install the Specify CLI with uv tool install specify-cli --from git+https://github.com/github/[email protected], replacing vX.Y.Z with a real release tag from the repo's Releases page (keep the leading v). It's also published on PyPI, so uv tool install specify-cli works too. Once installed, specify init my-project --integration copilot scaffolds a new project wired to the AI coding agent you choose. Upgrades run through specify self check (read-only), specify self upgrade --dry-run (preview only), and specify self upgrade (upgrades in place, detecting whether you used uv tool or pipx); pin a specific release with specify self upgrade --tag vX.Y.Z[suffix]. SPECIFY_UPGRADE_TIMEOUT_SECS caps how long that upgrade subprocess is allowed to run.

Spec-Driven Development Workflow

The workflow runs in a fixed order once your AI coding agent is open in the project directory. /speckit.constitution comes first and sets the project's governing principles, things like code quality, testing standards, UX consistency, and performance requirements, that later steps are meant to respect. /speckit.specify then takes a plain-language description of the feature you want, focused on intent and rationale; implementation detail comes later. /speckit.plan adds the technical side: framework, libraries, architecture choices. /speckit.tasks turns that plan into a concrete task list, and /speckit.implement executes it, building the feature out. /speckit.converge circles back after the fact, comparing the codebase to spec/plan/tasks and adding any leftover work as fresh tasks. Most agents expose these as /speckit.* slash commands; some integrations use a $speckit-* prefix instead when running in skills mode, per the README.

Approaches to AI-Assisted Software Development

Frequently Asked Questions

Best use cases

• •Greenfield builds where you'd rather write a specification once and regenerate the implementation than hand-maintain code an AI agent wrote from a one-off prompt.
• •Teams running the same coding agent across multiple projects who want a shared constitution, specify, plan, tasks, implement sequence instead of everyone prompting differently.
• •Brownfield modernization work, using the 'Iterative Enhancement' phase and the Evolving Specs guide to keep spec updates separate from routine tooling upgrades.
• •Organizations that need spec or plan templates to carry compliance language, via a preset, without hand-editing the core templates on every project.
• •Exploring the same feature built multiple ways, since the README lists 'Creative Exploration' (parallel implementations across different tech stacks) as one of its three development phases.

Who should try it — and who should skip

Try Spec Kit if you're already using an AI coding agent day to day and want a repeatable sequence (constitution, specify, plan, tasks, implement) instead of ad hoc prompting, especially if your team needs presets or extensions to enforce its own standards. Skip it if you just want quick inline code suggestions in your editor: Spec Kit adds a CLI, a template stack, and a multi-step command sequence that's more overhead than small, one-off changes need.

Related repositories