Spec Kit
Guide coding agents from requirements to implementation and verification with reusable, structured processes.
- Source repo
- github/spec-kit
- Stars
- ★ 142k
- Last updated
- 1d ago
- License
- MIT
- Primary language
- Python
- FA score
- 78/100 · Good
At a glance
- How it runs
- Works with
- Universal · cross-platform
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A development team wants requirements, a technical plan, and implementation tasks written down before building a feature; it can use SDD to produce those artifacts step by step.
- Not a fit if
- Teams that do not use a supported AI coding agent
- Users who need a fully graphical workflow
- Source review
- 78/100 · Good
What does this agent do, and when should you use it?
Spec Kit is an open source toolkit that gives coding agents structured processes, reusable templates, and documented outcomes through the Specify CLI. Its core process, Spec-Driven Development (SDD), moves from project principles to a feature specification, plan, and tasks, then through implementation and repeated convergence. Optional bug-fixing and idea-assessment extensions produce, respectively, a cause assessment with verification records and an evidence-backed go, clarify, or stop decision. Teams can adapt or assemble processes with extensions, presets, workflows, overlays, and bundles, and connect them to supported coding agents. Getting started requires Python 3.11+, uv, and a supported AI coding agent; the README names Linux, macOS, and Windows.
After installing the Specify CLI and initializing a project with specify init, users launch a coding agent in the project directory and run Spec Kit skills one at a time using that agent's command syntax. The SDD process produces a specification, plan, and tasks, then code that is iterated until /speckit-converge reports Converged. With the bug extension installed, /speckit-bug-assess, /speckit-bug-fix, and /speckit-bug-test record assessment, repair, and verification under .specify/bugs/. With the assess extension, /speckit-assess-intake, research, define, shape, and decide produce assessment materials under .specify/assessments/ and a go, needs-clarification, or kill decision. Extensions and presets adapt commands, templates, and methods; workflows automate prompts, commands, quality gates, and shell tasks.
- A development team wants requirements, a technical plan, and implementation tasks written down before building a feature; it can use SDD to produce those artifacts step by step.
- A maintainer has a reproducible bug and wants to document its cause, scoped fix, and verification; the bug extension provides that workflow and reports.
- A product team is unsure whether to invest in an idea; the assessment extension gathers evidence and records a go, needs-clarification, or kill decision.
- An engineering organization needs its standards reflected in development work; it can adapt templates, commands, or quality gates with presets, project overrides, or extensions.
- A team wants to use its own process; it can expose commands through an extension, automate checks with workflows, and share the setup in a bundle.
How do you install or deploy this agent?
You need Python 3.11+, uv, and a supported AI coding agent. The README lists Linux, macOS, and Windows. This example installs the CLI and initializes a project with GitHub Copilot:
uv tool install specify-cli
specify init my-project --integration copilot
cd my-projectBug fixing and idea assessment are optional extensions. Add them from the project directory when needed:
specify extension add bug
specify extension add assessHow do you use this agent?
Start your coding agent in the project directory, then run skills in its chat one at a time and review each result. The compact SDD flow is:
/speckit-constitution Create principles focused on code quality, testing, and maintainability.
/speckit-specify Build a photo organizer with albums grouped by date and a tile preview of each album.
/speckit-plan Use Vite with vanilla JavaScript. Keep images local and store metadata in SQLite.
/speckit-tasks
/speckit-implement
/speckit-convergeRepeat implement and converge for a feature until the result is Converged. Other coding agents use their own invocation syntax; bug-fixing and assessment start with commands from their respective extensions.
What are this agent's strengths and limitations?
- SDD, bug fixing, and idea assessment are independent entry points, so the latter two do not require an SDD feature workflow first.
- Extensions, presets, workflows, overlays, workflow steps, and bundles can be composed, from a one-off template adjustment to a shared team setup.
- SDD and bug fixing have explicit convergence or verification outcomes, making completion and symptom resolution easier to record.
- The documentation describes switching coding-agent integrations and lists Linux, macOS, and Windows support.
- The workflow requires Python 3.11+, uv, and a supported AI coding agent; the README does not state the agent service's cost.
- Users run commands and review results in a coding agent's chat, so this is not a standalone graphical application.
- Bug fixing and idea assessment require installing their extensions separately in a project.
- The README says community catalog listings are not audited or endorsed, so users should review community code before installing or running it.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Spec Kit This agent | 78 · Good | CLI | ★ 142k | 1d ago | Python | — |
| Smart Ralph | 73 · Some gaps | Agent plugin / skillFree + model costs | ★ 559 | 1mo ago | Shell | Codex · Claude Code |
| OSpec | 50 · Major gaps | CLIFree + model costs | ★ 396 | 1mo ago | JavaScript | Codex · Claude Code |
| Autoprompt Skill | 69 · Some gaps | CLIFree + model costs | ★ 1.2k | 16d ago | JavaScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The documentation says authentication is off by default, credentials are matched by host, and Authorization is stripped when a redirect leaves declared hosts; it recommends environment variables for tokens and owner-only access to auth.json. Initializing a non-empty directory requires explicit --force, and the guide advises review and backup first. HTTP requests, extension downloads, initialization scripts, and implementation tasks can still have effects; the supplied material does not fully show confirmation before every action, all data flows, or rollback mechanisms, so full marks are not justified. The repository provides a security disclosure channel and clear copyright attribution.
The README, installation guide, and quickstart consistently distinguish CLI setup from invoking skills in agent chat, and describe multiple processes and stages. pyproject.toml lists runtime dependencies and the Python minimum; the installation guide covers multiple channels, version pinning, and offline installation. The docs describe some error cases and upgrade recovery, but the material does not establish availability of every dependency, all failure paths, or complete consistency across the repository, so points are deducted.
The material clearly targets AI coding agent users and covers feature development, bug repair, and idea assessment. It supports multiple agents, script types, interactive and non-interactive setup, existing projects, and offline environments. The docs distinguish the CLI, agent skills, and optional extensions, but the supplied material does not fully enumerate capability boundaries or trigger rules, so this is not full marks.
The information architecture is clear, with installation, quickstart, reference, extension, and troubleshooting paths. Installation notes cover prerequisites, channels, integration choices, script types, upgrades, and verification. Command and directory names are generally stable in the docs, with step-by-step examples and some compatibility limitations. The MIT license is explicit; the release process explains tags, versions, changelog generation, and maintainer actions. Some limitations and migration details are distributed across documents, and the supplied files do not establish complete documentation consistency, so points are deducted.
The toolkit produces specifications, plans, tasks, and convergence reports, plus bug and idea-assessment workflows; it advises invoking and reviewing steps individually. Examples make common use concrete. The templates and processes show practical potential, but the supplied material provides no baseline comparison or cost data, so marginal value and cost-benefit are not scored at full marks.
Installation commands, prerequisites, command names, artifact paths, license, and release steps have concrete textual support; the README, installation guide, and quickstart corroborate the core usage path. The docs distinguish a local version check from proof of package origin and state that missing verification does not establish a successful fix. The supplied material includes only partial code and test excerpts, so repository-wide claims cannot be cross-checked here; fact and inference boundaries are comparatively clear.
- This is a low-confidence review of the supplied static material; the CLI, agent workflows, and tests were not executed.
- Initialization, extension downloads, and agent implementation may modify project files or access the network; review changes and back up before using --force on a non-empty directory.
- Authentication configuration can involve privileged tokens; the material recommends environment variables and restrictive permissions for auth.json.
FAQ
Do I have to adopt SDD to use Spec Kit?
Which coding agents does it support?
How do I know whether a bug fix succeeded?
.specify/bugs/<slug>/. Only the final verdict verified confirms a successful fix; partial and failed do not.What should I do if an assessment says needs-clarification?
.specify/assessments/<slug>/, rather than rerunning all stages.