What it is
Spec Kit is GitHub's open-source take on spec-driven development. The specify CLI sets up a project, and slash commands in your agent walk through a constitution, a feature spec, a technical plan, a task list, implementation and a converge step that checks the result against the documents.
Bug fixing and idea assessment are optional extensions. Presets, workflows, bundles and catalogs let teams change templates, add quality gates or host a private catalog of approved components. It installs with uv tool install specify-cli, needs Python 3.11 or newer, and runs on Linux, macOS and Windows.
Our review
Spec Kit is the most-starred spec-driven workflow on GitHub, and its value depends entirely on whether someone reads the specs. The repo has about 142,500 stars and ships fast: v1.0.0 arrived on 21 August 2026, and eight releases landed between 2 and 22 October, ending at v1.2.3. The integrations page lists 41 named coding agents, from Claude Code and Copilot to Cursor, Codex CLI, Kiro and Zed, plus a generic option.
Its strength is that the contract lives in files. The constitution, spec, plan and tasks are Markdown in the repo, so they go through pull requests like code, and the converge step checks implementation against them until it reports Converged. Presets can enforce an organisation's standards, and the bug extension records a cause, fix and verification verdict under .specify/bugs.
The cost is reading time. Wavect's production guide says it does not make AI-generated code production-ready by itself, and when Scott Logic's CTO rebuilt two features with it, review took about 3.5 hours and 2 hours against roughly half an hour of agent time each. For a one-line fix that overhead is absurd, which is why the bug extension skips the full feature flow.
Two governance points matter to a buyer. Community catalog entries are discovery only, not audits, so a team should run a private catalog of approved extensions, which Spec Kit supports along with offline installs. And the weekly release pace means templates and command names shift; pinning a release, as the install docs describe, keeps every engineer on the same process. BMAD covers similar ground with more personas, and running both on one project produces two competing sets of plans.
Where it earns its place
A tech lead on a 12-engineer product team using Copilot and Claude Code
Make agent-built features reviewable before code is written
Each feature gets a spec and plan in a pull request that a senior engineer approves before /speckit-implement runs. Free software, but budget hours, not minutes, of human review per non-trivial feature.
An engineering director at a regulated fintech
Show auditors how AI-assisted changes were specified and verified
A preset encodes the firm's standards in the constitution and a private catalog limits extensions to approved ones. The spec, plan and converge output in git form a paper trail; the risk is reviewers rubber-stamping long documents.
A solo developer fixing a production bug with an agent
Fix a crash without losing track of the cause
`specify extension add bug`, then assess, fix and test commands write a report whose verdict is verified, partial or failed. Lighter than the feature flow and costs only the agent's own usage.
Before you commit
- Setup
- `uv tool install specify-cli` then `specify init my-project --integration copilot` (Python 3.11+, uv, a supported agent); the first spec comes from a few slash commands in the agent's chat.
- Integrations
- 41 named agents including Claude Code, GitHub Copilot, Gemini CLI, Codex CLI, Cursor, Cline, opencode, Kiro CLI and Zed; extensions, hooks, presets, workflows and bundles for customisation.
- Data and privacy
- Runs locally and writes Markdown into your repo; your agent sends that content to its own model provider. Telemetry is not mentioned in the docs. MIT licence; no GitHub security advisories published.
- Cost at scale
- No paid edition. The real costs are agent usage on your existing plan and reviewer time: about 2 to 3.5 hours per feature in Scott Logic's published trial.
- Limits
- Weekly releases change templates; community catalog is unaudited; heavy for small changes; does not combine sensibly with BMAD on one project.
What's good
- Spec, plan and tasks reviewed in git like code
- 41 named agent integrations plus a generic option
- Private catalogs and air-gapped installs for companies
- Converge step checks code against the spec
What's not
- Hours of document review per non-trivial feature
- Overkill for small fixes
- Community catalog entries are unaudited
- Frequent releases change templates and commands
Best for
Teams that want written, reviewable specs before an agent writes code.
Alternatives
Spec Kit: common questions
- Is Spec Kit worth it?
- Spec Kit turns agent work into reviewable documents in git, and charges for it in hours of human reading per feature. We rate it 4.3 out of 5.
- How much does Spec Kit cost?
- Free and open source (MIT).
- What are the downsides of Spec Kit?
- Hours of document review per non-trivial feature; overkill for small fixes; community catalog entries are unaudited; frequent releases change templates and commands.
- Who is Spec Kit best for?
- Teams that want written, reviewable specs before an agent writes code.
- How much does Spec Kit cost for a team or in production?
- No paid edition. The real costs are agent usage on your existing plan and reviewer time: about 2 to 3.5 hours per feature in Scott Logic's published trial.
- Is Spec Kit safe for company data?
- Runs locally and writes Markdown into your repo; your agent sends that content to its own model provider. Telemetry is not mentioned in the docs. MIT licence; no GitHub security advisories published.
- What are the best Spec Kit alternatives?
- The closest alternatives we have reviewed are BMAD Method, Superpowers and Matt Pocock's Skills.
Answers checked Oct 23, 2026.
Make Spec Kit? Show this rating on your site
