Spec-first logo

Spec-first

Spec-driven development workflow for AI agents

Guides AI coding agents through specs, reviews, worktrees, and gated commits to deliver verified pull requests.

AI coding toolsspec-driven developmentcode reviewworktree managementcommit gating
Spec-first official website screenshot
spec-first.dev · Website preview · 2026-10-28View full screenshot ↗

What is Spec-first?

Spec-first is a method and plugin for AI coding agents that replaces unstructured vibe coding with a structured, spec-driven workflow. It helps you describe a feature in your own words, then guides the agent through four commands: create a spec, review it, apply it, and deliver it. Each step leaves a file behind, so the process is transparent and resumable.

You start by describing what you want. The agent reads your repository and memory to avoid asking unnecessary questions. Then /create-spec writes a plan with goals, steps, and done criteria. /review-spec has a fresh reviewer check the spec against your code. /apply-spec builds the feature in its own worktree, and /deliver-spec runs your checks before committing and opening a pull request.

Spec-first is MIT licensed and works as a Claude Code plugin, but it can be used with any agent. It supports running multiple specs in parallel, each in its own branch and worktree. The tool also logs every decision with its alternatives, so the reasoning stays next to the code. It's designed for developers who want more control and verification in AI-assisted coding.

Spec-first features

  • Spec creation: /create-spec writes a goal, steps, facts with file references, and done criteria.
  • Spec review: /review-spec has a fresh reviewer check the spec against your code before building.
  • Isolated worktrees: Each spec gets its own branch and worktree, so you can run several side by side.
  • Commit gate: /deliver-spec runs your checks before committing, ensuring only verified code is merged.
  • Resumable state: Context is saved in context.yaml, so you can restart and pick up where you left off.

What you can do with Spec-first

  • Describe a new feature in your own words and let the agent turn it into a spec.
  • Run multiple feature specs in parallel without conflicts, each in its own worktree.
  • Review a spec before any code is written to catch mistakes early.
  • Deliver a series of specs as one pull request with verified commits.

How to get started with Spec-first

  1. Install the Spec-first plugin via the marketplace command shown on the website.
  2. Describe your feature in your own words; the agent reads your repo and memory first.
  3. Run the four commands: /create-spec, /review-spec, /apply-spec, /deliver-spec.

Tips for better results with Spec-first

  • Start by describing your feature in plain words, then run /create-spec to generate a structured plan with goals, steps, and done criteria before any code is written.
  • Run /review-spec after creating a spec to have a fresh reviewer check it against your codebase, catching mistakes early and reducing back-and-forth during the build.
  • Use /spec-first:deliver-series for larger changes by cutting them into multiple specs, letting Spec-first build independent groups in parallel and open one pull request.
  • Keep your repository memory and context files updated, as Spec-first reads them to avoid asking unnecessary questions and to resume work after interruptions.

Spec-first pricing

Spec-first’s homepage does not list prices. Check spec-first.dev for current plans.

What to check before you rely on Spec-first

  • Verify your agent supports Spec-first's four commands, since it is a Claude Code plugin but can work with any agent, and confirm the installation method matches your setup.
  • Check that your repository has the required.spec-first directory structure and that context.yaml is writable, as Spec-first saves state there for resumability.
  • Confirm the commit gate runs your own checks before merging, and review the logged decisions with alternatives to ensure the reasoning aligns with your project's standards.

Who Spec-first is for

Spec-first suits developers, development teams and AI coding agent users. If that is not you, the AI coding tools below may fit better.

Similar tools compared with Spec-first

ToolWhat it isPricing
Spec-firstSpec-driven development workflow for AI agentsSee website
TabnineAgentic quality engineering platform for AI codeSee website
CodeRabbitAI pull request reviewer and prioritizerFree trial
CursorAI coding agent for ambitious softwareSee website
Claude CodeAI coding agent for terminal and IDEPaid

Spec-first FAQ

What is Spec-first?

Spec-first is a method and plugin for AI coding agents that structures development around specs. You describe a feature, and the agent creates a spec, reviews it, builds it in a separate worktree, and delivers it through your checks.

How does Spec-first work with AI agents?

It provides four commands: /create-spec, /review-spec, /apply-spec, and /deliver-spec. Each command handles a step in the workflow, leaving files behind for transparency. It works as a Claude Code plugin but can be used with any agent.

Is Spec-first free?

The website states that Spec-first is MIT licensed, which means the code is open-source and free to use. However, it doesn't mention any paid plans or pricing details, so check the GitHub repository for more information.

Who is Spec-first for?

It's for developers and teams who use AI coding agents and want more control and verification. It's especially useful if you're tired of agents making unverified changes and want a structured, reviewable process.

What file formats does Spec-first use for specs and decisions?

Spec-first stores everything as plain YAML and Markdown files directly in your repository. Specs, decisions, and context are saved in a.spec-first folder, so you can read, edit, and version them like any other code. The website does not mention exporting to other formats.

Which AI coding agents can run Spec-first besides Claude Code?

Spec-first is built as a Claude Code plugin, but the website says it can be used with any agent. It lists Claude Code, Codex, Cursor, and Copilot as compatible tools. You install it via the plugin marketplace command shown on the site, and the workflow commands work the same way.

Does Spec-first require a paid plan or subscription?

Spec-first is MIT licensed, which means the code is free to use and modify. The website does not mention any paid plan, subscription, or pricing tier. You can install it directly from the plugin marketplace without signing up for a separate account. Check the GitHub repository for any future licensing changes.

Spec-first alternatives

Other tools you can compare with Spec-first. Each does a similar job to Spec-first with a different focus, plan or price.

Tabnine logo

Tabnine

Deploys AI agents to test and quality-check AI-generated code across your software delivery lifecycle.

CodeRabbit logo

CodeRabbit

Automatically reviews every pull request, suggests fixes, and helps you prioritize and secure code changes.

Free trial

Cursor logo

Cursor

Cursor is an AI coding agent that turns ideas into code, letting you delegate tasks while you make decisions.

Claude Code logo

Claude Code

An AI coding agent that understands your codebase, edits files, and runs commands to help you ship faster.

Paid

Devin logo

Devin

Devin is an AI coding agent that plans, codes, tests, and opens pull requests for you.

Windsurf logo

Windsurf

Manage local and cloud coding agents from one desktop IDE, plan tasks, review work, and ship code.

GitHub Copilot logo

GitHub Copilot

AI coding assistant that suggests code, answers questions, and automates tasks in your editor and GitHub.

Freemium

Sourcegraph Amp logo

Sourcegraph Amp

Amp runs coding agents on remote machines you control from anywhere, so you can build and ship without juggling local setups.

Lovable logo

Lovable

Describe your idea in plain language and Lovable builds full-stack apps, websites, and internal tools you can deploy.

Freemium

OpenRouter logo

OpenRouter

One API key gives you access to 500+ AI models across 80+ providers with automatic fallback and cost control.

Freemium

Replit logo

Replit

Describe your idea and Replit builds a working app or website for you in minutes, no coding needed.

Freemium

Kimi logo

Kimi

Kimi is an AI assistant that helps you code, create slides, research, and run tasks with agents.

See every tool in this category

ChatbotsWritingImage generatorsPhoto editingVideo generatorsVideo editingAvatarsVoiceMusicTranscriptionCodingApp buildersAgentsSearchResearchStudentsDataProductivityPresentationsDesignMarketingBusinessCareers3D & gamesTranslationDev platformsDetectors
NewFree AI toolsSubmit a tool