OpenSpec

A lightweight spec-driven framework for AI coding agents

Free· Free and open source (MIT license). No paid tiers — install via npm.CodingAgent-agnostic; works with Claude (Opus recommended), GPT, Gemini, and others via 30+ coding-agent integrations

Visit website →

Best for

Engineering teams working on mature codebases with multiple AI agents who want a durable, reviewable planning layer that keeps intent and implementation aligned across sessions and contributors.

Skip if

Solo hackers on greenfield prototypes, one-shot script writers, or anyone who wants an all-in-one hosted agent — OpenSpec is BYO-agent scaffolding, not a coding assistant itself.

OpenSpec is a lightweight, open-source spec-driven framework that sits between your team and AI coding agents, giving both a shared, version-controlled understanding of what a change is supposed to do before any code gets written. Instead of feeding an agent a one-off prompt and hoping it infers intent from the codebase, you author plain Markdown specs organised by capability — each feature or change gets its own folder containing a proposal, design decisions, spec deltas, and a task checklist. AI assistants read those artifacts as context, generate proposals aligned to your stated requirements, and hand back implementations you can review by diffing the spec, not just the code. It installs globally via npm (`npm install -g @fission-ai/openspec@latest`) and initialises inside any existing repo with `openspec init`, requiring Node 20.19+. The workflow revolves around four slash commands surfaced inside your agent — explore (brainstorm design options), propose (create a change with specs and design), apply (implement the tasks), and archive (retire completed changes) — with syntax that adapts to whichever assistant you use. OpenSpec integrates natively with 30+ coding agents including Claude Code, Cursor, GitHub Copilot, Windsurf, Gemini CLI, Cline, and Amazon Q, and the maintainers recommend pairing it with high-reasoning models like Claude Opus for the best planning quality. It is aimed squarely at teams working on mature codebases where drift between intent and implementation is costly, and where surviving context — across chat sessions, contributors, and agent swaps — matters more than raw generation speed.

Editor's take

OpenSpec is the closest thing I've seen to a sane 'spec-as-code' convention for the agent era: it treats requirements as first-class, diffable artifacts rather than throwaway prompts, and it does so without picking a favourite agent. If your team already reviews PRs and cares about intent surviving turnover, the overhead pays back quickly. If you're a solo dev vibing changes into Cursor, it will feel like paperwork.

— The AI Tool Bible editorial team

Pros

✅ Open source under MIT with no paid tier or vendor lock-in
✅ Agent-agnostic: same specs work across Claude Code, Cursor, Copilot, Windsurf, and 25+ other tools
✅ Specs and change proposals are plain Markdown in your repo, so they diff, review, and version like code
✅ Spec deltas let reviewers judge intent without reading every line of generated code
✅ Persistent context outlives chat sessions and onboards new contributors or new agents cheaply
✅ Lightweight install (single npm package) and initialises non-destructively inside existing projects

Cons

⚠️ Adds an authoring step before coding — overhead is real on trivial changes and prototypes
⚠️ Value depends on the underlying agent; weaker models produce weaker proposals from the same specs
⚠️ Early-stage project with a small ecosystem and evolving command syntax across integrations
⚠️ No hosted UI, dashboard, or team-collaboration layer beyond what Git already provides
⚠️ Slash-command syntax varies by agent (/opsx:, @opsx-, $openspec-), which is a small papercut when switching tools

Use cases

Spec-driven change proposals for AI coding agentsReviewing AI-generated code via spec deltas instead of line diffsPersistent project context across Claude Code, Cursor, and CopilotDesign exploration before implementation in mature codebasesOnboarding new contributors or new agents to an existing repoTask decomposition and checklist generation for feature workArchiving completed changes with an auditable requirements trail