OpenSpec

Workflow

Spec-driven development CLI: agree on proposals and specs with AI before code is written, via slash commands.

by Fission-AIFission-AI/OpenSpec

Stars72.7k

Forks5k

Watchers286

Open issues206

LicenseMIT

Last pushtoday

Install

install

$ npm install -g @fission-ai/openspec@latest && openspec init

Update

update

$ npm install -g @fission-ai/openspec@latest && openspec init

Re-running the global install with @latest upgrades the CLI; re-run openspec init to refresh the scaffolded /opsx: commands.

When to use it

You want AI coding assistants to agree on a spec before writing code, instead of relying on chat history
Starting a new feature and want a structured proposal, requirements, design, and task checklist
You need a lightweight spec layer that works across Claude Code and 20+ other AI tools, not a heavyweight process
You want each change tracked in its own folder with delta specs showing what is changing

Commands & flags

openspec init: Scaffold the openspec/ directory and generate the /opsx: slash commands.
/opsx:propose <what-to-build>: Draft a proposal with requirements/scenarios, design, and tasks.
/opsx:apply: Implement the tasks from the proposal.
/opsx:archive: Archive completed changes.

How to use it

OpenSpec is an npm-installed CLI that wires slash commands plus agent instructions into your project for Claude Code and 20+ other AI assistants.

npm install -g @fission-ai/openspec@latest
cd your-project
openspec init

openspec init scaffolds an openspec/ directory and generates /opsx: slash commands. Drive the workflow from inside Claude Code:

/opsx:propose <what-to-build>   # proposal, requirements/scenarios, design, tasks
/opsx:apply                     # implement the tasks
/opsx:archive                   # archive completed changes

Each change gets its own folder (openspec/changes/<feature>/) with proposal.md, specs/, design.md, and tasks.md. Requires Node.js 20.19.0+.

Why it earns its place

OpenSpec puts a lightweight, reviewable spec layer between intent and code so AI assistants stop improvising from chat history. Its delta-spec model shows exactly what a change adds or modifies relative to current specs, each change is isolated in its own folder, and the same /opsx: workflow works across Claude Code and 20+ other AI tools, making it a portable team convention.

Try it

claude

> /opsx:propose add rate limiting to the public API endpoints