OpenSpec: Spec-Driven Development for AI Coding Assistants

TL;DR: Fission-AI/OpenSpec is the TypeScript, MIT project that turns spec-driven development into a workflow AI coding assistants can follow. 69,373 stars on GitHub, docs at https://openspec.dev/.

What it is

Fission-AI/OpenSpec describes itself as “Spec-driven development (SDD) for AI coding assistants.” That sentence positions the project at the intersection of two ideas: the older discipline of writing a PRD or design doc before code, and the newer category of AI coding tools that prefer structured context over freeform prompts. The repository is written in TypeScript, sits at 69,373 stars, and is published under the MIT license. The homepage field is https://openspec.dev/, which is where the project’s documentation lives. The GitHub topic list is short and on-message: ai, context-engineering, engineering, planning, prd. The repository was created on 2025-08-05, which makes it a recent entrant that has clearly found an audience.

The shape is a workflow you drop in front of an AI agent: write a spec, get the agent to confirm it understands the spec, then implement against it. The README on openspec.dev walks through the loop.

What it does well

The interesting question for any “AI workflow” tool is what it does that a plain prompt cannot. The topic list points at the answer: OpenSpec is less about generating code and more about structuring the conversation that produces the code.

• PRDs as a first-class artifact. The prd topic signals that OpenSpec treats a spec like a real document with sections, not just a paragraph stuffed into a prompt. That matters once a feature needs more than a paragraph to describe.
• Planning before code. planning and engineering in the topic list describe a workflow that pushes architectural decisions upstream, into the spec, rather than mid-stream as the agent is writing.
• Context engineering. The context-engineering topic is the technical name for the practice OpenSpec is built around — assembling the right slice of background, constraints, and acceptance criteria before asking the agent to ship.
• Tooling-neutral by design. ai is the only assistant-related topic, and the GitHub description explicitly mentions “AI coding assistants” in the plural. The workflow works the same whether the consumer is Claude Code, Cursor, or something else.
• MIT license. Commercial use is unblocked, which is why the project is reasonable to wire into a private product team’s process.

The 69,373-star count at 296 stars today reflects a community that has latched onto “spec first, code second” as a credible answer to the “agent hallucinated the wrong API” problem.

How to try it

OpenSpec ships as a CLI that initializes a spec directory and a workflow you drive from there.

# Install the CLI
npm install -g @fission-ai/openspec

# Initialize a spec project
openspec init

# Ask the agent to draft a change
openspec change "add user profile settings page"

# Validate the proposed spec
openspec validate

For AI-agent setup, the workflow runs through whatever shell-capable agent you already have. The agent reads the spec, asks clarifying questions, and only then writes code.

# Hand the spec to your agent
openspec show change "add user profile settings page" | claude

When it fits, and when it doesn’t

OpenSpec is the right tool when you have stopped trusting the “vibe coding” loop — when you want a paper trail for what the agent is about to build, and when the feature is large enough that a single paragraph prompt would lose important constraints. It is a strong fit for teams that already write PRDs and want the AI agent to honor them, and for solo developers working on non-trivial features where one round-trip is cheaper than ten small ones. It is less attractive for tiny one-file changes, for throwaway scripts, or for cases where the right answer is still “let me just see what the agent comes up with.” The MIT license means there is no friction on adoption.

The 69,373-star count, the recent creation date, and the topic list all point at the same conclusion: a workflow tool that has found product-market fit at the moment AI coding assistants are graduating from autocomplete to multi-step agents.

Source and Accuracy Notes

• Project page: https://github.com/Fission-AI/OpenSpec
• Source repository: Fission-AI/OpenSpec (verified via GitHub API html_url)
• License: MIT
• Stars at time of writing: 69,373
• Language: TypeScript
• Topics: ai, context-engineering, engineering, planning, prd
• Homepage: https://openspec.dev/
• Created: 2025-08-05