Transcription
If you have been following along with the AI coding space, you probably know that getting AI to build exactly what you want can be really difficult. Sometimes you get amazing results. Other times you get code that kind of works but does not really match what you had in mind. That is where OpenSpec comes into play. And trust me, this is not just another tool, but this one actually solves a very serious problem.
A while back, many of us were excited about SpecKit. It was this cool open-source toolkit that introduced us to specification-driven development. The whole idea was pretty smart. Instead of just throwing simple prompts at AI and hoping for the best, you would create detailed specifications first. The AI would then use those specifications to generate actual working code, not just random suggestions that sound good but break when you try to run them.
But SpecKit got some serious problems. For example, once your project started growing, keeping those specifications updated became very difficult with SpecKit. That is exactly why OpenSpec exists. OpenSpec takes the core idea of specification-driven development but makes it practical for real-world scenarios. It works beautifully whether you are starting fresh or working with an existing codebase that has been around for months.
The philosophy behind OpenSpec is simple but powerful. Before any code gets written, before your AI assistant starts generating files, you and the AI align on exactly what needs to be built. You log the intent first, then you implement. This means you get predictable, reviewable results instead of guesswork that you have to constantly fix.
What I really appreciate about OpenSpec is how lightweight it is. You do not need any API keys to get started. You do not need to sign up for yet another service or manage credentials. It just works right out of the box. Everything stays structured with proposals, task lists, and specification updates. If you need to change scope or modify requirements, everything is auditable and transparent. You can see exactly what changed, when it changed, and why it changed. This level of clarity is something you just do not get when you are working with raw AI prompts.
This is how OpenSpec compares to other tools in this space. Compared to Specate, OpenSpec handles evolving features much better because it uses separate folders for your current specifications and your proposed changes. This separation is brilliant because you always know what is your source of truth and what is being proposed. Compared to tools like Cursor with no specification framework at all, OpenSpec ensures you get exactly what you wanted. The difference is night and day once you experience it.
The best part is OpenSpec supports pretty much every major AI coding assistant you can think of. GitHub Copilot, Cursor, Wind Surf Client, or many others all work seamlessly with OpenSpec. More tools are being added regularly too. So no matter what your preferred development environment is, you can probably use OpenSpec with it. This flexibility means you are not locked into one ecosystem. You can switch between different AI systems while maintaining the same structured approach to development.
All right, let me introduce you to Engage, a Chrome extension that helps you skyrocket your LinkedIn growth. You see, to actually grow on LinkedIn, you need to engage: like, comment on posts, and stay active constantly. But let's be honest, writing thoughtful comments every day is a really time-consuming task. That's exactly where Engage comes in. It uses AI to automatically write relevant comments for you right inside the LinkedIn comment box. Just click the comment button, then hit the AI icon, and boom, it instantly generates a meaningful comment based on the post. You can also edit it if you want to, or let Engage handle multiple posts using the auto-suggest and auto-like features, making your profile look super active while saving you tons of time. So, if you want to grow really fast on LinkedIn, download Engage today for free. The link will be in the description.
Let me walk you through how to actually set this up and use it because seeing it in action makes everything click. First, you need to install OpenSpec globally on your system. Make sure you have the latest version of Node installed as the prerequisite. Once it is ready, you can open your terminal and run the `npm install` command for OpenSpec with the global flag. The installation is quick and straightforward. After it finishes, you can verify everything worked by running the `OpenSpec version` command. If you see a version number pop up, you are good to go.
Now, I'm going to show you how to use OpenSpec with a task management application. I have already created a basic template for this project with some starter files, but nothing is really implemented yet. This is where OpenSpec shines. We are going to initialize OpenSpec in our project directory, and it will set up all the necessary structure for us. You run the `openspec init` command in your project folder, and it will prompt you to choose which AI assistant you want to work with.
When you initialize OpenSpec, it asks you to select your preferred coding assistant from a list. You will see options like GitHub Copilot, Cursor, Client, and others. I'm going to select GitHub Copilot and press Enter. What happens next is really cool. OpenSpec automatically configures everything for your chosen assistant. It creates custom slash commands and tool-specific instructions that are optimized for how that particular AI works.
After initialization completes, you will notice some new files in your project directory. The most important one is the `agent.md` file. This file contains all the instructions and context that will be handed off to your coding agent. It is specifically tailored for GitHub Copilot in my case. So the AI knows exactly how to work with OpenSpec in this project. You will also see a `project.md` file that OpenSpec created. This is where we are going to define all the important details about what we are building.
The first step in the OpenSpec workflow is populating your project context. OpenSpec provides you with a template prompt that you can copy. Then, you open up GitHub Copilot or whatever assistant you are using. Make sure your project folder is added as context, and you paste in that prompt. The prompt tells the AI to read the OpenSpec project markdown file and help you fill out all the relevant details about your project. The AI will fill in all the information about your tech stack, your architecture preferences, your testing approach, and other important decisions.
Watch what happens next. When I send this prompt to GitHub Copilot, it reads the OpenSpec configuration and starts filling out the `project.md` file with comprehensive details about our task management application. It documents our technology choices, our project structure, and coding standards – everything that matters. This becomes the single source of truth that the AI will reference throughout the entire development process.
For this task management project, I want a full-featured task manager where users can create tasks, organize them into projects, set due dates, add priorities, and mark tasks as complete. I also want a clean dashboard where you can see all your tasks at a glance. Instead of just asking GitHub Copilot to build this directly, we are going to use OpenSpec to create a formal changes proposal.
First, I copy the feature request prompt that OpenSpec provides, and then I modify it for my specific needs. I tell GitHub Copilot that I am building a task management application and I want to create an OpenSpec changes proposal for the core task management feature. I list out what I want included, like task creation, project organization, and status tracking. Then I send this request to the AI.
What GitHub Copilot does next is amazing. It does not just start writing code immediately. Instead, it creates a detailed proposal that breaks down exactly how this feature will be implemented. The AI generates several proposal files in a `changes` folder. There is a `design.md` file that outlines the technical approach, a task list that breaks the feature into manageable chunks, and specification files that define the exact behavior of each component. All of this is generated automatically based on the project context we set up earlier.
I simply tell GitHub Copilot to implement the task management feature according to the OpenSpec proposal. I make it clear that it should follow the specifications exactly and not add any extra features beyond what was defined. The AI then goes to work systematically implementing each task from the proposal because everything is well-defined and structured.
While GitHub Copilot is working on implementation, you can actually monitor the progress using OpenSpec commands. If you open a terminal and run the `openspec view` command, you will see a visualization of all the tasks being worked on. It shows you which tasks are completed, which are in progress, and which are still pending. This visibility is really helpful for longer implementations because you can see the AI making steady progress through the task list instead of wondering what it is doing.
After the implementation finishes, you have a full, functional task management application. Let me show you what was generated. We have a clean interface where you can create new tasks with titles and descriptions. You can assign tasks to different projects, set up due dates using a date picker, and mark priority levels. There is a dashboard view that shows all your tasks organized by status. You can filter tasks by project or priority. You can mark tasks as completed, and they move to the completed section. The entire process from initialization to working application took maybe 30 minutes of actual work time. Most of that was reviewing the proposal and making small adjustments. This is the power of specification-driven development with OpenSpec. The implementation follows a clear task list. Changes are tracked and documented.
So, I have left all the relevant links in the description below if you want to try OpenSpec by yourself. So that's it from the video. I hope you enjoyed it. If you did, please hit the like button and subscribe to this channel, and I will see you in the next video.