Transcription
Around a month ago, we had taken a look at GitHub spec kit, an open-source toolkit designed to help you get started with spec-driven development. This approach can greatly improve both your code quality and development efficiency. It's actually quite incredible because it allows AI agents to focus directly on specifications that are executable, meaning they can generate real working implementations, not just vibe code or random suggestions that sound right but don't actually run.
Now, this framework is truly impressive and it can greatly elevate your coding workflow, bringing structure, clarity, and precision as to how AI collaborates with you. But the main thing is spec kit works best for brand new projects when you're working from scratch. Once your codebase evolves, managing updates and changes across specs becomes quite messy. It's not ideal for ongoing development, but that's where OpenSpec comes in.
OpenSpec is a new toolkit that builds on spec-driven development. It aligns humans and AI coding assistants on what to build before any code is written. It's quite lightweight and it's easier to work with than spec kit. It requires no API keys and it ensures your intent is locked before implementation and it's going to give you more deterministic, reviewable results instead of AI guesswork. It keeps everything structured with proposals, tasks, as well as spec updates, making scope changes auditable and transparent. And unlike toolkits like spec kit or something like Curo, which shines for new features, OpenSpec evolves when modifying existing systems, cleanly separating your current truth from proposed updates. In short, it's going to bring more control, clarity, and reliability while being lightweight to spec-driven AI development at any stage of your product.
But you may wonder, how does OpenSpec compare to these other toolkits like spec kit? Well, the thing is it's able to handle evolving features better with separate folders for current specs and changes. With Kurode, it is something that does better at grouping all updates for a feature in one place for easier tracking. And if it's something that's compared to something that doesn't have any sort of spec-driven toolkit, it is going to ensure more predictable, agreed-upon outputs instead of vague AI code that is generated from a natural prompt. And what's great is that it's supported by many tools like Cloud Code, Cursor, Open Code, Kilo Code, CodeX, and so many others. And there's going to be other tools that will be added soon like Root Code, when code. You even have Gemini CLI.
So apparently Zapier is dead again, right? Every time a new AI tool drops, people rush to say it's the end of Zapier. But here's the truth. It's never been more alive. Recently, we saw the shiny new OpenAI tool, Agent Kit, which is about answer optimization, not workflow orchestration. It helps ChatGPT think better. But Zapier helps you do more. Zapier doesn't just guess the right answer. It orchestrates real actions across 8,000 plus apps with deterministic reliability. And guess what? Zapier even plays nice with OpenAI's new MCP. So you can route answers through ChatGPT and automate results with Zapier. Zapier has been here before. Platform-native integrations, API pass tools, even ChatGPT itself. They were all called Zapier killers. Yet here we are, still leading, still integrating, still lighting the way for the future of automation. And now with Zapier Cobot, you can describe what you need in plain English and it will build a full production-ready workflow for you. Plus, the new Opus Clip integration lets you automate your video pipeline and repurpose content at scale. So the next time you see a Zapier killer, remember Zapier isn't dying. It's redefining AI orchestration. So use the link in the description below so that you can build smarter today.
But now let's get started and showcase how you can set this up. To install this globally, you can use the npm install command, but just make sure you have the prerequisites installed, having the latest version of Node. And once you have that installed, we can open up our terminal like Command Prompt and then paste in the npm install command to install OpenSpec globally. Once that is installed, you can then verify the installation by running the `openspec version` command. See if it's installed, which it is.
Now we can initialize OpenSpec into our project directory. So, I've gone ahead and I have created a template app for my AI detection tool that I'm going to be building out with this new OpenSpec kit. Now, what we're going to be doing is initializing OpenSpec within that project directory. So, we can then paste in the `openspec initialize` command, which will open up OpenSpec for me. Now, it is saying that you can configure OpenSpec with whatever AI assistant you want that we had mentioned. So if you want to use any of these different tools, you can now specify which one you want to work with. So you can click Enter and then you can select amongst these different options like GitHub Copilot, Kilo Code, or something like Cloud Code. I'm going to actually use Kilo Code in my case. So we're going to go ahead and click Enter. And after initializing it, you're going to see a couple of new files within this template for our AI detection tool.
Here it has created an `agents.md` file. And essentially, if you select the supported tools, say like Cursor or Cloud Code or something like Kilo Code in my case, OpenSpec will automatically configure custom slash commands and tool-specific instructions for that particular AI agent. So this `.md` file will then be handed off to my coding agent and it's going to be associated within my project file, which you saw that it had actually generated over here.
But now that we have everything ready, the first step is where we're going to populate our project context. So what we can do is first copy this natural prompt that OpenSpec provides and then we can go into whatever coding agent we're working with. So in this case, we're working with Kilo Code. I've gone ahead and added my folder as context. And then I can send in this prompt to read the `openspec.md` and help me fill out all the details that are necessary about my project. And you can see that the coding agent is going to get the instructions to read this file and make the necessary changes for my tech stack. So now you can see that it has filled out all the necessary details within the `project.md` with the comprehensive details about my AI detection tool.
But now the next step is where we're going to be able to give OpenSpec all the features that we want it to thoroughly follow. So in this case, it is going to say that you copy this prompt and then you add your features over here, like creating me the OpenSpec change proposal for this feature. So right now I'm saying that I am building an AI detection tool. Please create an OpenSpec change proposal for this feature. And within that `project.md` file, it is going to write up all the necessary details that are needed for the AI to follow in terms of the architecture pattern, the state management, testing strategies, and all the necessary details that will make sure that our app is proficient. So we can then go ahead and send in our feature request to Kilo Code, the agent that will then execute it autonomously. And the reason why I actually use Kilo Code over any of the other AI coding agents is because it's going to be able to understand all the different commands that would be necessary to use at any point, like using the `openspec list` command to take a look at all the different tasks that it has generated, and it will make sure that it will process all of them individually autonomously, as well as keeping you in the loop.
And after it has generated all the details for our proposal, it is going to allow you to review through all of these proposal files, like the design `.md`, has generated all the different task lists where you can review through everything and see if you want to actually follow through with it. If there's certain sections that you do not want to work on, you can take it out and add your own implementation. And essentially, this is the thought-out plan that it is going to work on deploying with the agent to create your AI detection tool or whatever you prompted in the initial section. Now all you've got to do is just implement the change, which is what they specified within the GitHub repository. So now I can just simply instruct the AI agent to implement the AI detection tool according to the OpenSpec proposal. All of these different proposals that it has listed in terms of the design, the task list, as well as the different specifications. I also told it, do not add any extra features beyond what's specified. So now we can have it work on tackling this task.
And you can see that it only costed 48 cents in comparison to what SpecKit did, which took approximately, I believe, 20 minutes to generate my proposed plan for the task list that I told it to generate. And it spent probably around $5, I believe, in my last video, which is a drastic difference from this, which is only 51. And as Kilo Code is working on executing this task of generating our AI detection tool with the specifications that the OpenSpec kit has developed, we can actually go within our terminal and we can run the `openspec view` command. This will give us a good understanding of the progress of the generation right now. Right now, it is working on 46 different tasks. We just started it. So as it progresses, it will give us a visualization of all the different changes being made. There we go. It has finished generating our AI detection tool. And this is what we have gotten. This is where we can paste in any sort of AI slop, like this AI story. And then we can analyze the text. And you can see right away that it is confident that this is AI detected, which is true, 'cause I had it generated by ChatGPT. And it talks about the analysis of what is AI-generated, like the different contents of this, as well as the metrics like the perplexity as well as the burstiness. And all of this was fully generated with Kilo Code, powered by OpenSpec, which is costing me $2 to fully generate alongside with the plan. And you can see how detailed it was. It made sure that it's functional in one shot and it added all the necessary things that we had requested, like the detection tool having different results being analyzed and a lot more. This is the power of OpenSpec. And I know the front end doesn't look good, but that's just odd code, or sorry, Cloud Sonic 4.5, which doesn't really generate the best front ends at the moment, but you can see that the quality is there in terms of the output.
If you like this video and would love to support the channel, you can consider donating to my channel through the Super Thanks option below. Or you can consider joining our private Discord where you can access multiple subscriptions to different AI tools for free on a monthly basis, plus daily AI news and exclusive content, plus a lot more. But that's basically, guys, for today's video on OpenSpec. This is a pretty cool new tool that can turn AI proposals into a structured, reviewable implementation plan that you can give to whatever AI agent. It's lightweight, which is why I really like it. And it's something that you can easily work with right away, where you can seamlessly move from specs to writing code a bit better than any other AI coding agent. But let me know what you guys think. This is something that is completely free and open source that you can easily get started. So, I'll leave all these links in the description below. Make sure you go ahead and take a look at the second channel if you haven't already. Join the newsletter, join our Discord, follow me on Twitter, and lastly, make sure you guys subscribe, turn on the notification bell, like this video, and please take a look at our previous videos 'cause there is a lot of content that you'll truly benefit from. But with that thought, guys, have an amazing day, spread positivity, and I'll see you guys really shortly. He suffers.