AI + Chainlink Coding: How To Use Agent Skills To Build 10x Faster

Hey folks, Dave here. I'm really excited to share something special with you today. It's 2026 and most of us are building together with AI tools. Stuff like Claude Code and Cursor have really elevated the productivity level of what we can do as developers. To me, it almost feels like I have an entire team behind me when I'm coding. So, I'm excited to talk to you today about something our team has built to make using Chainlink with your AI friends even easier. It's called Chainlink Agent Skills, and I think it's kind of a gamecher. Chainlink Agent Skills is an open-source library you can start building with today. It's part of the already popular Chainlink smart contract kit. So you may have already downloaded it. The library is public. And with this library, we've built MCP skills specifically for CR development. This enables AI assistants like cursor and claw to help build your CR workflows. It understands CR patterns, triggers, capabilities, and it's trained on the official Chainlink developer documentation. If you haven't used an MCP before, no worries. This space is moving so fast and there's no time like the present to jump in. MCP stands for model context protocol and it's a standardized way to allow AI assistants to use all of these specialized tools and skills. Think of it as plugins for AI coding assistants. CRA, the Chainlink runtime environment, is an orchestration workflow engine for building institutional-grade smart contracts that bridge onchain and off-chain execution. You've probably seen me do a bunch of videos on this already. Think of it as connecting both the DeFi and traditional finance world in a secure, auditable, private, and consensus focused environment. As a CR developer, you already changed project timelines from like weeks to days and sometimes hours. And now with Chainlink agent skills, your AI tools can start to build right alongside you. Think of this too as a way to learn CR patterns through guided examples. TypeScript and Go language support are already there. And because it's part of the Chainlink smart contract kit, it's completely open- source and communitydriven. Let me show you how simple this is to get started. Okay, so I'm in cursor. I just created a folder, called it chain link agent skills demo. The only thing I did was copy over the license and the read me from the repo just so we could look at it. Uh this is going to give you information about the format, what the YAML looks like for the skills, what current support is currently CRA uh supports. That's what we're going to cover today. Uh how to get set up, how to move those skills into uh Claude so that cursor can pick it up and understand everything. And then you're going to do a /c skills command to be able to use everything. And you can pick your, you know, your favorite uh API model. I like Opus 4. 6. six. Find it does well. So, we're going to use that. So, if I have a new chat up here, you'll see I have Opus 4. 6. Should feel free to use uh whatever you'd like. I'm going to ask it. This is just a way to see what you've got installed and what cursor has. So, what skills are available for use? So you see it's picked up the chain link runtime environment and this helps you learn CR quickly and it references the docs so we can create some workflow doing typescript and go like so that sounds good to me. Uh the other step that you're going to need to do which you probably already have on your system just make sure you have uh CR right so I have CR installed you can see version if you have it logged in just do that CR login if you haven't signed up for CR just go to chain. link/c link/c. You only need an email address and you'll be able to log in and then that'll get you all started. Remember, everything you can run uh locally so you got a nice uh simulator to run locally. All right, so we've got this one and what we're going to do now is start a new guy and let's ask it. So using and then I see I did the command here. We got CR skills and we can see everything's working there. So using CR skills, create a simple hello world workflow in Typescript. Okay, we're just going to keep this super simple and see how it does.

Now, as it runs, I've already done uh a white list on the Chain Link documentation portal. So, if you do get that, uh, you can go ahead and tell it to use that moving forward. That's really the only other step. You can see it's going through the markdown files to figure out how everything's running here. And it's following the documentation. So, it's checking the version it's got. It's got 1. 2. It's going to check for any — um prerequisites or anything like that it may need. And then it's thinking uh it's lookings like it's creating the files here. You can see our main TypeScript file. Uh it's creating a git ignore an environment file. Seeing if we've got bun installed. All our dependencies look good. Now remember uh AI is non-deterministic. So if the order of everything you see here in my videos a little different than yours, don't worry. AI is like human beings. Everybody's special and a little different and every time it's a little different. So uh you can see here though, uh it looks like it ran uh it created the project structure for us here, right? Right. So we can see our hello world workflow environmental file our git ignore. It's showing you some of the workflow code. So you can see on cron trigger hello world workflow triggered return a hello world initiate the workflow. Gives you a nice little description here. So I've got a trigger handler and an entry point in main. It's saying the simulation ran successfully and we can actually run it ourself. So, if we want to copy and never trust, always verify, right? So, we're going to come in here. We're going to go to see Halo World. And then we're going to simulate using the same thing. We uh move this up for you. And you can see this is just like if you were using Siri in the terminal, running it with a cron trigger. Hello world. Workflow trigger workflow simulation result hello world. So, look at that. If we come in here into the folder, you'll see it's set up the git ignore for us. Uh we've got our project. yamel with information here. What kind of uh you know chains that we might be using for our RPC calls, which basically I'm sure you're familiar with uh environment. So if you want to go ahead and add your private key here, you would need to create a wallet and use that. You can see the code that we just looked at. Hello world. In here, it's got a little readme. You can see the YAML file that sets up our workflow for our CR workflow. And uh that's good. We have uh asked the AI to go ahead and create a simple workflow for us. And it did. Exciting. Okay, let's move up a little bit back in. Yep, we got our hello world example. Let's go ahead and start a new chat. Now, there's all sorts of like places you could take this You could actually ask the AI to go and convert that sample that we had into Golang if you wanted. You know, it's very good. AI is pretty good at like converting between languages and things like that. I think what we'll do is a more advanced one just to kind of show how you can pull in some real world stuff. So, I already wrote this. It's a longer prompt. So, let me just paste it in here for you all. and I'll walk you through what it's doing. Okay, so we're going to use CR skills and we're going to create a workflow, but this time we want to read cryptocurrency prices from the Chainlink data feeds services. And if you've ever done anything with chain link, you're probably familiar with Chainlink data feeds. Uh, in fact, it's something like over 80% uh on Ethereum chains uh are running and using it backed by chain link. So, it's a pretty popular service. It runs on top of the Oracle network and it will give you a deterministic price. So, it's going to go and pull from all over the place and actually give you a single price for different tokens. We're going to use the Go language for this workflow. So, just so I can show you the different language and we're going to use the read data feed template pattern to read uh we're going to do Bitcoin and ETH. So, we're going to read from the chain link data feeds on the arbitrum uh blockchain. We're then we're going to have a cron trigger every 30 seconds and then we're going to log the prices and return them as a JSON. And then we're going to ask the AI here, let's name the workflow read data feeds. So, when this is done, we should be able to kick off that workflow anytime we'd like to with read data feeds. Let's fire it off. We're going to go ahead and allow GitHub here, too. So, it can go ahead and pull. It needs to pull the data feeds

template. I'm allowing to pull everything here. All right. Looks like we're getting a workflow here. Read data feeds. Going to go up on arbitro. I can review its code changes here if I'd like. Setting up our AMD file. All right. So, we got our project structure. You remember this is in Golang, so it's going to look a little bit different. We're going to have our go files. We got our workflow. We've got our configuration files, our env trigger again every 30 seconds. And then for each feed, so bitcoin and ethereum, we're going to have uh decimals the latest answer. We're using arbitrum one. We're going to return JSON. Uh then generate the go bindings. This creates go bindings under. If you try to do this with uh typescript, you might get some like, you know, EVM errors. I find that go works good here. It's just why I'm doing it in go. Uh, and then you can see workflow simulation results. Give the simulation a try and let me know if you got any errors. So, let's go ahead and give that. Okay. So, this is that non-deterministic output I talked about. So, you may or may not get this, but when my AI created this, it made a couple of mistakes. And this is the fun part, you know, it's like working with another developer. You get that high when you uh you know, you finish and you solve you complete the bug, right? So when I ran this, we were getting some errors like stuff was missing. You know, you could see everything up here failed to compile the workflow. So CR couldn't I asked it, I said, "Hey, I'm getting errors. Uh can't set the project context. No project settings found in the current directory. " And you could see the AI came back and said, "Oh, I can see two issues from the terminal output. The generate bind failed because it expects the AI uh in a different location, right? Probably, you know, pretty common bug. And then the go build fails because it doesn't have that. So, let me fix the structure. " And you can see here it went and it made a new directory and it moved stuff around and it uh corrected it and said, "All right, try running this stuff and we're going to go ahead and generate the bindings and then simulate the workflow. " And uh you know it's pretty confident as we all are. I think it's definitely the AVI path, right? It's definitely that's where it was. And uh like I got the following error, dude. And uh and they went through and you can see that error is down here. Uh right. And it's basically looking for uh an AI extension, right? Not. json. So he goes in and he moves the stuff around, fixes it, puts in the extension and says, "Hey, try rerunning this from the workflow directory. We're going to generate the bindings for the EVM. Go mod tidy and then simulate the workflow. " So I went, "Hith it uh generated all the bindings, right? We did the go mod tidy and uh then it downloaded all the dependencies that it needed and then it was able to compile it, right? So the workflow was compiled. So then the simulator kicked it off. Let's go ahead and uh simulate it. And you can see pulled the feed. Uh got the latest answers. Bitcoin's coming in at around 69,133. A little bit up today. And then 2355. So simulation results. You can see right here it's formatting. This is the JSON file that it returns. Right. And um let's go ahead and run that again just so you can uh you can see it. Let me uh run it simulation. We should get back uh so you'll see a different uh price. So this is uh you know we had a hello world and here's our read data feed. So you can see everything up here. We have our smart contracts been generated. We have our workflow the workflow. yaml that you can see up here right our environment file everything uh and then our project YAML. So this is a real uh CRV workflow. Let's go ahead and run that again. And boom. See 68933 and E 2030. So look at that. Using AI and debugging with AI, we were able to actually create a CR workflow that uses the chain link data feeds on arbitum and gives us real-time pricing for both Ethereum and Bitcoin. Oh, awesome. That was so much fun. I feel empowered as a developer

when I'm working with these AI tools. And I can't help but think this is the future of development. Humans and AI building together. And we're building here the future of finance. For more info and what we built here today, check out CRA at chain. link/cre. and make sure you head over to our GitHub repo at smartcontractkit/chainlink- aent. skills and then you can start building these AI tools today. Thank you.