My GitHub Actions Failures Now Start a Local AI Agent
About this video
GitHub Actions failed again. Instead of sending my coding agent the same message, I connected the failure to the agent. In this walkthrough, I use ClawTab to send a failure webhook through its relay to my Mac. The local daemon starts a coding agent to inspect the failed workflow and push a fix. I verify the connection with a controlled test: add a workflow that fails on purpose, then instruct the repair agent to remove it. This checks that the webhook starts the local agent and that the agent can push a change. I also explain how the same approach could connect other workflows, and how to run the agent on a server so your Mac doesn't have to stay open. ClawTab: https://clawtab.cc Chapters 0:00 Why I wanted to automate CI failures 1:09 Connecting GitHub to ClawTab 1:43 How the relay starts a local agent 2:28 Testing the webhook and repair loop 3:16 Checking the result 3:33 Setup and other workflow triggers 4:09 Running the agent on a server #GitHubActions #CodingAgents #BuildInPublic
Transcript (automatic captions)
Okay, hello everyone and welcome to this quick video about how to manage your GitHub workflow failures and send them off to your agents to resolve them automatically. And you can see I have a lot of failures here for my for my workflows and they run for a for a long relatively long time as you can see 16 minutes. And and they they they fail regularly. So this is the my Bmax project which is the web browser itself as you can see here. It's I have never looked at them myself. I always ask the AI to solve them. And how I do it is I as you can see my my latest command here, I just sent the GitHub actions failed again, can you fix? And I have asked it to do this for I don't know 10 times already and now I want to then have it automated. So how to do it? So so this feature uses CloudTab, my other project that allows you then to to send the the hooks to your local computer via the the relay here in remote.cloudtab.cc and because this is a I think very very very useful setup where GitHub can call your local computer, start an agent here that will check the workflow failure and then fix it and and push it. So essentially we have the CloudTab relay, uh we have the CloudTab daemon running locally and uh and we have GitHub hooks. So we have uh failure in GitHub, then it will send a hook to the to the relay here. And this relay has uh is always on and connected to our local computer and then send it it sends the command here to start an agent. And agent will fix it and push a uh new commit and then hopefully the the issue is is resolved. Now, if the AI says that it set it up for me, the only thing to do is is now to just verify it. So, we we can make a setup where we make the one of the checks fails fail and see that it it starts a new agent here locally and that it fixes it. So, let's ask the agent to do it. Can you set up a test case where you add a new simple workflow that fails immediately and add a comment here that the repair agent should just remove this workflow to to make it succeed the next time. So, this is a test to verify that the hook works and the new agent starts at my computer and pushes a fix. Uh this one failed. We are now on the green. Yes, we are. So, we can see that uh it is fixed. The The workflow failure is fixed uh via this agent in our local computer now. And uh if you want to set it up, then then you can go to cloudtab.cc and and install the local demon and and the relay to your computer. And this has been uh just a a single feature that is available, but you can imagine you can hook this up to any of the systems or is it Zapier or Slack or whatever workflow you are using and then call the hook in in the Cloud Relay and then send and start the agent in your local machine. The other option is is to have the Cloud Relay running on your server as well. And then you don't have to have your local Mac always open, but you can send the relay command to your agent in in the server. And let's make our workflows better together. Thanks for watching and have a good day.