Post

What's the best way to get GitHub Copilot in agent mode inside VSCode to repeatedly make modifications to be executed in a GitHub Actions workflow, check the results, then iterate until the workflow runs successfully? It's great at automatically building, running and fixing stuff *locally*

1 repost2 likes

Hopefully someone has already gone through the pain? Is there an MCP server for GitHub Actions yet? Otherwise I'm thinking Copilot could be instructed to call the GitHub CLI with a flow that pushes to a branch, triggers a workflow, checks the logs, etc.

I told it to start using the GitHub CLI, and it did. It would still be nice to have this more deeply integrated though, especially when it comes to watching workflow runs, instead of polling to then pull the logs, etc.

Also, I noticed one weird glitch: when it runs the command to change directory to then call "git commit", it hangs waiting for the command even though it has completed. I have no idea why it does an explicit "cd" before "git commit". Entering "ls" in the terminal unblocks it

"on GitHub itself" - I've got VSCode with my local git clone, I don't want to do that elsewhere. Ideally I'd just do it from a single place, which I've been doing this morning by just asking Copilot to call the GitHub CLI to test the workflow and check for status and output

I mean, if you are relying on changes pushed to github, and on Github CI as source of truth for whether it works, it just seems easier to avoid local clone altogether, and let it keep working on problem even when you are offline. But that's just me.

Maybe I don't fully understand, would it provide a better way to repeatedly make and test changes to a GitHub Actions workflow differently from using a local git clone? It would still have to create a commit and trigger the workflow anyway?