Set GitHub Action workflow for the AI agent (based on the devcontainer)

[gvegayon]

[olivia-banks]

I'm not actually entirely sure what this is. Talking to Claude, it thinks that it's the action that runs in GitHub Agents to set up their environment. This should be pretty simple. We can use the Containerfile as the image for the action, which already has everything we need installed, and then just run the post-install script in .devcontainer to install language servers and things.

Might take a little bit of looking into to see how we can expose the LSP as an MCP, or maybe the agent is smart enough to automatically do this.

[gvegayon]

This can be done by doing the following:

1. Create a GitHub action that builds and uploads the container image to ghcr.io (here).
2. Once that's up and running, you can use the container: option of the GitHub action, like this.
3. An example of a dev environment for the agent is here. This is for an R package, but it should be pretty much the same (you just add container:).

[olivia-banks]

This can be done by doing the following:

1. Create a GitHub action that builds and uploads the container image to ghcr.io (here).
2. Once that's up and running, you can use the container: option of the GitHub action, like this.
3. An example of a dev environment for the agent is here. This is for an R package, but it should be pretty much the same (you just add container:).

Sounds good! Will the build step wait for GHCR or might it sometimes run on an outdated image if it hasn't been built yet?

[gvegayon]

Yes, it won't wait, so the first run will almost certainly fail. Since it is the first run, it should be alright. Worst case scenario, if we add dependencies or change something in the infrastructure, the AI agent will trip when loading the older version of the image, but it will be only once (and probably the agent will try to figure it out on its own).