GitHub

Your repositories' issues, as tasks in their project.

Status: syncing is up and running and has been tested with real issues. The agent needs some setup in each repository (see below).

The GitHub screen

In GitHub (in the sidebar or, on your phone, under Go to in your account sheet) you have everything in one place:

The design also has a switch to pause a repository. It doesn't exist yet: to stop syncing a repository, unlink it.

Besides the GitHub screen, in Settings › GitHub you'll see your projects, each with a picker for its repository. Open it and choose one from the list. A notification with Undo confirms it. To unlink it, choose No repository.

The list shows your private repositories that syncing has access to, plus all your public repositories. When you pick a public one, Muninn quickly checks that syncing can use it; if it can't, it tells you and doesn't link it.

Each repository goes with just one project: if it's already linked to another, it shows up grayed out and tells you which one ("in Huginn").

Can't find the repository you're looking for, or it says it has no access? Syncing can only use the repositories you gave it access to when you set it up. Give it access to that repository on GitHub and pick it again.

If the list can't load (for example, in demo mode), you'll see a field instead: type user/repository or paste the GitHub address and press Enter.

On that same screen you can see whether the GitHub module is turned on and whether any issue has arrived yet.

What syncing does

Tasks that come from GitHub

For the agent

You can ask an agent (Claude Code) to solve an issue for you: move its task to the agent column and it works in the repository, opens a pull request, and the task moves forward on its own.

Status: brand new. It needs some setup in the GitHub repository that can't be done from Muninn yet (installing the Claude app, a key and a workflow file). Until that's done, moving a task to the column only adds a label to the issue.

Each time you send a task to the agent, it uses up a GitHub Actions run and some Claude usage. Only send issues you've read: the agent reads the whole issue, and in a public repository anyone can write in it.