The repository was the easy part

GitHub spent almost eight hours on Monday returning errors across git operations, pull requests, Actions, Pages and Copilot. By the afternoon an Ask HN thread about alternatives to GitHub was on the front page with a few hundred comments.

What held my attention was the service list rather than the downtime. Hosting, continuous integration, code review and the coding assistant now sit in one account and one failure domain, and the assistant is the newest thing to have moved in there. GitHub’s own postmortem names one detail that makes the point sharper than I expected: VS Code’s retry logic amplified Copilot token traffic from a normal 7-9K requests per second to 70-100K during the outage. The assistant did not just share the failure domain. Its retry traffic amplified the incident. The thread mostly discussed moving repositories, which is the one piece that was always portable, because a git remote is a git remote and a mirror takes an afternoon.

The parts that are genuinely hard to leave are the workflow files, the review history and whatever the assistant has been granted access to. Nobody was shopping for those on Monday. That is why the answer to the thread is usually “not really”.