In December 2025, I tried to sign in to another application through GitHub and found that my usual login no longer worked. I looked into it: GitHub had flagged my account. I hadn’t received an email explaining why.
I could still sign in to GitHub, push code and work. Other people, though, couldn’t see my profile or access the organizations I belonged to. Third-party application authorization stopped working, and I lost code search in my repositories. The account wasn’t completely blocked, but some of the features I relied on were gone.
In early 2026, I mostly stopped publishing code there. I was upset that the service had restricted my account without an explanation. I had no migration plan, and for several months I left the problem unresolved. I only started looking for alternatives after that pause.
Why this affected more than my code#
At first, I thought the priority was to avoid losing my repositories. But I also used GitHub to sign in to Tailscale, which ran the private network connecting my machines and services. The GitHub restriction cut off my usual way into the network’s admin console. Even if the devices could still communicate, I could no longer manage them through that account.
I never recovered access to that Tailscale account. Instead, I set up Headscale on my own server—a self-hosted control server for Tailscale clients—and moved all my devices over to it. Managing the network no longer depended on signing in through GitHub. Maintaining the Headscale server was now my responsibility, though.
That made the connections between my services clear: an account restriction on a code hosting platform had forced me to change how I accessed my own machines.
Looking for alternatives#
First, I tried Radicle, a system for decentralized Git repository hosting. I ran nodes on my machines and tested it with one of my side projects. I liked the idea, but everyday use meant dealing with another tool’s network, rules and maintenance. I wanted to get back to working on my code, without taking on another infrastructure project.
Then I started using Hermes, an AI agent I later worked with to build a Hugo blog and publish it on Codeberg Pages. The blog went live in June 2026. I’ve written separately about Hermes and my setup.
After the blog, Hermes and I worked on my Astro site. We deployed it directly to Cloudflare, without making GitHub a required part of publishing. In July, I moved the blog there too. Publishing a website and hosting its Git repository had become separate questions. I no longer needed to recover my old GitHub account to publish.
I also tried to resolve the restriction itself. On July 27, I contacted GitHub Support, described the symptoms and offered to verify my identity. I explained the Tailscale situation and my logins from Ukraine and Sweden. I suspected an automated check might have treated them as suspicious activity, but I have no evidence that this was the cause. I received an automated acknowledgment of my request. In October, I followed up. I still haven’t received a substantive reply.
By November, I had put my old GitLab account back to use after leaving it unused. On November 3, I created a second GitHub account, phantom-di. By then, I wasn’t looking for a single replacement for everything GitHub offered. Headscale handled my network, and Cloudflare hosted my site and blog. What remained was to move the code I needed and set up copying between platforms.
What needed to move#
The old account had 62 repositories, 53 of them forks: study assignments from 2022, someone else’s Django template and someone else’s Odoo compose file. I wasn’t going to copy everything indiscriminately. A substantial part of my own work was only stored locally anyway.
I scanned my disks and found 91 repositories. From those local projects and the repositories on the old account, I selected 23. I finished moving them on November 4: all 23 were on both the new GitHub account and GitLab.
GitHub took some persistence. I was creating and uploading repositories too quickly: after nine in a row, I got “too many repositories, too quickly” and a request to wait about an hour. That was what happened during my migration; I didn’t establish the exact rate-limit threshold. Some repositories went through on the second attempt. One repository took four attempts, with ten-minute pauses. All 23 went through on GitLab on the first try.
Why I kept the original commit history#
I wanted my old commits to appear under the new account, so I initially considered changing the email addresses throughout Git history. But a commit’s GPG signature is tied to its contents, including the author and committer information. Rewriting those fields changes the SHA, the commit’s identifier. The original signature won’t verify the new object; that would need a new signature.
I checked 150 commits in one working repository. All had valid signatures. I wasn’t going to rewrite that history just to make GitHub display a different account name.
GitHub commit attribution and cryptographic signatures are separate mechanisms. GitHub links commits to accounts by email address. If I can add the old address to the new account, commits made with that address may be attributed to it without rewriting history. GitHub’s noreply addresses have restrictions, and GPG signature verification has its own requirements.
I left the history intact and configured new commits for the new account. I still need to check whether I can move the old email association before deleting the old profile.
Technical details: commit attribution and GPG signature verification.
How the GitLab mirror works#
I kept GitHub as the main place I work with my code and made GitLab a second host for copies of the repositories. GitLab Free supports push mirroring, which sends changes from GitLab to another host. I need the reverse: regularly copying changes from GitHub into GitLab. Built-in pull mirroring is a paid feature, so I wrote my own script.
For repositories already present on both platforms, the script uses separate bare clones: copies of the Git repositories without checked-out working files. It fetches changes from GitHub and pushes branches and tags to GitLab, including deletions. It doesn’t touch my working directories. It’s scheduled to run every six hours.
There are limits to this setup. A mirror is not a backup: deleting a branch at the source can delete it from GitLab too. The schedule depends on the machine running the script. When it’s off or asleep, nothing syncs. New repositories don’t appear on the second platform automatically either; I have to create them there first.
I could have used GitHub Actions to sync after each push. That would remove the need to keep my laptop on, but I’d have to configure credentials and permissions to push to GitLab. For now, I’ve kept the local script with those limitations.
Documentation: GitLab repository mirroring.
Why I stayed on GitHub#
I wanted to leave. To me, GitHub had gotten worse since Microsoft bought it. But other people’s projects, forks, integrations and familiar links were still there. Leaving completely would mean changing that part of my work too.
The second account solves a specific problem: I can use GitHub again without the old profile’s restrictions. This time, I no longer rely on it for everything else. My Headscale server manages the private network, Cloudflare handles the site and blog, and my code is also on GitLab and stored locally.
That doesn’t make me completely independent. Headscale needs maintenance, Cloudflare is still an external provider, and two Git hosts aren’t a substitute for tested backups. But restricting one GitHub account should no longer automatically lock me out of managing my network or publishing my site.
What’s still left to do#
I plan to delete the old account after checking the migration. Preserving Git history isn’t enough: I also need to check email addresses, commit attribution, issues, pull requests, wikis and links. Not all of that moves with the repository.
There’s another complication: deleting a restricted account through its settings may be unavailable. Even that step could require contacting support. GitHub documents this limitation.
As of November 14, 2026, I still haven’t received a substantive reply. But my ability to keep working no longer depends on a reply. I use GitHub where it’s useful to me, while managing my network and publishing my site no longer require signing in through that account.
