SSH, Actually 100% signal, 0% noise.

Git over SSH

concept
depth 3/3 − deeper + surface

With GitHub, GitLab, and similar services, SSH is used as authentication, not login. You never get a shell—you only prove identity, the service decides what you can access based on that identity, and each git operation is a separate authentication.

Two models, one protocol.

The traditional model (server login): you get a shell, you run commands, you browse the filesystem, the connection stays open.

The git hosting model (authentication only): no shell, ever. You prove identity; the service does the rest.

When you run:

git clone [email protected]:user/repo.git

behind the scenes:

  1. Git opens an SSH connection to github.com
  2. You prove identity using your SSH key
  3. GitHub checks: "Does this key have access to this repo?"
  4. If yes → sends repository data
  5. Connection closes

No shell. No commands. Just cryptographic proof of identity.

Why this model is powerful:

  • No passwords ever transmitted
  • No credentials to steal from GitHub's servers
  • Phishing-resistant—you're not typing anything

Same cryptography—different purpose. This is why learning SSH well matters even if you never manage servers: every time you push code, you're using SSH authentication.

The model has a distinctive fingerprint: because the service identifies you by which key you presented, the username in the URL is always just git—your actual identity comes from the key, not the username. And because authentication is a signature, not a typed secret, this is the same phishing-resistant shape that FIDO2 brought to the web—modern setups go one step further and back the git key with a hardware device.