GitHub: the operating system of shipping

Git is a version-control program on your machine. GitHub is a company hosting remotes with a workflow product on top. The confusion between them causes real mistakes, and untangling it takes ten minutes.

The mental model behind running an entire shipping operation on GitHub: one repo per concern, private by default with a deliberate public surface, and the repo as durable memory rather than a code drop.

Auth scopes gate what gh can push, pr/run/repo are the daily verbs, and gh api plus --jq answers every question no subcommand covers.

Branch-per-task naming conventions, why main never takes a direct commit, and why a scheduled-content branch that lives for days is doing its job rather than going stale.

Why every change goes through a PR even when you're the only committer, what squash merge quietly does to your commit history, and the concurrency setting that can cancel a sibling deploy mid-flight.

How to build a fast GitHub Actions gate that runs typecheck and build on every push and PR, then use that same mechanism to enforce content and policy rules automatically.

How two sibling GitHub Actions workflows deploy a static site and a PHP backend to the same shared host without stepping on each other.

Cron and workflow_dispatch triggers, concurrency groups that queue instead of collide, and two real repos where the right call was to keep a job off GitHub Actions entirely.

A single global core.hooksPath repo that blocks name leaks, ties commits to roadmap ticks, and treats bypass as a documented token rather than a silent escape hatch.

Treating GitHub's REST API as a queryable database for your own sites and tools, with the workshop XP feature and this week's repo-inventory sweeps as the worked examples.

GitHub Actions' buried minutes model, the mid-month block nobody warns you about, and the three-command diagnosis that caught a permanently-failing suite burning 980 minutes.

The recovery repertoire for solo shipping on GitHub: rejected pushes, orphaned PRs holding a stale lock, transient workflow failures, quota-blocked Actions, and the devlog entry that turns each one into a lesson you don't relearn.

Real developers spend more time recovering than memorising. Five drills — detached HEAD, the deleted branch, the bad force-push, the merge conflict, the lost commit — and the Recovery Ladder that turns panic into procedure.