A repository is a folder that remembers every change made to it, and because coding agents now work inside repos, knowing how to read one has become a way to supervise AI.
- Git keeps the memory. Each saved state is a commit, a snapshot with a note on what changed and why, so you can walk back through the project's history.
- A handful of moves cover most of it. Clone, branch, commit, merge, and the pull request, where changes are reviewed before they land.
- GitHub is the public square around git. Hosts like it add issues, reviews, forks, stars and releases on top of the tool itself.
- Read a new repo in order. The README first, then the license, then the issues and commit dates to see whether anyone still maintains it.
- AI made this everyone's business. Agents open pull requests for people to review, AI-assisted work needs versioning, and open-weight models are shared through repo-style hubs.
You have seen the word everywhere — "check the repo," "it's on GitHub," "the agent opened a pull request." This is the whole idea, from zero, in plain language.
1 · The folder with a memory
An ordinary folder knows only its present. A repository — a folder managed by git — remembers every state it has ever been in. Each saved state is a commit: a snapshot plus a note saying what changed and why. The chain of commits is the project's history, and you can walk it backward at will. That property — a recorded history you can go back through — is a large part of why so much software is built this way. It is not a guarantee: history can be deliberately rewritten, so treat it as a strong safety net rather than a promise that nothing is lost.
2 · The five moves that cover most of it
Clone — copy a repo, with its whole history, to your machine. Branch — split off a parallel timeline to try something without disturbing the main one. Commit — snapshot your progress with a message. Merge — weave a branch's changes back into the main line. Pull request (PR) — the polite version of merge: "here are my changes; someone review before they land." Most workflows you will meet are these five in costume.
3 · GitHub is the town, git is the road
Git works alone on your laptop. GitHub (and GitLab, Codeberg, and peers) add the public square: hosting, issues (the shared to-do and bug list), PR reviews with comments, forks (your own copy of someone else's repo, the seed of most open source), stars, and releases. When people say "open source lives on GitHub," they mean the town, not the road.
4 · How to read a repo you just found
Open the README first — the front door, saying what this is and how to run it. Then the LICENSE — the actual legal terms; "public" does not mean "free to use for anything." Then glance at issues and the commit dates: a repo with recent commits and answered issues is alive; years of silence means you are visiting a ruin, which can still be useful, but you should know which one you are in.
5 · Why AI made this everyone's business
Three reasons. Coding agents now clone repos, write code, and open pull requests — the PR review is where a human checks the machine's work, which makes repo literacy a supervision skill, not a developer skill. Second, the artifacts of AI-assisted building — your prompts, configs, and generated apps — need versioning exactly because generation is cheap and iteration is constant. Third, open-weight models themselves are distributed through repo-shaped hubs, licenses attached. The folder-with-a-memory became the standard container for human–machine collaboration.
If you work with coding agents, learn to read a pull request. That review is where a person checks the machine's work before it lands, which makes it a supervision skill rather than a developer one.
6 · Starting, in one afternoon
Make an account on a host, create a repo, put one text file in it, commit, edit, commit again, and look at the diff. The moment you see your own history laid out — what changed, when, why — the entire concept installs itself. Continue with the Craft when coding agents come up.
Part of the Stay Human record. Continue: how AI skills and tools work →