Exploring Git History

Drive AI agents through a million lines of code you did not write. Recon, surgical navigation, codemods, characterization tests, and a toolkit of four reusable skills you keep.
I know the size and roughly the languages. Now I want the history. What has actually gone into this project, and where is the team pointed?
I could sit here running git log incantations. Instead I open Claude Code with Opus 5, because this is a thinking job, and I ask it to explore the history, tell me how old the repo is, how many commits, and what the last year has been about. Run the commands, then write me a report.
While it works, I open a second window and check /usage. Session cost zero. That is my baseline, and I want it, because the thing I am going to watch all course is how much Claude is talking to me. Code is cheap in tokens. Code is terse. Documents and long chat sessions are where the money goes.
What comes back:
- First commit, 2015, Erich Gamma. "Hello, code." The Gang of Four guy.
- 162,000 commits. 3,000 authors, which is far more than the VS Code team, so they are taking outside code.
- Commits per year are consistent, but this year is the busiest ever. 2024 was the quietest. January through August of this year already beat every full year before it.
Does that mean they are working hard, or that they are using AI? I would say AI.
Then the headline. Over the last year VS Code stopped being an editor with an AI panel and became an agent host. File work moved into the chat window. A new peer to Workbench appeared, a second application shell. There is now a first-class service that spawns, supervises, and restarts coding agents, some of them on remote servers.
They are chasing Cursor. That is exactly why I am doing my own thing, so this reads as validation.
Two takeaways.
One, I now know what the team cares about, and it is not the editor.
Two, doing this by hand would have taken you a week and a pile of emails. I did it over one coffee.
Last move: I have Claude write it down. A docs directory in the root, with decisions and history inside it, and the whole history summary plus a note on where they seem to be heading. That is the project record. I keep docs on main and update it whenever I finish work, so the next session has somewhere to start.
Drive AI agents through a million lines of code you did not write. Recon, surgical navigation, codemods, characterization tests, and a toolkit of four reusable skills you keep.