I made Buildermark to see exactly how much of my code is generated by my coding agents vs what I was writing by hand. For this project, it ended up being 364 agent conversations writing 94% of the code. Browse all of those here:
https://demo.buildermark.dev/ It matches edits by your coding agents (scans history in `~/.claude`, '`/.codex', etc.) with git commit diffs. It does formatting-agnostic matching to be robust against auto-formatting and reorganizing code. It's open source and local only, not even any analytics. Everything is automatic... just run the app, open the localhost:55022 frontend, and choose the project to import. It supports Claude Code, Codex, Gemini, and Cursor, so far. There's also a browser extension to import from cloud agents. I'm also working on a Team Server to aggregate results and make the project sustainable. |
No comments yet