Show HN: Claude's Code – tracking the 19M+ commits generated by Claude on GitHub

I was curious about the amount of code on GitHub that is generated with Claude Code, and this is my attempt at finding that answer.

Spoiler alert: It's a lot - around 19M commits by my count

In a nutshell it is a dashboard that presents some basic, and hopefully, interesting stats about commits signed by Claude Code on GitHub - in public repos.

Not all commits are signed (via the author field, or a commit "trailer"), and many repos are private, which means Claude's reach is probably wider than what you see here. But I think it's enough to see the spread and learn a bit about how it's used.

Technology wise, it's a pretty basic Next.js app with Recharts for graphing and PostgreSQL for the DB. I started off with using BigQuery because I estimated I would need the analytical scale, but I eventually pivoted to Postgres because the small writes, and frequent reads for deduplication became too expensive.

The ingestion/backfill job is the more interesting part since I went from severely under-engineering it (start smol and all that) to ending up with a bare-bones, but capable, ETL pipeline.

Primarily, the challenge to overcome in reading the data, was GitHub's rate limits - both on their search API, and on their GraphQL API. On search it's 30 req/min, on GraphQL it's 5000 req/hour - per access token. Because of this difference, and response time differences, I split the work:

  1. There is a batch of search workers that write basic commit info to a table - paging and splitting to find as many of the commits as we can.

  2. Enrichment workers read from that table and fill in some of the info that we can't see in the search. Lines added/deleted, and repo information, is added this way.

There is a bit of lag in reading these commits currently, and it is still pulling historical commits, which is why the most recent dates are a bit low on commits, and why some repos don't yet have a language set yet.

I wouldn't say it's 100% done - I want to improve the ingestion still, and I think there is more I can extract from the data - but I have definitely enjoyed looking at what I have so far.

Let me know if you have an idea for what I can add to the dashboard, or can think of something else I should also be reading.

For some more info on my methodology and the evolution of the backfill job, head to the About page. :-)