DeepSeek Harness Crosses 100,000 Stars

The counter on the stats page rounds to a number that used to belong to a different class of project: DeepSeek Harness has passed 100,000 GitHub stars.

The shape of the run

Four days ago the same snapshot read about 88,000. Today’s build-time fetch reads past 108,000. More than twenty thousand stars inside four days is not a long tail accumulating — that is a project still in its vertical phase, and the plugin ecosystem described in our ecosystem survey is growing on the same curve.

For comparison, that is faster than the early adoption curves of the frameworks this one is routinely discussed alongside. Whether the pace holds matters less than what it already proves: there was a large population waiting for exactly this shape of tool.

What the milestone does not mean

Two caveats worth pinning to any celebration:

It does not mean stable. The README is explicit — developer preview, iterating rapidly, compatibility-breaking changes promised. Teams adopting now should pin versions and expect churn.

It does not mean versioned. As of this writing the repository has no GitHub releases or tags at all; the only version signal is the npm package @deepseek-ai/dsh, currently shipping release candidates in the 0.1.0 line. Anyone scripting against “latest” is really scripting against an rc.

How we count

A note on methodology, because this site’s whole premise is verifiability: star counts on DSHSeek are never typed by hand. The build fetches them from the GitHub API into a committed snapshot (refreshed daily at 03:00 UTC), and every page — home, map, stats — renders from that snapshot. When the number you see here disagrees with the badge on GitHub, the build is stale, not the claim.

The map’s 53 verified entries are the same idea applied to links: every one of them answered 200 at the last build.