@Yangtze_Seventh: A new paper is out, and the algorithm team is incredibly productive! Let me briefly introduce it. Right now on GitHhb you can search for all kinds of skills, each of which looks pretty good, and you star and bookmark them, but very few are actually put into use. Some skills, once downloaded, don't always work well and you run into a lot of pitfalls...

A new paper is out—the algorithm team has been incredibly prolific!

Let me give a brief introduction. Right now, on GitHhb, you can find all kinds of skills. Each one looks impressive, and you might star or bookmark it, but very few actually end up being used for real.

Some skills, once downloaded, don’t always work well in practice—you run into a lot of pitfalls.

The work in this paper crawled roughly 821K public skills and curated them down to 96,401. Then they tested the full capabilities across 3 benchmarks, 2 test frameworks, and 2 open backbones.

They improved all three benchmarks, with the largest combined gain of +7.5pp on SkillsBench.

We’ve already done the evaluation and filtering work for you—all you need to do is just use them, no need to try them out with a trial-and-error mindset.

If you’re academically interested, feel free to read the paper; if you want to use it directly, just head over to SkillHub!

EverMind (@evermind): Yet another paper.

This time, we asked a practical question: Do public SKILL.md files actually make agents better? The open skill ecosystem has a quality problem.

The hard part isn’t finding more skills. It’s knowing which ones are unique, useful, safe, permissively licensed,