Show HN: Deep Skill Finder – Find agent skills using real execution benchmarks

Deep Skill Finder appeared on Hacker News as a Show HN under the AI Agents lane, pointing at https://www.meyo.life/skill. The listing framed the product as a way to find agent skills using real execution benchmarks rather than catalog marketing copy. At capture it had 3 points and 0 comments on item 49030664, so the thread was still early and had no public discussion to lean on.

The product pitch centers on ranking or surfacing agent skills from measured run behavior instead of name match or vendor claims alone. That implies a discovery loop where skill candidates are exercised in real agent runs, results are scored, and those scores drive what builders see first. Without published architecture notes in the submission itself, the usable signal is the evaluation surface: execution outcomes as the primary filter for skill selection.

For engineers shipping agent stacks, skill discovery is often the weak link. Tool and skill registries grow faster than any team can trial by hand, and a skill that looks right in a README can fail under real tool use, latency, or failure modes. A finder that prioritizes benchmarked execution gives a concrete way to shortlist integrations before wiring them into production agents or evaluation harnesses.

The competitive context is a crowded agent-skill and plugin ecosystem where many listings compete on description quality and ecosystem affiliation. Most discovery still works like a marketplace search: title, tags, and popularity. A benchmark-first index is a different ranking thesis—closer to CI and eval dashboards than to app stores—and that is the wedge Deep Skill Finder is testing against description-led directories.

What to watch next is whether the site exposes how benchmarks are defined, which agents and tasks they run against, and how results stay comparable as new skills land. On the HN side, item 49030664 is still at 3 points with no comments, so the useful follow-up is whether early users report that execution scores match their own agent runs, and whether the finder stays useful once more skills and more noisy tasks enter the pool.