Five sub-scores, one combined score, recomputed daily.
Usage, installs, GitHub stars, views and saves, each log-scaled so one huge skill doesn't flatten the rest.
Trust-weighted upvotes, saves, comments and ratings, plus how often people come back to a skill.
24-hour, 7-day and 30-day momentum, volume-weighted: going from 3 to 9 uses counts far less than 3,000 to 9,000.
Maintenance recency, documentation depth, completeness of the listing, ratings (shrunk toward a prior until there are enough) and real output examples.
A verified source, a verified maker, the security scan result, repository reputation and maintenance history.
Each sub-score is 0–100. They’re combined with a weighted formula we tune over time and don’t publish, so it can’t be gamed to the decimal. Security acts as a gate, not a weight: skills with high-risk findings are heavily discounted and blocked skills score zero regardless of popularity.
Trending favours momentum, anchored to quality and trust so a burst of suspicious traffic can’t push junk to the top. Top · today / week / month ranks by real activity in that window: uses, installs, votes and saves. All time is the LaunchFable score itself. Rising is pure week-over-week growth.
You can’t buy placement. What moves a skill: real output previews, a clear install path, a maintained source, and people actually using it.
