PUBLIC METHOD · NO PAID BOOSTS
Popularity, momentum, recency, quality, and safety are different questions. SkillSignal keeps them separate and shows the evidence date behind each answer.
Leaderboard entries carry captured skills.sh rank, installs, and weekly history. Editorial entries add source-grounded Skills without inventing popularity. An editorial Skill can rank on quality or topic fit, but never receives an all-time or weekly install value unless a source provides one.
All-time ranks cumulative installs. Trending uses only the latest observation in the eight-week source series. Rising calculates week-over-week growth and requires at least 25 installs in the previous week. Newly indexed sorts by the date SkillSignal checked the source—not the author's publication date.
The 100-point quality signal uses published weights: source trust 25%, maintenance 20%, documentation 20%, compatibility 15%, community evidence 10%, and on-site trend 10%. The same formula applies to every Skill and paid placement cannot change it.
Source checked, pinned revision, profile updated, and runtime reviewed are separate fields. A pinned commit proves which source was read; it does not prove the Skill is safe or behaves correctly. Runtime review remains explicitly unverified until an independent execution record exists.
A refresh first creates a review candidate and diff report. It accepts current weekly telemetry when supplied, preserves the previous series only when the source omits it, and appends dated rank and install snapshots. The rolling archive keeps up to 26 snapshots for future movement comparisons.
Install counts can reflect age, distribution, brand reach, duplicate listings, or curiosity. Metadata scores cannot replace source review, controlled testing, permission checks, or organizational approval. Treat every ranking as a discovery aid, not a security certification or universal recommendation.
NEXT STEP
Choose the ranking that matches the question you are actually asking.