How Skillshelf grades skills
Every grade is computed at build time from two public GitHub signals - maintenance (recency of commits) and adoption (stars) - on a 0-100 scale. There are no paid rankings and no manual thumb on the scale. Archived or abandoned repos drop to a D even if they have thousands of stars.
Most skill directories either show no quality signal at all, or stamp nearly everything with the same green “A”. Neither helps you decide what to actually install. Skillshelf grades every listing on a transparent 0-100 scale, and shows you the exact breakdown on each skill’s page. Here is precisely how it works.
The two signals we score
A grade is built from two signals, both pulled from the public GitHub repository at build time:
| Signal | Max points | What it measures |
|---|---|---|
| Maintenance | 55 | How recently the repo was updated - the single best signal for abandonware |
| Adoption | 45 | How widely the skill is used, measured by stars |
Together they sum to 100. A skill that is actively maintained and widely adopted scores 100. Nothing is hidden behind connector-only or vendor-only bonuses, so a perfect community skill really does reach a perfect score.
Maintenance (up to 55)
We are deliberately generous to stable projects - a mature, “finished” skill does not need a commit every week to be healthy. We only dock hard once a repo goes genuinely quiet.
| Last updated | Points |
|---|---|
| Within the last ~4 months | 55 |
| Within the past year | 42 |
| 1-2 years ago | 22 |
| Over 2 years ago | 6 |
| Repo archived | 0 |
Adoption (up to 45)
Stars are an imperfect but honest proxy for “do other people rely on this.” We bucket them so a 1,000-star repo and a 100,000-star repo are clearly distinguished.
| Stars | Points |
|---|---|
| 50,000+ | 45 |
| 20,000+ | 41 |
| 7,000+ | 36 |
| 2,000+ | 30 |
| 500+ | 22 |
| 100+ | 14 |
| under 100 | 3-8 |
From score to letter grade
| Grade | Score |
|---|---|
| A | 80-100 |
| B | 60-79 |
| C | 40-59 |
| D | below 40 |
Because both signals matter, a repo with huge adoption but no recent maintenance is pulled down. That is the whole point: an archived repo with 2,000 stars correctly grades a D, because however popular it once was, it is no longer maintained. A directory that grades everything an A cannot tell you that.
What we do not do
- No paid rankings. You cannot buy a grade. Sponsored placements, if and when they exist, are labelled as sponsored.
- No manual thumb on the scale. The grade is computed, not hand-assigned. “Official” listings get a badge for context, not free points.
- No fake numbers. If a repo has not been synced, it shows as Unrated rather than a guessed grade.
How fresh the grades are
Signals are re-fetched on a schedule and baked into the static page, so the grade you see reflects the repo’s real state as of the last sync - the date is shown on every listing. There is no live database call slowing the page down; the number is just always current.
That is the entire methodology. No black box - you can read every grade off the breakdown on the skill’s own page.
Frequently asked
Can a skill pay to rank higher or get a better grade?
No. Grades are computed purely from public GitHub signals at build time. Sponsored placements, if any, are labelled as sponsored and never disguised as a quality grade.
Why is a popular skill only graded a B or C?
Adoption is only part of the score. If a repo has many stars but has not been updated in a long time, its maintenance score drops, which pulls the grade down. A high-star, long-abandoned repo can land a C or D.
What does "Unrated" mean?
It means we have not been able to sync that repo's public signals yet, so we show no grade rather than guessing. We never fabricate a score.
Do you grade security?
Not as the headline grade. Some directories show a security-scan pass that nearly everything gets. We grade maintenance and adoption, and surface security context (like whether a connector uses authentication) separately so it is not lost in an inflated badge.