Trust, in numbers

We check everything. Here is the proof.

A skill is just text, and text can hide instructions. So a human reads every one before it goes live. These numbers grow as our nightly review runs.

8,422+
security and privacy checks run
60
Skills published
each read line by line
6
Rejected as not safe
publicly documented
86/100
Average scan score
of what we publish
7
Categories curated
small on purpose

What a check actually means

Before we list any skill, a person reads the whole thing and checks: Can it be tricked into doing something you did not ask for? Does it quietly collect or send your data? Does it ask for more access than it needs? Only the ones that pass get listed. The ones that fail on safety grounds do not just quietly disappear. We explain what we found.

Copied to clipboard. Go back to ChatGPT or Claude and paste it to teach the skill.