# How we audit

Only security-audited skills are listed. Before a skill goes up, we check every file in it. When we are not sure a skill is safe, we leave it out.

## Every file is scanned

A scanner reads every file of a skill and marks anything that reaches the internet, runs commands, installs software, reads passwords or keys, starts itself when your computer starts, or hides what it does.

## An AI model checks every skill

For a skill with scripts, or with anything the scanner marked, the model reads every script and every instruction the agent follows in full, line by line.

## Code only from well-known makers

A skill that runs code on your computer, or installs a tool, is listed only when its maker is a well-known company, such as Anthropic, OpenAI or Google.

## What we turn away

A skill that sends your keys, passwords or files somewhere its job does not need. A skill that holds someone's private details. A skill whose links pay its maker a commission, or that advertises its maker's paid product. And any skill whose code we could not read in full.

## Checked again when it changes

A new version is checked before it replaces the one listed. Until then, you get the version we checked.

## Workotron's own workflows

Workotron workflows are made by Workotron. Our team reviews and tests each one before it ships.
