Please report security problems privately, with Report a vulnerability on the repository's Security tab, not in a public issue. Say which skill, what an attacker could do, and how to reproduce it. We'll reply there.

A skill is instructions and small scripts that your agent runs with your permissions. What it can reach depends on what you've connected and allowed:

• MCP servers you connected. The skills read and change Whimsical boards through the Whimsical MCP, with your account's access, and read from the other services each skill's guide lists. They create new boards rather than edit existing ones, unless you name one.
• Files. Scripts read the files you point them to and write working files in a folder named after the task, in the folder your agent was started in.
• Network. Scripts only reach the hosts their skill's guide names. App Wireframe, for example, fetches from mobbin.com and Mobbin's image host, with size limits, and never follows a redirect elsewhere. Scripts that use packages download them through uv from PyPI the first time they run.

Nothing is collected or sent anywhere else.

Skills read content you didn't write: Mobbin pages and screenshots, emails, CRM records. The skills treat it as data to describe, never as instructions to follow. If you find content that makes a skill do something you didn't ask for, please report it as above.