Report it privately. Please do not open a public issue for a security problem: an issue is visible to everyone the moment you file it, including whoever would use the bug.

Include what you did, what happened, and what you expected. A proof of concept helps. Reporters are credited in the fix notes unless they would rather not be.

Nothing but your own code, on your own machine. Skill Builder reads the repository you point it at and writes its results under .agents/skill/ in that repository. It connects to no service of ours, and nothing leaves your machine except what your AI tool sends to its own model.

No telemetry, no backend of ours, no account to create.

Your code, only through a pull request. Fixes go on a branch and arrive as a pull request that lists each change with the finding it closes. Nothing is merged or pushed to a live system for you.

Code, comments, READMEs and agent instruction files in the repository you scan were written by someone else, and some carry text aimed at AI agents. The skills treat all of it as data to report on, never as instructions.

Read, run and pull apart anything here. Nobody but the maintainer can change this repository, so nothing you do while investigating puts it at risk.

The care is owed to the people and services the plugin talks to, not to the code. When testing, use your own accounts and your own content. If a test could affect anyone but you, stop and send a private report first.

Research done in that spirit is welcome, and nothing here is a trap.

The latest released version gets fixes. Given the size of this project, older versions do not.