Please report security issues privately to the SnappCloud team rather than opening a public issue. Use GitHub's private vulnerability reporting on this repository.

Include what you found, how to reproduce it, and what an attacker could reach with it. We will confirm receipt and keep you updated while we work on a fix.

The bot's entire purpose is to answer questions about clusters without widening anyone's access, so anything that breaks that is a security issue, not a bug:

• A caller receiving data from a namespace or cluster they are not authorized for — including through an aggregate, a metrics query, or a tool result that should have been filtered.
• A way to reach a tool that the configuration does not permit, or to call a write operation through a read-only server.
• Credentials appearing in an answer, a log line, or an error message.
• Anything that lets the model's output influence authorization, rather than authorization constraining what the model can see.

Bypassing a rate limit or a size budget is a bug; bypassing a namespace scope is a vulnerability.

The latest release is supported. Fixes are published as a new tag.