Hey everyone, I’ve been working on this problem where LLM agents can call tools, but there’s no real way to validate what they’re doing or control access.
Like, if you give Claude access to your file system or database, how do you make sure it doesn’t do something stupid?
You can check it out here: