Nice project, but saying "Run AI agents in production without the risk" isn't quite accurate.

Even if some tool makes it impossible for an AI agent to delete things in a way that isn't recoverable, there are other risks such as data exfiltration that need to be managed separately.