With tools like Moltbot(https://clawd.bot/) gaining traction, I got thinking about AI agents executing shell commands with full permissions by default.

I built veto to add a safety layer:

1.Intercepts commands before execution

2.Risk-based authentication (Touch ID for critical, simple confirm for low)

3.Custom rules for project-specific patterns

4.Full audit trail

Currently supports Claude Code (via hooks). Looking into Moltbot integration but it doesn't expose hooks yet — if anyone has ideas, I'm all ears.

Feedback welcome.

https://github.com/runkids/veto