Security researcher launches AgentsAegis, a trap-based training tool that caught 75% of developers blindly approving dangerous AI-suggested commands

Software engineer with 14 years experience created AgentsAegis after realizing he was approving Claude Code permission requests without reading them first

The tool works as a Go proxy between AI assistants and APIs, intercepting responses and injecting realistic trap commands to catch users approving blindly

A 2-minute quiz version (free, no signup required) has already attracted significant participation with 80+ responses showing high failure rates

Open-source approach prioritized over closed-source to build user trust, addressing risks like accidental production destruction via Terraform commands