2 pointsby dreynow4 hours ago2 comments
  • AlphaTheGoat4 hours ago
    Love the scoped, cryptographically verified tokens. Way safer than just hoping an AI won't rm -rf /.
    • dreynow4 hours ago
      Thanks! That's exactly the problem - "hoping" isn't a security model. The token either grants the scope or it doesn't. No policy engine, no prompt engineering, just Ed25519 math.

      If you want to try it: pip install kanoniv-auth && kanoniv-auth init && kanoniv-auth delegate -s code.edit,test.run -t 4h

  • dreynow4 hours ago
    [dead]