I'm a solo engineer with a systems-engineering background. I build Phionyx, an open-source runtime that makes AI agents abstain when they are outside their competence and leave a signed, independently-verifiable record of what they decided. It ships as an MCP server and governs its own development under its own fail-closed gates. My current focus is making that evidence layer cryptographically verifiable, so an agent's decisions can be reconstructed and checked by anyone.