29/10/2026
When you ask an AI to change your code, it usually writes it, tells you "done, it works!", and you find the problems later. That's like letting a student grade their own homework.
OMEGA-LOOP adds one rule: whoever writes the code can never judge it.
Here's what happens instead:
° It makes a plan and waits for your approval
° One AI writes small, targeted changes
° A different, independent AI reviews only the diff and its job is to find flaws
° Build and tests must pass before anything is committed
It runs inside Claude Code. It uses OpenAI's Codex as the reviewer if you have it, and a fresh, separate Claude session if you don't. If one runs out of tokens, the other takes over. And if no reviewer answers, the result is "rejected", never approved by mistake.
👉 github.com/priore/omega-loop
