A small, readable self-improving agent loop. Observe, Orient, Decide, Act, and the second A, Adjust: the phase that makes an agent get smarter across runs instead of only faster within one.