◇ Could this help me?
Yes, and this is the version of the loop idea worth keeping. The reason it works where the vaguer autonomous-loop pitches do not is that it has a real verifier: the review step compares the result against a plan that was written down first, so success is defined before the work rather than inferred afterwards, and the explicit no-scope-creep constraint on the build step is what keeps the comparison meaningful. Two things to hold onto: an output graded against its own checklist has a ceiling, since a loop cannot catch a problem the checklist never anticipated, so read the review yourself rather than waiting for it to declare itself done. Also discount the closing argument, which is that prominent people use loops so the approach must be right, because a list of names is not evidence and this technique stands perfectly well on its own logic. General practice.
Set up a three-stage loop for this project rather than working from a single prompt. Stage one, help me write a proper spec: interview me until the goal, the constraints, what is out of scope, and what done looks like are all written down explicitly. Stage two, implement exactly that spec and nothing else, and tell me if you think something outside it is needed rather than adding it. Stage three, review what was actually built against the spec from stage one, list every gap and every deviation, and feed those back into stage two. Loop stages two and three, but show me the review output each time rather than deciding on your own when it is finished.