It seems most people directly issue review commands to the LLM, whereas my approach is to use different skills to perform repeated cross-checks, and after each

It seems most people directly issue review commands to the LLM, whereas my approach is to use different skills to perform repeated cross-checks, and after each check I add corresponding tests to prevent regressions.