The easiest way is to actually review the prompt input, put it through a LLM to catch the typical missing steps and automatically forward the result as review comment to the MR.

Let the LLM wars start!