Migrating a Hand-Tuned CoT Prompt to a Reasoning Model

Your team has a well-tuned prompt for a code-review-comment generator: it includes a persona ("You are a senior staff engineer reviewing this diff"), an explicit "think step by step, listing each potential issue before writing your final comment" instruction, and 4 few-shot examples of the target format. It runs on a non-reasoning model today at 82% "comment is accurate and non-redundant" on your golden set. A teammate wants to migrate it to a reasoning model with an extended-thinking budget, and proposes keeping the prompt exactly as- is (persona, step-by-step instruction, and few-shot examples) to "keep the reliable parts."

1. Which parts of the existing prompt are likely to become redundant or counterproductive on a reasoning model, and why, specifically?
2. Which parts are likely to still earn their place, and why?
3. Describe the actual migration process you'd follow, including what you would and would not assume without measuring.