A code increment is a coherent, reviewable set of changes that produces a stable intermediate version of a codebase. In migration work, its size is a deliberate tradeoff: tiny increments create process overhead, while oversized increments hide architectural mistakes and increase review difficulty. section 2.12
Treat each increment as an operational checkpoint:
- Define one migration step and commit it clearly on a dedicated branch. section 2.3.2
- Review the complete diff before accepting AI-generated changes; reject inferred future requirements, over-engineering, and architectural missteps. section 2.12 section 2.3.2
- Test and encapsulate every step so later changes remain comparable with the legacy application. section 2.12 section 2.3.2
- Keep sub-tasks separately committed when that improves traceability, even when they belong to one larger migration step. section 2.3.2
Layer-by-layer migration
A migration branch can contain one pull request for the data layer and another for the service layer. Cursor Plan Mode can propose the changes, show a unified diff, summarize the result as a conventional commit, and execute the commit only after review. section 2.3.2
