defensive-development

AI coding mistakes during refactoring are prevented by enforcing a verification-first protocol that requires reading, mapping, listing, presenting, waiting, executing, and verifying before applying edits or deletions.

Safely renaming functions across an unfamiliar codebase requires grepping all references and presenting a comprehensive list of dependencies before executing the refactor, ensuring no usages are missed.

The read-verify protocol works by enforcing a strict sequence: read the code, map dependencies, list proposed changes, present them for confirmation, wait for approval, execute edits, and verify the final state.

Adding logs before changing code is a core defensive debugging practice that verifies actual program behavior and runtime property values before executing any modifications or refactors.

API properties and constructors are verified through API discovery practices that require agents to inspect getters, setters, and constructors in the codebase before attempting to integrate or call them.

Line count thresholds for deleting code require explicit confirmation based on the volume of lines targeted, preventing accidental bulk deletions and ensuring safe refactoring outcomes.