agentic-coding-workflow

To prevent scope creep during multi-file code changes, require an explicit intent block listing the goal, scoped file list, and approach before editing, then enforce scope guards to avoid collateral refactors and preserve existing behavior.

An intent-first coding workflow requires agents to present a concise block containing the goal, affected files, approach, and done criteria, securing user confirmation before any code modifications begin to keep edits reviewable.

Make AI-generated code refactors reviewable by enforcing task sizing rules for single-pass or split-pass edits, verifying dependencies exist, and producing a post-change summary detailing what changed, how to verify, and flagged assumptions.

Surface assumptions for database or auth-impacting changes whenever the task touches multiple files or affects production behavior, ensuring design choices are explicitly declared and verified before modifications to reduce regression risks.

Yes, the agent workflow supports chunking large refactors by applying task sizing rules based on touched-file counts, splitting large changes into smaller, reviewable passes to maintain operational discipline and reduce unreviewable diffs.

AI agents invent dependencies during automated code edits when operational rules are absent, requiring explicit scope guards and dependency verification checks before modifications to ensure changes remain within declared scope.