Refactoring

:DwightRefactor handles large-scale refactoring that touches a feature and all its dependents. It finds every file that imports the feature, builds a comprehensive context, decomposes the refactoring into sub-tasks, and executes them through the Auto pipeline.

How It Works

1. Pick a feature — :DwightRefactor auth selects the feature to refactor. Without arguments, a picker opens.
2. Importer scan — Dwight scans the entire project for files that import or reference the feature's files.
3. Context assembly — builds a context package: full source of feature files + signatures and import lines from dependent files.
4. Decomposition — the AI generates up to 8 ordered sub-tasks: structural changes first, then importer updates, then cleanup.
5. Review — the plan is shown for your approval before execution.
6. Execution — sub-tasks run through the Auto pipeline with verification gates and git checkpoints.

:DwightRefactor                          " Feature picker
:DwightRefactor auth                     " Refactor the auth feature
:DwightRefactor auth Extract JWT into a separate module

Importer Analysis

Before refactoring, Dwight scans project files (.go, .py, .js, .ts, .rs, .rb, .java, .lua, and more) for import statements that reference the feature's files. This ensures the refactoring plan includes dependent code that needs updating.

The scan uses pattern matching on:

• import "path" and from "path" import (Python, JS/TS)
• require("path") (Lua, Node)
• Package-level imports (Go, Java)
• File basename matching

Importers appear in the context as signatures and usage lines — enough for the AI to understand the dependency without including full file contents.

Sub-Task Ordering

The decomposition follows a strict order:

1. Structural changes — rename, extract, or restructure within the feature
2. Importer updates — update dependent files to match the new structure
3. Cleanup — remove dead code, update documentation, consolidate

Each sub-task is small enough for one agent session and specifies which files to create, edit, or delete.

Tips

• Include a prompt for better results. :DwightRefactor auth Extract the JWT logic into a separate module gives the AI a clear goal. Without a prompt, it refactors based on code quality heuristics.
• Review the decomposition carefully. The plan preview shows every sub-task before execution. Remove or reorder tasks if needed.
• Commit before refactoring. The Auto pipeline creates git checkpoints, but a clean starting point means you can always roll back entirely.
• Use for cross-cutting changes. Renaming a shared type, changing an API signature, or splitting a module — these are cases where importer analysis matters.

Commands

See Also

• Auto Mode -- executes the refactoring sub-tasks with verification gates
• Feature Management -- browse and preview features before refactoring
• Inline Editing -- the /refactor mode for smaller, scoped refactoring on a selection