Refactor
Refactor code matching $ARGUMENTS across the codebase using parallel Snipper agents for maximum speed.
Process
Phase 1: Discovery
- •
Parse the refactor request from
$ARGUMENTS- •Identify: method name, component name, pattern, or code smell
- •Choose search strategy: Grep for code patterns, Glob for file patterns
- •
Find all affected files
- •Use Grep/Glob to search the codebase
- •Exclude: node_modules, .git, dist, build
- •If >15 files found, ask user to confirm or narrow scope
Phase 2: Create Instructions
- •
Create task folder:
.claude/tasks/refactor-{timestamp}/ - •
Create ONE instruction file at
instructions.md:
markdown
# Refactor: {title}
## Objective
{What to refactor - derived from $ARGUMENTS}
## Pattern to Find
{Exact code pattern or structure to locate}
## Transformation
{How to transform - be specific and adaptive}
## Examples
Before:
{current code}
After:
{refactored code}
## Constraints
- Only modify code matching the pattern
- Preserve all existing functionality
- Follow codebase conventions
Phase 3: Execute in Parallel
- •
Group files into batches of max 3 files each
- •
Launch Snipper agents in parallel - ALL batches in a SINGLE message:
code
Using instructions in .claude/tasks/refactor-{id}/instructions.md, refactor:
- {file_1}
- {file_2}
- {file_3}
Read instructions first, then apply the refactor to each file.
Phase 4: Verification
- •
Validate changes
- •Run
pnpm lint(if available) - •Run
pnpm tsc(if TypeScript)
- •Run
- •
Report summary
- •Files modified
- •Errors encountered
- •Next steps if needed
Example
User: /refactor rename getUserData to fetchUserProfile
- •Grep finds 12 files containing "getUserData"
- •Creates
.claude/tasks/refactor-1702489200/instructions.md - •Groups into 4 batches of 3 files
- •Launches 4 Snipper agents in parallel
- •Runs lint, reports: "Refactored 12 files"