refactor

name	refactor
description	Scan the project for refactoring opportunities, then fix them. No behavior/visual changes.

Refactor

Identify and fix code quality issues across the project without changing behavior, visuals, or public API.

Constraints

• NEVER change behavior, visuals, or public API
• NEVER add new dependencies
• NEVER add features or functionality
• Run pnpm check after all changes to verify no type errors were introduced

Phase 1: Scan (read-only, use Agents for token efficiency)

Launch an Explore agent to scan src/ for issues in these categories:

1. Dead code: unused imports, unreachable branches, commented-out code, unused exports
2. Oversized components: .svelte files > 300 lines that mix unrelated concerns and could be split (skip files with a // NOTE: Large file by design comment — these have been reviewed and deemed unsplittable)
3. Redundant state: $state that duplicates or could be $derived from other state
4. Copy-paste: near-duplicate code blocks across files that should share a helper
5. Bug risks: unguarded async, missing cleanup, event listener leaks, race conditions
6. Inefficiency: unnecessary re-computation, work in hot paths, unbounded allocations

Skip lines marked with // REFACTOR-OK: <reason> — these have been reviewed and intentionally kept as-is.

Output a numbered list of findings with file paths and line numbers, sorted by severity (high → low). Do NOT fix anything yet.

Phase 2: User approval

Present the findings list to the user. Ask which items to fix (e.g. "all", specific numbers, or "skip"). Wait for the user to respond before proceeding.

Phase 3: Fix approved items

Fix only approved items, one at a time. After all fixes, run pnpm check to verify.

Briefly summarize what was changed.