vibeeval/vibecosystem/refactor

Trust tier 2
  • community
LLM review pending
Source check pending

Checking that the SKILL.md file is still in the repository.

Code refactoring workflow - analyze → plan → implement → review → validate

Install

npx vskill@latest install vibeeval/vibecosystem/refactor

SKILL.md

/refactor - Refactoring Workflow

Safe refactoring with review gates.

When to Use

  • "Refactor X"
  • "Clean up this code"
  • "Extract this into a module"
  • "Improve the architecture of Y"
  • Large-scale code restructuring
  • Technical debt reduction

Workflow Overview

┌──────────┐ ┌────────────┐ ┌──────────┐ ┌──────────┐ ┌───────────┐
│ phoenix │───▶│ plan- │───▶│ kraken │───▶│plan-reviewer│───▶│ arbiter │
│ │ │ agent │ │ │ │ │ │ │
└──────────┘ └────────────┘ └──────────┘ └──────────┘ └───────────┘
Analyze Plan Implement Review Verify
current changes refactor changes tests pass

Agent Sequence

#AgentRoleOutput
1phoenixAnalyze current code, identify improvement areasAnalysis report
2plan-agentCreate safe refactoring planStep-by-step plan
3krakenImplement the refactoringCode changes
4plan-reviewerReview changes for correctnessReview report
5arbiterVerify all tests still passTest report

Refactoring Principles

  1. Tests first: Ensure adequate test coverage before refactoring
  2. Small steps: Each change should be independently verifiable
  3. Behavior preserved: No functional changes during refactor
  4. Reviewable: Changes should be easy to review

Execution

Phase 1: Analyze
Task(
subagent_type="phoenix",
prompt="""
Analyze for refactoring: [TARGET_CODE]
Identify:
- Current pain points
- Code smells
- Improvement opportunities
- Risk areas
- Test coverage gaps
"""
)
Phase 2: Plan
Task(
subagent_type="plan-agent",
prompt="""
Plan refactoring: [TARGET_CODE]
Analysis: [from phoenix]
Create:
- Step-by-step refactoring plan
- Each step should be:
- Small and focused
- Independently testable
- Reversible
- Identify files affected
- Risk mitigation strategy
"""
)
Phase 3: Implement
Task(
subagent_type="kraken",
prompt="""
Implement refactoring: [TARGET_CODE]
Plan: [from plan-agent]
Requirements:
- Follow plan exactly
- Run tests after each step
- Stop if tests fail
- NO behavior changes
"""
)
Phase 4: Review
Task(
subagent_type="plan-reviewer",
prompt="""
Review refactoring: [TARGET_CODE]
Changes: [git diff from kraken]
Check:
- Behavior preserved
- No unintended changes
- Code quality improved
- Patterns consistent
"""
)
Phase 5: Validate
Task(
subagent_type="arbiter",
prompt="""
Validate refactoring: [TARGET_CODE]
- Run full test suite
- Verify no regressions
- Check type errors
- Run linting
"""
)

Refactoring Types

Extract Module
phoenix → plan-agent → kraken → plan-reviewer → arbiter
Rename/Restructure
phoenix → kraken → arbiter (simpler, skip detailed planning)
Architecture Change
phoenix → plan-agent → [kraken → plan-reviewer] × N phases → arbiter

Example

User: /refactor Extract the validation logic into a separate module
Claude: Starting /refactor workflow...
Phase 1: Analyzing current structure...
[Spawns phoenix]
Found: Validation logic spread across 4 files
- form.ts (lines 45-120)
- api.ts (lines 200-280)
- user.ts (lines 15-45)
- order.ts (lines 88-130)
Phase 2: Planning extraction...
[Spawns plan-agent]
Plan:
1. Create src/validation/index.ts
2. Extract common validators
3. Update imports one file at a time
4. Run tests after each change
Phase 3: Implementing...
[Spawns kraken]
Completed all 4 steps, tests green after each
Phase 4: Reviewing changes...
[Spawns plan-reviewer]
✅ All behavior preserved
✅ DRY improved (removed 45 duplicate lines)
✅ New structure consistent
Phase 5: Final validation...
[Spawns arbiter]
✅ 312 tests passing, 0 regressions
Refactoring complete!

Safety Flags

  • --dry-run: Plan but don't implement
  • --step-by-step: Pause after each change for approval
  • --coverage-check: Require >80% coverage before proceeding

Security

  • Security scanPassed, 100 of 100Pattern checks on the skill's files.
  • Model reviewNot runA model review is scheduled for the next rescan.

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
534
Repository forks
45
Repository npm downloads, 7 days
6
7-day trend score
45

Stars and forks are for the whole GitHub repository, not this skill alone. Downloads are for the repository's npm package. Refreshed 29 Apr 2026, 6 months ago.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/vibeeval/vibecosystem/refactor/badge)](https://verified-skill.com/skills/vibeeval/vibecosystem/refactor)

Something wrong with this skill or its page? Tell us.