SC Refactor Review Skill

Reference guide for routing review and refactoring requests to specialized agents.

Quick Reference

Commands

Command	Use When	Invokes
/sc-refactor:sc-refactor	Refactor code (plan or execute)	sc-refactor-planner + sc-refactor-executor
/sc-refactor:sc-review-pr	Reviewing a PR for quality with ticket context	sc-code-reviewer + sc-structural-reviewer
/sc-refactor:sc-pr-comments	View unresolved PR review comments	Scripts (GraphQL)
/sc-refactor:sc-resolve-pr-parallel	Batch resolve all PR comments	sc-pr-comment-resolver (parallel)
/sc-refactor:sc-cleanup	Post-AI session cleanup (debug statements, duplicates)	4 agents (dead-code, duplication, naming, test)
/sc-refactor:sc-audit	Verify structural completeness (wiring, configs)	sc-structural-reviewer
/sc-refactor:sc-codebase-health	Full codebase analysis	All 6 agents in parallel
/sc-refactor:sc-adversarial-review	Multi-model adversarial review	Codex CLI + Gemini CLI + Claude

Agents

Agent	Focus	Color
sc-refactor-planner	Analyze code, recommend refactoring opportunities	yellow
sc-refactor-executor	Execute ONE surgical refactoring with test verification	green
sc-structural-reviewer	Change completeness, orphaned code, dev artifacts	blue
sc-duplication-hunter	Copy-paste, structural, logic duplication	yellow
sc-abstraction-critic	YAGNI violations, over-engineering, wrapper hell	orange
sc-naming-auditor	Convention violations, semantic drift	cyan
sc-dead-code-detector	Unreferenced exports, orphan files, commented code	red
sc-test-organizer	Test structure, missing tests, fixture sprawl	green
sc-pr-comment-resolver	Implement PR review comment changes	blue

External dependency (from simpleclaude-core):

• sc-code-reviewer - Bugs, security, CLAUDE.md compliance

Routing Table

Match the user's request to the appropriate command or agents:

User Intent	Route To
"refactor", "extract method", "rename", "consolidate", "inline"	/sc-refactor
"refactoring opportunities", "what should I refactor"	/sc-refactor (plan mode)
"review PR", "check my PR", "PR review"	/sc-review-pr
"PR comments", "view comments", "unresolved comments"	/sc-pr-comments
"resolve comments", "fix PR feedback", "address review"	/sc-resolve-pr-parallel
"clean up", "after AI session", "find debug", "console.log"	/sc-cleanup
"audit", "structural check", "verify wiring", "missing config"	/sc-audit
"health check", "full analysis", "comprehensive"	/sc-codebase-health
"adversarial review", "red team", "break this code", "multi-model"	/sc-adversarial-review
"dead code", "unused", "orphan"	sc-dead-code-detector
"duplicate", "DRY", "repeated"	sc-duplication-hunter
"simplify", "YAGNI", "over-engineer"	sc-abstraction-critic
"naming", "consistency", "convention"	sc-naming-auditor
"test organization", "test structure"	sc-test-organizer
"structural", "complete changes"	sc-structural-reviewer

Agent Spawning

Single Agent

Task(
  subagent_type: "sc-refactor:sc-[agent-name]",
  prompt: "Analyze [target] for [focus area]. Report findings with confidence scores."
)

Multiple Agents (Parallel)

Spawn all relevant agents in a single message with run_in_background: true:

Task(subagent_type: "sc-refactor:sc-duplication-hunter", run_in_background: true, prompt: "...")
Task(subagent_type: "sc-refactor:sc-abstraction-critic", run_in_background: true, prompt: "...")

Full Health Check (6 Agents)

Task(subagent_type: "sc-refactor:sc-structural-reviewer", run_in_background: true, ...)
Task(subagent_type: "sc-refactor:sc-duplication-hunter", run_in_background: true, ...)
Task(subagent_type: "sc-refactor:sc-abstraction-critic", run_in_background: true, ...)
Task(subagent_type: "sc-refactor:sc-naming-auditor", run_in_background: true, ...)
Task(subagent_type: "sc-refactor:sc-dead-code-detector", run_in_background: true, ...)
Task(subagent_type: "sc-refactor:sc-test-organizer", run_in_background: true, ...)

Output Synthesis

After agents complete, synthesize findings by priority:

## [Category] Analysis

### High Priority (confidence >= 90)
- [finding]: [description] ([file:line])

### Medium Priority (confidence 80-89)
- [finding]: [description] ([file:line])

### Recommendations
- [actionable next step]

Command Details

sc-review-pr

Context-aware PR review with ticket integration. Gathers:

• PR metadata (title, body, commits, changed files)
• Linked ticket context (Jira, GitHub Issues, beads)
• Relevant CLAUDE.md guidelines

Then runs parallel review agents focused on the PR diff.

sc-cleanup

Post-AI session cleanup. Spawns 4 agents to find:

• Debug statements (console.log, print, debugger, binding.pry)
• Code duplication (AI rewrote instead of reusing)
• Naming inconsistencies
• Test organization issues

Offers auto-fix for immediate issues (debug removal, dead code deletion).

sc-audit

Structural completeness verification. Uses sc-structural-reviewer to check:

• Route registration
• ENV variable documentation
• Database migrations
• Barrel file exports
• Documentation updates
• Related file renames (CSS, tests, stories)

Reports PASS/FAIL per category with fix suggestions.

sc-refactor

Surgical code refactoring workflow with two modes:

Plan Mode (no specific action given):

1. Invokes sc-refactor-planner to analyze target
2. Presents prioritized recommendations
3. User selects one to execute

Execute Mode (specific action given):

1. Invokes sc-refactor-executor with the specific refactoring
2. Executor runs tests, makes change, verifies behavior preserved
3. Reports success/failure with files modified

Example usage:

• /sc-refactor src/utils - Analyze for opportunities
• /sc-refactor extract duplicate validation in auth.ts - Execute specific refactoring

Key constraints:

• ONE refactoring per invocation
• Tests must pass before and after
• Zero behavioral changes
• Does NOT auto-commit

sc-codebase-health

Comprehensive health check running all 6 agents in parallel. Produces:

• Executive summary
• Quick wins (fixable in <30 min)
• Recommended refactors
• Tech debt backlog
• Items to skip (not worth effort)

Offers auto-fix for quick wins after analysis.

sc-adversarial-review

Multi-model adversarial review using external AI CLIs. Runs OpenAI Codex, Google Gemini, and Claude in parallel against the same target with an attacker's mindset.

Value: Model diversity. Different architectures find different blind spots. Findings categorized as:

• Consensus (multiple models agree) — highest confidence
• Unique (one model only) — investigate
• Divergent (models disagree) — most interesting

Requires: codex and/or gemini CLI installed. Degrades gracefully to whichever is available + Claude.

Example usage:

• /sc-adversarial-review — Review uncommitted changes
• /sc-adversarial-review branch — Review current branch vs main
• /sc-adversarial-review src/auth/ — Review specific directory