メニュー
Use when the user asks to refactor code, review code quality, or fix code smells — orchestrates the full refactoring pipeline from detection through fix, with parallel worktree isolation per independent refactoring group
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Use when the user invokes /evo:loop or asks to run autonomous improvement cycles, self-evolving development, compound discovery, or multi-cycle code improvement with research, build, audit, and learning phases
Use after build has produced build-report.md. Validates the build via four parallel sub-auditors (eval-replay, lint, regression, build-quality) and produces ALL-PASS verdict. Adversarial mode default-on per CLAUDE.md.
Use when the user asks to explain how a feature/subsystem works and how the loops built it, or what the concurrent loops/lanes/cycles did and why. Default output = per-FEATURE HTML explainer pages (two movements — how it works today, then how it got built, with verified file:line claims and drift callouts) under docs/explain/; per-batch/per-lane story pages are the fallback framing. Derived from geoffreylitt/explain-diff-html, customized for evolve-loop.
Use when writing or changing any production code or tests — new features, bug fixes, refactors, or test authoring, in any language, by any model on any CLI.
Use when running any LLM agent below Fable 5 — Claude (Opus/Sonnet/Haiku), GPT via codex, Gemini via agy, or local models via ollama — to adopt Fable 5's operating discipline — evidence-first investigation, premise verification, root-cause-only fixes, adversarial self-review, calibrated autonomy, and honest failure reporting. Load at session start or as a persona overlay for phase agents on any CLI.
Use after ship completes. Three sub-reflectors (instinct, gene, failure) run in parallel to extract lessons, update gene pool, and analyze any failures. Off the latency-critical path.
| name | refactor |
| description | Use when the user asks to refactor code, review code quality, or fix code smells — orchestrates the full refactoring pipeline from detection through fix, with parallel worktree isolation per independent refactoring group |
| argument-hint | [scan|fix|security|performance|review|auto|arch|complexity|health|diff|hotspots] |
Orchestrate the complete refactoring workflow: detect smells, prioritize by weighted scoring, partition into independent groups, execute in parallel worktrees via subagents, merge and verify. Enhanced with cognitive complexity scoring, architecture analysis, and speed-optimized parallel scan pipelines.
| Section | Purpose |
|---|---|
| Auto Mode Detection | Detect bypass/yolo mode for autonomous execution |
| Git Isolation | Worktree-based isolation for all refactoring work |
| Workflow Summary | 5-phase pipeline overview |
| Quick Modes | Scoped invocations for targeted refactoring |
| Cross-Reference Map | Skill routing by issue type |
| Reference | Detailed phase docs, catalogs, techniques |
Before starting, check if the user is in bypass/yolo mode (auto-accept permissions enabled). Detection signals:
--dangerously-skip-permissions or equivalentWhen auto mode is detected: Skip all confirmation prompts. Automatically select all issues, partition, launch parallel subagents, and merge passing branches without pausing. This enables fully autonomous refactoring.
When auto mode is NOT detected: Pause for user confirmation at each checkpoint described in reference/workflow.md.
All refactoring work MUST be done in isolated git worktrees branched from main. Never commit directly to main.
The orchestrator selects the execution mode based on the partition result from Phase 3:
| Condition | Mode | Worktrees |
|---|---|---|
| 1 group or all issues touch the same files | Sequential | 1 worktree, 1 branch |
| N independent groups with no file overlap | Parallel | N worktrees, N branches, N subagents |
../refactor-wt-<group-slug> → branch: refactor/<group-slug>
Examples:
../refactor-wt-extract-auth → refactor/extract-auth
../refactor-wt-simplify-payment → refactor/simplify-payment
../refactor-wt-cleanup-utils → refactor/cleanup-utils
Phase 1 SCAN ──────────── on main (read-only, parallel tool pipeline)
Phase 2 PRIORITIZE ────── on main (read-only, weighted scoring)
Phase 3 PLAN & PARTITION ─ on main (read-only, graph-based critical pair analysis)
Phase 4 EXECUTE ────────── in worktrees (parallel subagents, up to 3 passes)
Phase 5 MERGE & VERIFY ── back on main (sequential merge, final scan)
For the full phase-by-phase protocol including subagent prompts, isolation rules, and merge handling, read reference/workflow.md.
For scan tool launch and speed optimizations, read reference/scan-pipeline.md.
The user can scope the refactoring with arguments:
| Command | Behavior |
|---|---|
/evo:refactor | Full pipeline on current file or user-specified files |
/evo:refactor scan | Phase 1 only — detect and report, no changes |
/evo:refactor fix <smell> | Skip to Phase 3-4 for a specific known smell |
/evo:refactor security | Security-focused scan using security-patterns-code-review |
/evo:refactor performance | Performance-focused scan using performance-anti-patterns |
/evo:refactor review | Full code review using review-cheat-sheet as guide |
/evo:refactor auto | Force auto mode — plan, partition, execute in parallel, merge without confirmation |
/evo:refactor arch | Architecture-only analysis — circular deps, boundaries, centrality |
/evo:refactor complexity | Cognitive complexity report only — no fixes |
/evo:refactor health | Composite health score per function — multi-metric scoring |
/evo:refactor diff | Scan only files changed since last commit |
/evo:refactor hotspots | Git-history analysis — find high-churn, high-smell files |
| When you find... | Use skill... |
|---|---|
| Code smells | detect-code-smells |
| Anti-patterns | anti-patterns-catalog |
| Which technique to use | refactoring-decision-matrix |
| Long/complex methods | refactor-composing-methods |
| Misplaced responsibilities | refactor-moving-features |
| Data handling issues | refactor-organizing-data |
| Complex conditionals | refactor-simplifying-conditionals |
| Bad method interfaces | refactor-simplifying-method-calls |
| Inheritance problems | refactor-generalization |
| FP improvements | refactor-functional-patterns |
| Type safety gaps | type-system-patterns |
| Design pattern opportunity | design-patterns-creational-structural, design-patterns-behavioral |
| Architecture violations | architectural-patterns |
| Security vulnerabilities | security-patterns-code-review |
| Performance issues | performance-anti-patterns |
| Database problems | database-review-patterns |
| Error handling gaps | error-handling-patterns |
| Concurrency bugs | concurrency-patterns |
| Test quality issues | testing-patterns |
| Observability gaps | observability-patterns |
| DI/coupling problems | dependency-injection-module-patterns |
| Language anti-idioms | language-specific-idioms |
| API contract issues | review-api-contract |
| SOLID violations | review-solid-clean-code |
| Full code review | review-cheat-sheet |
| File | When to read |
|---|---|
| reference/workflow.md | Detailed Phases 1-5 with subagent prompts and merge protocol |
| reference/scan-pipeline.md | Phase 1: parallel tool execution + speed optimizations |
| reference/continuous-integration.md | PR review integration, debt tracking, velocity dashboards |
| File | When to read |
|---|---|
| reference/code-smells.md | Phase 1: detecting code smells (22 smells, thresholds) |
| reference/complexity-scoring.md | Phase 1: computing cognitive complexity scores |
| reference/architecture-analysis.md | Phase 1: circular deps, boundaries, fan-in/fan-out |
| reference/health-scoring.md | Phase 2: computing composite priority scores |
| File | When to read |
|---|---|
| reference/refactoring-techniques.md | Phase 3: selecting technique for each detected smell (66 techniques) |
| reference/smell-to-technique-map.md | Quick lookup: smell → fix technique + skill |
| File | When to read |
|---|---|
| reference/safety-protocols.md | Phase 4: LLM safety, RefactoringMirror, re-prompting |
| reference/prompt-engineering.md | Phase 4: crafting subagent prompts (specificity ladder) |
| File | When to read |
|---|---|
| reference/language-notes.md | Phase 3-4: TS/JS, Python, Go, Java specifics |
| reference/worked-example.md | First time: full pipeline walkthrough (UserService) |