SK-ClaudeCode

Unified agent and skill system for Claude Code

29 agents · 74 core skills (+141 scientific domains in full mode) · 18 magic keywords — all in one package.

🚀 Installation

# Marketplace (Recommended)
claude plugin marketplace add cool25th/sk-claudecode
claude plugin install sk-claudecode
/sk-claudecode:setup

# NPM (Direct)
npm install -g sk-claudecode

# Local (Development)
git clone https://github.com/cool25th/sk-claudecode && cd sk-claudecode
npm install && npm run build && npm run setup

| Mode | Command | Size | |------|---------|------| | Standard | /sk-claudecode:setup | ~2MB | | Minimal | /sk-claudecode:setup --minimal | ~1MB | | Full | /sk-claudecode:setup --full | ~17MB |

💡 Core Concept: Think → Build → Check

🧠 Think          ⚡ Build          🔍 Check          📚 Help
┌──────────┐     ┌──────────┐     ┌──────────┐     ┌──────────┐
│ architect │     │ executor │     │ reviewer │     │ explore  │
│ planner   │────▶│ designer │────▶│ qa-tester│     │ writer   │
│ critic    │     │ scientist│     │ security │     │researcher│
└──────────┘     └──────────┘     └──────────┘     └──────────┘
      ▲                                 │
      └─────────────────────────────────┘
                 Iterate until ✅

🤖 Agents (29)

🧠 Think

| Agent | Model | When to use | |-------|-------|-------------| | architect | Opus | System architecture, debugging strategy | | planner | Opus | Feature planning, requirements analysis | | critic | Opus | Review and challenge a work plan |

⚡ Build

| Agent | Model | When to use | |-------|-------|-------------| | executor | Sonnet | General implementation tasks | | ultra-executor | Opus | Complex multi-file autonomous tasks | | designer | Sonnet | UI/UX development | | build-fixer | Sonnet | Fix build errors (TS, Go, etc.) | | git-master | Sonnet | Commits, rebasing, history management | | scientist | Sonnet | Data analysis, research, experiment design | | mobile-developer | Sonnet | iOS / Android / Flutter apps | | finance | Opus | Trading systems, portfolio management | | ontology | Sonnet | Ontology design, implementation, review | | document-writer | Sonnet | Document creation (PDF, PPTX, XLSX, DOCX) |

🔍 Check

| Agent | Model | When to use | |-------|-------|-------------| | code-reviewer | Opus | Comprehensive code review | | security-reviewer | Opus | Security vulnerabilities (OWASP Top 10) | | database-reviewer | Opus | SQL optimization, schema design | | go-reviewer | Opus | Go-specific best practices | | python-reviewer | Opus | Python-specific (PEP 8, type hints) | | qa-tester | Sonnet | Interactive CLI testing (tmux) | | tdd-guide | Opus | Test-Driven Development | | e2e-runner | Opus | End-to-end browser testing | | tracer | Opus | Runtime tracing, profiling, race conditions | | scientist-reviewer | Opus | Research methodology validation | | designer-reviewer | Opus | UI/UX accessibility, consistency |

📚 Help

| Agent | Model | When to use | |-------|-------|-------------| | explore | Sonnet | Search codebase for files and patterns | | researcher | Sonnet | External docs and reference research | | writer | Haiku | README, API docs, comments | | vision | Sonnet | Analyze images, PDFs, diagrams | | refactor-cleaner | Opus | Find and remove dead code |

🚀 Orchestration Modes

| Command | How it works | |---------|-------------| | 🔄 Full Cycle | | | ralph | 🧠→⚡→🔍 + retry until all pass | | autopilot | 🧠→⚡→🔍 auto-selects best agents | | pipeline | 🧠→⚡→🔍 you define the agent chain | | ecomode | 🧠→⚡→🔍 Haiku/Sonnet only (60%+ cheaper) | | ⚡ Parallel | | | ultrawork | All phases in parallel across files | | ultrapilot | Splits by file owner, parallel per partition | | swarm | N coordinated agents share the cycle | | 🔍 Check | | | ultraqa | Build → Check → loop until pass |

ralph:       @planner → @executor → @code-reviewer ── fail? → @build-fixer → retry
ultrawork:   ┌─ @executor → @code-reviewer       (3 lanes
             ├─ @executor → @security-reviewer     in parallel)
             └─ @designer → @designer-reviewer

📋 Commands & Skills

| Command | Purpose | |---------|---------| | /sk-claudecode:ralph | Full cycle, retry until done | | /sk-claudecode:ultrawork | Max parallelism (ulw) | | /sk-claudecode:plan | Project planning with spec | | /sk-claudecode:code-review | Code review (auto-detects language) | | /sk-claudecode:tdd | Test-driven development | | /sk-claudecode:ultraqa | Build/test/lint verification loop | | /sk-claudecode:deep-interview | Socratic requirement extraction | | /sk-claudecode:ai-slop-cleaner | Remove AI-generated anti-patterns | | /sk-claudecode:handoff | Session handoff for continuation | | /sk-claudecode:learner | Persistent cross-session learning | | /sk-claudecode:finance | KR/US market, quant strategy | | /sk-claudecode:mobile | iOS / Android / Flutter | | /sk-claudecode:ontology | Palantir-style ontology | | /sk-claudecode:scientist | 141 scientific domains | | /sk-claudecode:web-search-browser | Browser-based web search (agent-browser) | | /sk-claudecode:cancel | Stop any active mode |

🪄 Magic Keywords

Type anywhere in your prompt — no slash command needed. 한국어 지원.

| Keyword | Triggers | Effect | |---------|----------|--------| | ultrawork | ultrawork ulw uw | Parallel agent orchestration | | search | search find grep 검색 | Exhaustive codebase search | | analyze | analyze debug audit 분석 | Deep analysis with context | | ultrathink | ultrathink think reason | Extended reasoning | | refactor | refactor cleanup 리팩터 | Plan-first safe refactoring | | tdd | tdd write test 테스트 작성 | RED→GREEN→REFACTOR cycle | | security | security owasp 보안 | OWASP Top 10 checklist | | plan | plan spec blueprint 기획 | Spec-first, no code until approved | | review | code review cr 코드 리뷰 | Structured code review | | docs | document readme 문서화 | Documentation mode | | perf | optimize slow 성능 | Measure-first optimization | | browser | browser e2e 브라우저 | Browser automation | | websearch | websearch web-search 웹검색 | Browser-based web search | | deep-interview | interview clarify 인터뷰 | Requirement extraction | | ai-slop | slop over-engineered 슬로프 | Remove AI code patterns | | tracer | trace profile 추적 | Runtime tracing & profiling | | handoff | handoff 인수인계 세션 이관 | Session context handoff | | ecomode | ecomode cheap 절약 | Haiku-first, minimize tokens |

💻 Usage Examples

# Full Cycle — retry until done
/sk-claudecode:ralph implement user profile page with avatar upload

# Parallel — maximum speed
/sk-claudecode:ultrawork refactor authentication to use OAuth2

# Save tokens (60%+ cheaper)
/sk-claudecode:ecomode fix all lint errors in src/

# Clarify before coding
/sk-claudecode:deep-interview "I want to build a task management app"

# Remove AI slop
clean slop in src/services/

# Runtime profiling
trace execution of the login flow to find the bottleneck

# 📦 Session Handoff — preserve context for next session
handoff this session

💡 Tip: ralph automates Think → Build → Check and keeps iterating until everything passes.

✨ Highlights

• Real-time HUD — Status line shows active agents and progress
• Smart Model Routing — Saves 30-50% on tokens automatically
• Automatic Parallelization — Complex tasks distributed across agents
• Persistent Execution — Won't give up until the job is verified complete
• Learn from Experience — Extracts and reuses problem-solving patterns

📚 More Documentation

• Ultra Suite Guide — Advanced workflow skills
• Ontology Guide — Ontology development step-by-step
• AGENTS.md — Complete agent reference

🙏 Inspired By

oh-my-claudecode · superpowers · claude-scientific-skills · claude-mem · oh-my-opencode

📜 License

MIT — Components retain their original licenses.

Made with ❤️ by merging the best of Claude Code ecosystems

Testing

npm test                                    # run all tests
npm run test:unit -- --grep "module-name"   # filtered tests
npm run test:full                           # lint + all tests
npm run test:ci                             # CI-safe full check