Skip to content

Transconnectome/claude-code-coord

Folders and files

NameName
Last commit message
Last commit date

Latest commit

Β 

History

2 Commits
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 

Repository files navigation

coord β€” Intelligent Complex Task Workflow

License: MIT Python 3.10+ Node.js 18+ Claude Code

/coord is a 6-stage pipeline built on top of Claude Code that orchestrates memory retrieval, planning, parallel research, specialist agent delegation, and anti-sycophancy verification for complex engineering and research tasks.

/coordλŠ” λ©”λͺ¨λ¦¬ 검색, κ³„νš 수립, 병렬 λ¦¬μ„œμΉ˜, μ „λ¬Έκ°€ μ—μ΄μ „νŠΈ μœ„μž„, 아첨 λ°©μ§€ 검증을 μ‘°μœ¨ν•˜λŠ” 6단계 νŒŒμ΄ν”„λΌμΈμž…λ‹ˆλ‹€.


Quick Start (λΉ λ₯Έ μ‹œμž‘)

# Step 1 β€” Clone this repository (μ €μž₯μ†Œ 볡사)
git clone https://github.com/Transconnectome/claude-code-coord.git
cd claude-code-coord

# Step 2 β€” Copy framework files (ν”„λ ˆμž„μ›Œν¬ 파일 볡사)
cp -r .claude/ ~/.claude/

# Step 3 β€” Install MCP servers, then verify (MCP μ„œλ²„ μ„€μΉ˜ 및 검증)
npm install -g @modelcontextprotocol/server-sequential-thinking
claude code --print "/coord 'Hello, is this working?'"

Full installation details: docs/SETUP_GUIDE.md


Architecture (μ•„ν‚€ν…μ²˜)

6-Stage Pipeline

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚                   /coord β€” 6-Stage Pipeline                         β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚  Stage 0 β”‚  BOOTSTRAP β€” detect task complexity, choose mode         β”‚
β”‚  μ€€λΉ„     β”‚  (μž‘μ—… λ³΅μž‘λ„ μΈ‘μ •, μ‹€ν–‰ λͺ¨λ“œ 선택)                         β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚  Stage 1 β”‚  MEMORY PRE-CHECK β€” parallel 3-source scan               β”‚
β”‚  λ©”λͺ¨λ¦¬   β”‚  Auto-memory β€– Serena MCP β€– Graphiti knowledge graph     β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚  Stage 2 β”‚  PLAN β€” sequential-thinking MCP + parallelization map    β”‚
β”‚  κ³„νš     β”‚  (닀단계 μΆ”λ‘  + 병렬화 κ°€λŠ₯ν•œ μž‘μ—… 뢄리)                     β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚  Stage 3 β”‚  RESEARCH β€” parallel subagents (Explore β€– web β€– docs)   β”‚
β”‚  λ¦¬μ„œμΉ˜   β”‚  (탐색/μ›Ή/κ³΅μ‹λ¬Έμ„œ μ„œλΈŒμ—μ΄μ „νŠΈ 병렬 μ‹€ν–‰)                    β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚  Stage 4 β”‚  EXECUTE β€” specialist agents + intermediate validation   β”‚
β”‚  μ‹€ν–‰     β”‚  (μ „λ¬Έκ°€ μ—μ΄μ „νŠΈ + 단계별 검증)                            β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚  Stage 5 β”‚  REVIEW β€” critic agent + optional premium model review   β”‚
β”‚  리뷰     β”‚  (critic μ—μ΄μ „νŠΈ + 선택적 프리미엄 λͺ¨λΈ κ²€ν† )               β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚  Stage 6 β”‚  MEMORY UPDATE β€” route lessons to correct store          β”‚
β”‚  λ©”λͺ¨λ¦¬   β”‚  Auto-memory > Graphiti > Serena, update MEMORY.md       β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

Memory Layer (λ©”λͺ¨λ¦¬ 계측)

The pipeline always reads from three memory sources in parallel before planning. Each source serves a distinct purpose.

νŒŒμ΄ν”„λΌμΈμ€ κ³„νš 전에 항상 μ„Έ κ°€μ§€ λ©”λͺ¨λ¦¬ μ†ŒμŠ€λ₯Ό λ³‘λ ¬λ‘œ μ½μŠ΅λ‹ˆλ‹€.

Source What it stores (μ €μž₯ λ‚΄μš©) Persistence (지속성)
Auto-memory (~/.claude/projects/…/memory/) User preferences, session lessons, project state (μ‚¬μš©μž μ„ ν˜Έ, μ„Έμ…˜ κ΅ν›ˆ, ν”„λ‘œμ νŠΈ μƒνƒœ) Cross-session (μ„Έμ…˜ κ°„)
Serena MCP Code symbols, file locations, workspace structure (μ½”λ“œ 심볼, 파일 μœ„μΉ˜, μ›Œν¬μŠ€νŽ˜μ΄μŠ€ ꡬ쑰) Within session + project (μ„Έμ…˜ λ‚΄ + ν”„λ‘œμ νŠΈ)
Graphiti (graphiti-memory) Domain knowledge, research decisions, meeting records (도메인 지식, 연ꡬ κ²°μ •, 회의둝) Long-term knowledge graph (μž₯κΈ° 지식 κ·Έλž˜ν”„)

MCP Tool Mapping Table (MCP 도ꡬ λ§€ν•‘)

Task type (μž‘μ—… μœ ν˜•) Primary MCP Fallback
Multi-step reasoning (닀단계 μΆ”λ‘ ) sequential-thinking Native Claude
Web search (μ›Ή 검색) tavily WebSearch
Official library docs (곡식 λ¬Έμ„œ) context7 Tavily
UI component generation (UI 생성) magic Manual HTML
Browser automation (λΈŒλΌμš°μ € μžλ™ν™”) playwright Manual testing
Symbol operations (심볼 μ‘°μž‘) serena Grep + Edit
Bulk pattern edits (일괄 νŒ¨ν„΄ νŽΈμ§‘) morphllm MultiEdit
Document analysis (λ¬Έμ„œ 뢄석) notebooklm Read + summarize
Knowledge graph memory (지식 κ·Έλž˜ν”„) graphiti-memory Auto-memory

Installation (μ„€μΉ˜)

Prerequisites (사전 μš”κ΅¬μ‚¬ν•­)

  • Python 3.10 or higher (Python 3.10 이상)
  • Node.js 18 or higher (Node.js 18 이상)
  • Claude Code installed and authenticated (μ„€μΉ˜ 및 인증 μ™„λ£Œ)
  • API keys: TAVILY_API_KEY, OPENROUTER_API_KEY (optional for premium review) (선택적 β€” 프리미엄 리뷰용)

Step-by-Step (단계별 μ„€μΉ˜)

1. Clone the repository (μ €μž₯μ†Œ 볡사)

git clone https://github.com/Transconnectome/claude-code-coord.git
cd claude-code-coord

2. Copy the .claude/ framework directory (ν”„λ ˆμž„μ›Œν¬ 디렉토리 볡사)

# Back up your existing ~/.claude/ first (κΈ°μ‘΄ μ„€μ • λ°±μ—… ꢌμž₯)
cp -r ~/.claude/ ~/.claude.backup/

cp -r .claude/ ~/.claude/

3. Install MCP servers (MCP μ„œλ²„ μ„€μΉ˜)

# Required servers (ν•„μˆ˜ μ„œλ²„)
npm install -g @modelcontextprotocol/server-sequential-thinking
npm install -g @modelcontextprotocol/server-memory

# Recommended servers (ꢌμž₯ μ„œλ²„)
npm install -g @tavily/mcp-server
npm install -g @21st-dev/magic

Full MCP installation instructions: docs/MCP_SERVERS.md

4. Copy and configure settings (μ„€μ • 파일 ꡬ성)

cp settings/settings.json.template ~/.claude/settings.json
# Edit ~/.claude/settings.json β€” fill in API keys and hook paths
# ~/.claude/settings.json νŽΈμ§‘ β€” API ν‚€ 및 hook 경둜 μž…λ ₯

5. Verify the installation (μ„€μΉ˜ 검증)

cd tests/
python -m pytest test_hooks.py -v

Usage (μ‚¬μš©λ²•)

Basic invocation (기본 호좜)

# Explicit invocation (λͺ…μ‹œμ  호좜)
/coord "Implement JWT authentication for the FastAPI service"

# With @sub shorthand (λ°±κ·ΈλΌμš΄λ“œ μ—μ΄μ „νŠΈ 단좕 호좜)
@sub Implement JWT authentication for the FastAPI service

Auto-activation keywords (μžλ™ ν™œμ„±ν™” ν‚€μ›Œλ“œ)

/coord activates automatically when the user message contains signals of complexity. You do not need to type /coord explicitly.

λ³΅μž‘μ„± μ‹ ν˜Έκ°€ ν¬ν•¨λœ λ©”μ‹œμ§€μ—μ„œλŠ” /coordλ₯Ό λͺ…μ‹œμ μœΌλ‘œ μž…λ ₯ν•˜μ§€ μ•Šμ•„λ„ μžλ™ ν™œμ„±ν™”λ©λ‹ˆλ‹€.

Signal category (μ‹ ν˜Έ λ²”μ£Ό) Example keywords (μ˜ˆμ‹œ ν‚€μ›Œλ“œ)
Scope ambiguity (λ²”μœ„ λͺ¨ν˜Έ) "not sure", "thinking about", "explore"
Multi-domain (닀쀑 도메인) "frontend + backend", "infra and code"
Long pipelines (κΈ΄ νŒŒμ΄ν”„λΌμΈ) "step by step", "refactor the whole", "migrate"
Debugging (디버깅) "figure out why", "root cause", "broken since"
Architecture (μ•„ν‚€ν…μ²˜) "design a system", "how should I structure"
Research (λ¦¬μ„œμΉ˜) "find the best", "compare approaches", "survey"

When NOT to use /coord (μ‚¬μš©ν•˜μ§€ 말아야 ν•  경우)

  • Simple one-file edits (λ‹¨μˆœ 단일 파일 μˆ˜μ •)
  • Quick factual questions (λΉ λ₯Έ 사싀 질문)
  • Single-tool operations under 3 steps (3단계 미만 단일 도ꡬ μž‘μ—…)
  • Casual conversation (일상적인 λŒ€ν™”)

For detailed use-case guidance: docs/WHEN_TO_USE.md


MCP Servers (MCP μ„œλ²„)

Server Role (μ—­ν• ) API key required (API ν‚€ ν•„μš”)
sequential-thinking Multi-step structured reasoning (닀단계 ꡬ쑰적 μΆ”λ‘ ) No
memory Key-value session persistence (μ„Έμ…˜ λ‚΄ ν‚€-κ°’ μ €μž₯) No
tavily Real-time web search (μ‹€μ‹œκ°„ μ›Ή 검색) Yes β€” TAVILY_API_KEY
context7 Official library documentation (곡식 라이브러리 λ¬Έμ„œ) No
magic UI component generation via 21st.dev (UI μ»΄ν¬λ„ŒνŠΈ 생성) Yes β€” MAGIC_API_KEY
playwright Browser automation and E2E testing (λΈŒλΌμš°μ € μžλ™ν™”) No
morphllm Bulk pattern-based code edits (일괄 νŒ¨ν„΄ 기반 μ½”λ“œ μˆ˜μ •) No
serena Semantic code navigation and memory (μ‹œλ§¨ν‹± μ½”λ“œ 탐색) No
notebooklm Document analysis and knowledge base (λ¬Έμ„œ 뢄석 + 지식 베이슀) No (Google auth)

Full per-server documentation: docs/MCP_SERVERS.md


Hook System β€” Anti-Sycophancy (ν›… μ‹œμŠ€ν…œ β€” 아첨 λ°©μ§€)

The hook system runs Python scripts at four Claude Code lifecycle events. It catches agreement bias before responses are generated and after they complete.

ν›… μ‹œμŠ€ν…œμ€ Claude Code 수λͺ… 주기의 λ„€ μ΄λ²€νŠΈμ—μ„œ Python 슀크립트λ₯Ό μ‹€ν–‰ν•©λ‹ˆλ‹€. 응닡이 μƒμ„±λ˜κΈ° μ „κ³Ό ν›„ λͺ¨λ‘μ—μ„œ λ™μ˜ 편ν–₯을 κ°μ§€ν•©λ‹ˆλ‹€.

Hook event Script Purpose (μ—­ν• )
SessionStart chavis_session_init.py Load sycophancy lessons from prior sessions (이전 μ„Έμ…˜ κ΅ν›ˆ λ‘œλ“œ)
UserPromptSubmit chavis_prompt_classify.py Classify risk level of incoming prompt (ν”„λ‘¬ν”„νŠΈ μœ„ν—˜ μˆ˜μ€€ λΆ„λ₯˜)
UserPromptSubmit chavis_strategic_challenge.py Fire Strategic Challenge Template on scope pivots (λ²”μœ„ 피봇 μ‹œ μ „λž΅ 도전 ν…œν”Œλ¦Ώ 생성)
Stop chavis_stop_audit.py Audit completed response for compliance drift (μ™„λ£Œλœ μ‘λ‹΅μ˜ κ·œμ • μ€€μˆ˜ 편차 감사)
Stop chavis_persistent_logger.py Append evaluation to audit trail (감사 좔적에 평가 μΆ”κ°€)

Full hook documentation: docs/HOOKS.md


Agents (μ—μ΄μ „νŠΈ)

Specialist agents are delegated via subagent_type. The pipeline selects the right agent automatically based on task classification.

μ „λ¬Έκ°€ μ—μ΄μ „νŠΈλŠ” subagent_type을 톡해 μœ„μž„λ©λ‹ˆλ‹€. νŒŒμ΄ν”„λΌμΈμ€ μž‘μ—… λΆ„λ₯˜λ₯Ό 기반으둜 μžλ™μœΌλ‘œ μ μ ˆν•œ μ—μ΄μ „νŠΈλ₯Ό μ„ νƒν•©λ‹ˆλ‹€.

Agent When activated (ν™œμ„±ν™” μ‹œμ )
explore Codebase discovery, 3+ search queries (μ½”λ“œλ² μ΄μŠ€ 탐색)
plan Implementation strategy before coding (μ½”λ”© μ „ κ΅¬ν˜„ μ „λž΅)
deep-research-agent Multi-source web research + synthesis (닀쀑 μ†ŒμŠ€ μ›Ή λ¦¬μ„œμΉ˜)
critic Review and falsifiability challenge (κ²€ν†  및 반증 κ°€λŠ₯μ„± 도전)
backend-architect Server-side system design (μ„œλ²„ μ‚¬μ΄λ“œ μ‹œμŠ€ν…œ 섀계)
frontend-architect UI/UX architecture decisions (UI/UX μ•„ν‚€ν…μ²˜ κ²°μ •)
python-expert Python-specific implementation (Python νŠΉν™” κ΅¬ν˜„)
quality-engineer Test coverage and quality gates (ν…ŒμŠ€νŠΈ 컀버리지 및 ν’ˆμ§ˆ 게이트)
security-engineer Threat modeling and vulnerability review (μœ„ν˜‘ λͺ¨λΈλ§ 및 취약점 κ²€ν† )
performance-engineer Profiling and optimization (ν”„λ‘œνŒŒμΌλ§ 및 μ΅œμ ν™”)
technical-writer Documentation and user guides (λ¬Έμ„œν™” 및 μ‚¬μš©μž κ°€μ΄λ“œ)
requirements-analyst Ambiguous request clarification (λͺ¨ν˜Έν•œ μš”μ²­ λͺ…μ„Έν™”)
system-architect Full system design across domains (전체 μ‹œμŠ€ν…œ 섀계)
sci-method Scientific hypothesis-evidence-validation workflow (과학적 κ°€μ„€-증거-검증 μ›Œν¬ν”Œλ‘œμš°)

Full agent documentation: docs/AGENTS.md


Testing (ν…ŒμŠ€νŠΈ)

cd tests/
python -m pytest test_hooks.py -v          # Hook system unit tests (ν›… μ‹œμŠ€ν…œ λ‹¨μœ„ ν…ŒμŠ€νŠΈ)
python -m pytest test_pipeline.py -v       # 6-stage pipeline integration tests (νŒŒμ΄ν”„λΌμΈ 톡합 ν…ŒμŠ€νŠΈ)
python -m pytest test_agents.py -v         # Agent routing tests (μ—μ΄μ „νŠΈ λΌμš°νŒ… ν…ŒμŠ€νŠΈ)

Documentation Index (λ¬Έμ„œ 인덱슀)

File Contents
docs/WORKFLOW.md 6-stage pipeline deep dive (6단계 νŒŒμ΄ν”„λΌμΈ 상세)
docs/AGENTS.md All specialist agents with examples (μ „λ¬Έκ°€ μ—μ΄μ „νŠΈ λͺ©λ‘ 및 μ˜ˆμ‹œ)
docs/HOOKS.md Anti-sycophancy hook architecture (아첨 λ°©μ§€ ν›… μ•„ν‚€ν…μ²˜)
docs/WHEN_TO_USE.md Use-case guide with scenarios (μ‚¬μš© μΌ€μ΄μŠ€ κ°€μ΄λ“œ)
docs/SETUP_GUIDE.md Step-by-step installation for beginners (초보자용 μ„€μΉ˜ κ°€μ΄λ“œ)
docs/MCP_SERVERS.md Per-server configuration reference (μ„œλ²„λ³„ μ„€μ • μ°Έμ‘°)

License

MIT License. See LICENSE for full text.

About

🧠 /coord β€” Intelligent Complex Task Workflow for Claude Code: 6-stage pipeline with Memory, MCP, Specialist Agents & Anti-Sycophancy hooks

Resources

License

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors