Claude Code Team Setup

Elevate Claude Code from 50 to 95 points with this shared team configuration.

Based on 4 core systems built by a developer who used Claude Code hardcore for 6 months.

한국어 문서 보기

---

Why You Need This

Without proper setup, Claude Code suffers from:

• Goldfish memory: Forgets decisions made earlier as conversations grow longer
• Context blindness: Doesn't know why things were built a certain way, goes in the wrong direction
• Missed bugs: Skips error checking after code edits
• Single-agent bottleneck: One AI doing everything instead of specialists working in parallel

This setup solves all 4 problems systematically.

---

How It Works

Overall Flow

Project starts
    ↓
Run /init → generates CLAUDE.md / PLAN.md / CONTEXT.md / CHECKLIST.md
    ↓
Claude reads all 4 files as a navigation compass before starting work
    ↓
Code is written / modified
    ↓
Claude finishes a turn → automatic quality check runs (tsc / eslint / ruff / etc.)
    ↓
Errors detected → Claude immediately recognizes and fixes them
    ↓
CHECKLIST.md auto-updated

4 Core Systems

① Auto Manual System (OMC Plugin)

Claude detects the type of work and automatically activates relevant guidelines.

• Hooks fire at session start, on every prompt, and before/after tool use
• "Editing a TypeScript file" → TypeScript rules auto-activated
• "Security-related work" → Security guidelines auto-activated

② Task Memory System (/init command)

Solves the goldfish memory problem with 3 external files.

File	Role	Contents
PLAN.md	Blueprint	What to build, architecture, completion criteria
CONTEXT.md	Decision log	Why this tech was chosen, constraints, references
CHECKLIST.md	Progress tracker	What's done, in progress, and up next

Claude reads these files at the start of every session to restore context. Even after losing memory mid-conversation, it can immediately get back on track.

③ Automatic Quality Check (Stop Hook)

Runs automatically each time Claude finishes a turn. Even if 10 files were edited, it only runs once at the end.

File Type	Check
.ts / .tsx	tsc --noEmit — TypeScript type errors
.js / .jsx	ESLint (only if config file exists)
.py	python3 -m py_compile + ruff
.sh	bash -n — Shell syntax errors
.java	Gradle compile reminder message

When errors are found, Claude recognizes them on the next turn and fixes them automatically.

④ Specialist Agents (OMC Agent Catalog)

Instead of one AI doing everything, specialized AI teammates handle each role.

Requirements analysis (analyst)
    → Task planning (planner)
    → Code implementation (executor)
    → Test writing (test-engineer)
    → Quality review (quality-reviewer)
    → Completion verification (verifier)

---

Installation

⚠️ Follow the order exactly. OMC must be installed first to prevent CLAUDE.md from being overwritten.

Step 1 — Add OMC Marketplace (inside Claude Code)

Open Claude Code and run:

/plugin marketplace add https://github.com/Yeachan-Heo/oh-my-claudecode

Step 2 — Install OMC Plugin (inside Claude Code)

/plugin install oh-my-claudecode

Restart Claude Code after installation.

Step 3 — Run the Script (terminal)

git clone https://github.com/iops-leo/claude-code-setup.git ~/.claude-code-setup && chmod +x ~/.claude-code-setup/setup.sh && ~/.claude-code-setup/setup.sh

What setup.sh handles automatically:

• /init command → ~/.claude/commands/init.md
• Auto quality check script → ~/.claude/hooks/quality-check.sh
• Hook registration → ~/.claude/settings.json
• Project Memory System rules → ~/.claude/CLAUDE.md

Step 4 — Restart

Restart Claude Code and all settings will take effect.

---

Updating (Existing Users)

When the configuration changes, run this from anywhere:

~/.claude-code-setup/update.sh

Already-installed items are automatically skipped. Only new additions are applied.

---

Usage

Starting a New Project or Feature

/init

For existing projects, say "analyze it". For new projects, answer 4 questions — then these 4 files are generated:

project-folder/
├── CLAUDE.md      ← Project-specific AI instructions (conventions, rules)
├── PLAN.md        ← Blueprint
├── CONTEXT.md     ← Decision background
└── CHECKLIST.md   ← Progress status

Claude automatically reads these files at the start of every session.

Note: PLAN.md / CONTEXT.md / CHECKLIST.md are personal working notes and should not be committed to Git. Running /init automatically adds them to .gitignore.

During Normal Work

Everything runs automatically — no extra steps needed.

• After Claude finishes work → quality check runs automatically
• If errors are found → Claude auto-fixes on the next response
• When making important decisions → prompted to log in CONTEXT.md

When You Need to Restore Context

When conversations get too long or you start a new session:

Read PLAN.md, CONTEXT.md, CHECKLIST.md and catch me up on the current state

---

File Structure

~/.claude-code-setup/             ← Fixed install path
├── README.md                     ← This file (English)
├── README.ko.md                  ← Korean version
├── setup.sh                      ← Initial install script
├── update.sh                     ← Update script (run from anywhere)
├── claude-md-addition.md         ← Rules appended to ~/.claude/CLAUDE.md
├── commands/
│   └── init.md                   ← /init command definition
└── hooks/
    └── quality-check.sh          ← Stop hook — auto quality check script