3. Setup & Configuration

Prerequisites

Required:

• ✅ Claude Code v2.1.32 or later
• ✅ Opus 5 model (/model opus)
• ✅ Git repository (for coordination)

Recommended:

• ✅ Understanding of Sub-Agents
• ✅ Familiarity with git workflows
• ✅ Budget awareness (token-intensive feature)

Method 1: Environment Variable

Simplest approach: set env var before starting Claude Code:

# Enable agent teams for this session
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1

# Start Claude Code
claude

Persistent setup (bash/zsh):

# Add to ~/.bashrc or ~/.zshrc
echo 'export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1' >> ~/.bashrc
source ~/.bashrc

Method 2: Settings File

Persistent configuration: edit ~/.claude/settings.json:

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

Advantages:

• ✅ Persistent across sessions
• ✅ No need to remember env var
• ✅ Can be version-controlled in dotfiles

After editing, restart Claude Code for changes to take effect.

Verification

Check if enabled:

# In Claude Code session
> Are agent teams enabled?

Claude should confirm:

"Yes, agent teams are enabled (experimental feature). I can spawn multiple agents to work in parallel when appropriate."

Alternative verification (check env var):

echo $CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS
# Should output: 1

Multi-Terminal Setup

Pattern (from practitioner reports):

# Terminal 1: Research + bugfix
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
claude

# Terminal 2: Business ops
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
claude

# Terminal 3: Infrastructure
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
claude

Benefits:

• Isolation of contexts (research vs execution vs setup)
• Parallel progress on independent workstreams
• Reduced context switching cognitive load

Note: This differs from automatic teammate spawning. Here you're manually creating multiple team lead sessions, and each can spawn its own teammates.