Claude Code Integration

Guidance for interacting with Claude Code sessions running in iTerm2.

When to Use

• Detecting which iTerm2 sessions are running Claude Code
• Checking the status of Claude Code sessions (idle, working, waiting for input)
• Coordinating multiple Claude Code sessions
• Monitoring Claude Code session activity
• Setting up multi-agent Claude Code workflows

When NOT to Use

• For general terminal automation (use session-orchestration skill instead)
• When not working with Claude Code specifically
• For Claude API interactions (this is for the CLI tool in terminals)

Quick Start

# List sessions and identify Claude Code instances
it2 session list --format json | jq '.[] | select(.name | contains("claude"))'

# Get Claude-specific session info
it2 session claude "$SESSION_ID"

# Get rich Claude Code status
it2 session claude-status "$SESSION_ID"

# Check if session has a modal dialog (permission prompt, etc.)
it2 session has-modal "$SESSION_ID"

Detecting Claude Code Sessions

By Process Name

# Find sessions running Claude
it2 session list --format json | jq -r '.[] |
  select(.foreground_job.command | test("claude"; "i")) | .session_id'

By Session State

# Get comprehensive state analysis
it2 session get-state "$SID"

Claude Session Indicators

A session is likely running Claude Code if:

• Process name contains "claude" or "node" (Claude Code's runtime)
• Screen content shows Claude Code UI elements (╭─, tool use blocks, etc.)
• Session has specific environment variables set

Session Status Detection

Modal Detection

Claude Code shows modal dialogs for permissions, confirmations, etc.

# Check for modal
if it2 session has-modal "$SID"; then
  echo "Session has pending modal"
fi

Activity Detection

# Check if session is actively working
it2 session is-active "$SID"

# Suggest intervention action
it2 session suggest-action "$SID"

Session Linking

Link iTerm2 sessions to their Claude Code session files:

# Get Claude session link info
it2 session claude "$SID"

# This reveals:
# - Session file path
# - Working directory
# - Session metadata

Multi-Claude Workflows

Setting Up Parallel Claude Sessions

# Create splits for multiple Claude instances
MAIN=$ITERM_SESSION_ID
RESEARCH=$(it2 session split --vertical -q)
IMPLEMENT=$(it2 session split --horizontal -q)

# Badge each for identification
it2 session set-badge "$MAIN" "$(echo $MAIN | cut -c1-8)\nMain"
it2 session set-badge "$RESEARCH" "$(echo $RESEARCH | cut -c1-8)\nResearch"
it2 session set-badge "$IMPLEMENT" "$(echo $IMPLEMENT | cut -c1-8)\nImplement"

# Start Claude in each
it2 session send-text "$RESEARCH" "claude --continue"
it2 session send-text "$IMPLEMENT" "claude"

Monitoring Multiple Sessions

# Watch all sessions for state changes
it2 session watch --all

Screen Content Analysis

Getting Claude's Output

# Get visible screen
it2 session get-screen "$SID"

# Get more context from buffer
it2 session get-buffer "$SID" --lines 200

Parsing Claude UI Elements

Claude Code's TUI has recognizable patterns:

• ╭─ / ╰─ - Message box borders
• ● - Working indicator
• ? - Question/prompt indicator
• Tool use blocks with specific formatting

Best Practices

1. Always check for modals before sending input to Claude sessions
2. Use badges with session ID prefix for visual tracking
3. Monitor state before automated interactions
4. Verify screen content matches expected state before proceeding
5. Use quiet flags (-q) when capturing session IDs for scripts
6. Include response hints when messaging other sessions - always tell them how to respond back (see workflows/multi-agent.md)

Advanced Usage

See references/state-detection.md for detailed state analysis. See references/ui-patterns.md for parsing Claude Code output. See workflows/multi-agent.md for coordinating multiple Claude instances.