---

name: collaborate-gemini description: Use when delegating coding tasks to Gemini CLI agent, when you need parallel AI execution, or when tasks benefit from Gemini's 1M+ context window.

Collaborate with Gemini CLI

Overview

Gemini CLI in headless mode (-p) cannot execute shell commands or write files. Use tmux send-keys to control Gemini interactively while monitoring via capture-pane.

Quick Reference

Target: {right} for split pane, gemini for named session.

Status Markers

Detect Gemini state by grepping capture-pane output:

Workflow

Copy# 1. Start Gemini
tmux split-window -h -d "cd ~/project && gemini --yolo"
sleep 5

# 2. Send task (TWO separate calls - critical!)
tmux send-keys -t {right} 'Build the app per PLAN.md'
tmux send-keys -t {right} Enter

# 3. Poll for completion
while true; do
  output=$(tmux capture-pane -t {right} -p -S -50)

  # Check if idle (task done)
  if echo "$output" | grep -q "Type your message"; then
    break
  fi

  # Handle loop detection
  if echo "$output" | grep -q "potential loop"; then
    tmux send-keys -t {right} '2'
    tmux send-keys -t {right} Enter
  fi

  sleep 10
done

# 4. Check result
tmux capture-pane -t {right} -p -S -200 | tail -100

Common Mistakes

Custom Commands

For repeating tasks, create .gemini/commands/:

Copy# .gemini/commands/improve-design.toml
[command]
description = "Improve app design"

[[steps]]
prompt = "Improve design: gradients, shadows, animations. Run build when done."

Call: tmux send-keys -t {right} '/improve-design' + Enter

When NOT to Use

• Read-only analysis → gemini -p "analyze" --output-format json
• Simple questions → direct API call
• Need deterministic output → headless with JSON schema