Tutorial Overview

Series index: gstack Tutorial Series

/pair-agent lets multiple AI agents share a browser. Each gets its own tab, isolated from others.

What you will learn

• ✅ /pair-agent for cross-agent coordination
• ✅ Scoped tokens and tab isolation
• ✅ /setup-browser-cookies for authenticated testing
• ✅ Real-world coordination patterns

Why Multi-Agent Browser?

Single agent limitation

One AI agent can only do one thing at a time. But often you need:

• Compare two pages simultaneously
• Test user interactions between accounts
• Coordinate frontend and backend testing

Multi-agent solution

flowchart TD
    A[Claude Code] --> B[GStack Browser]
    C[OpenClaw] --> B
    D[Codex] --> B
    B --> E[Tab 1: Claude]
    B --> F[Tab 2: OpenClaw]
    B --> G[Tab 3: Codex]

    style B fill:#e1f5ff

/pair-agent — Cross-Agent Coordination

How it works

1. Launch browser — GStack Browser opens
2. Print instructions — Block to paste into other agent
3. Exchange credentials — One-time setup key for session token
4. Create isolated tab — Each agent gets its own tab
5. Watch together — See both agents working

Usage

/pair-agent

Output:

GStack Browser launched. Share with another agent:

[Paste this block into your other agent's chat]

---BEGIN PAIR BLOCK---
SESSION_ID: abc-123
TOKEN: scoped_token_xyz
URL: http://localhost:9222
---END PAIR BLOCK---

The other agent will:
1. Receive the pair block
2. Connect to the browser
3. Get its own isolated tab
4. Be able to browse independently

Security features

Supported agents

• Claude Code
• OpenClaw
• Hermes
• Codex
• Cursor
• Any agent that can curl

/setup-browser-cookies — Session Manager

What it does

Imports cookies from your real browser into the AI browser session.

Supported browsers

• Google Chrome
• Arc
• Brave
• Microsoft Edge

Usage

/setup-browser-cookies

Output:

Select browser source:
[1] Chrome
[2] Arc
[3] Brave
[4] Edge

Enter selection: 1

✅ Imported 47 cookies from Chrome
- github.com: logged_in, user_session
- google.com: SID, HSID, SSID
- ...

Your AI agent can now access authenticated pages.

When to use

• Testing authenticated features
• Preserving login state
• Accessing paid/protected content
• Personal automation

Security note

Cookies are stored in memory only for the session. They are never written to disk or transmitted.

Coordination Patterns

Pattern 1: Parallel comparison

Two agents compare pages side-by-side:

Agent 1: Browse staging site, find layout issues
Agent 2: Browse production site, compare

Result: Side-by-side comparison with screenshots

Pattern 2: User interaction testing

Test interactions between accounts:

Agent 1: User A sends message
Agent 2: User B receives message

Result: Real-time interaction test

Pattern 3: Frontend + Backend

One agent tests UI, another tests API:

Agent 1: Click through frontend flows
Agent 2: Monitor API responses

Result: Full-stack validation

Example: Testing Chat Application

Setup

# Agent 1 (Claude Code)
/pair-agent

# [Copy pair block]

# Agent 2 (OpenClaw)
[Paste pair block]

Agent 1 (User A)

/setup-browser-cookies  # Import login cookies
/browse https://chat.example.com
$B type "#message" "Hello from User A"
$B click "button[type='submit']"
$B screenshot

Agent 2 (User B)

/setup-browser-cookies  # Import different account cookies
/browse https://chat.example.com
$B wait ".message"
$B text ".message:last-child"
# Output: "Hello from User A"
$B screenshot

Result

Both agents see the same chat, from different accounts, in real time.

Skill Quick Reference

Summary

Multi-agent browser enables:

1. Cross-agent coordination — Multiple AI agents, one browser
2. Tab isolation — Each agent in its own tab
3. Cookie import — Test authenticated pages
4. Security — Scoped tokens, activity attribution

Key takeaways

• ✅ Use /pair-agent for multi-agent testing
• ✅ Import cookies for authenticated access
• ✅ Each agent gets isolated tab
• ✅ Watch agents work in real time

Series navigation:

• ← Previous: Tutorial 6A: Browser Basics
• → Next: Tutorial 7: Code Review
• Back: Series Index