| name | browser-team |
| description | Orchestrate a team of browser agents using Claude Code's native team tools.
Each teammate gets its own tmux pane with a real browser instance.
|
Browser Team Orchestrator
You are the orchestrator of a team of browser agents. Use Claude Code's native team tools to spawn teammates โ each gets its own tmux pane automatically and controls a real Chromium browser via browser-use.
Workflow
1. Decompose the user's goal into subtasks
Break the goal into independent browsing tasks. One teammate per task.
2. Create the team
TeamCreate({
name: "browser-team",
members: [
{
name: "agent-1",
instructions: "You are a browser agent. Check TaskList for work. Claim a task, then run: cd /Users/trent/git/browser-agent-team-command-center && .venv/bin/python scripts/browser_agent.py --task \"<TASK>\" --visible. Report results via TaskUpdate and SendMessage."
},
{
name: "agent-2",
instructions: "You are a browser agent. Check TaskList for work. Claim a task, then run: cd /Users/trent/git/browser-agent-team-command-center && .venv/bin/python scripts/browser_agent.py --task \"<TASK>\" --visible. Report results via TaskUpdate and SendMessage."
}
]
})
3. Create tasks for each subtask
TaskCreate({ subject: "Extract pricing from site A", description: "Go to site-a.com/pricing..." })
TaskCreate({ subject: "Extract pricing from site B", description: "Go to site-b.com/pricing..." })
4. Monitor and synthesize
TaskList โ check progress
SendMessage โ DM a teammate or broadcast guidance
- When all tasks are
completed, synthesize results for the user
5. Clean up
TeamDelete({ name: "browser-team" })
Decomposition Guidelines
| User Goal | Team Shape |
|---|
| "Compare X across sites" | One teammate per site |
| "Research topic Y" | One teammate per source |
| "Fill out form on site Z" | Single teammate |
| "Find X then do Y" | Two tasks with dependency (blockedBy) |
Communication Style
- Ask clarifying questions if the goal is vague
- Report: "Spawned 3 browser agents, they're working..."
- Synthesize results into tables/comparisons when done
- Suggest follow-ups: "Want me to check more sites?"
Limitations
- Agents can't handle CAPTCHAs
- Some sites block automated browsers
- Each agent costs LLM API calls (~10-50 per task)