How to keep track of several coding agents at once: status monitors vs orchestrators
The hard part of running three agents isn’t the code. It is noticing that one of them has been waiting for your approval for twenty minutes.
SHORT ANSWER
Give each agent its own task and its own folder or git worktree. Then decide what you actually need: an orchestrator starts agents and splits work between them; a status monitor watches the agents you already run and tells you which one is working, waiting or done. Claude Code’s claude agents view does this for Claude Code sessions in a terminal. For Claude Code, Codex, Cursor and other agents together, with their permission requests in one queue, you need a cross-agent status layer such as Pet Island.
Before anything else: one task, one folder
Two agents editing the same files will undo each other’s work. Give each agent a clear task and its own working copy. Git worktrees are the simplest way: each worktree is a separate folder on its own branch, sharing one repository.
Start one agent in each folder. Merge the branches when they are done.
Two kinds of tools: orchestrators and status monitors
| Orchestrator / agent manager | Status monitor / overlay | |
|---|---|---|
| Starts agents for you | Yes | No, you start them as usual |
| Splits work between agents | Yes | No |
| Where agents run | Inside the tool | In your own terminals and editors |
| Shows who is working, waiting or done | Usually, inside its own window | Yes, always on screen |
| Answer permission requests | Inside its own window | From the overlay, without switching windows |
| Changes your workflow | A lot | No |
If you want a tool to plan and hand out the work, you want an orchestrator. If you already know what each agent should do and only keep losing track of them, a status monitor is enough, and it works with the editors and terminals you already use.
What the agents show on their own
- Claude Code agent view.
claude agentsis “one screen for all your background sessions”, with states for working, needs input, idle, completed, failed and stopped. While it is open, Claude Code notifies you when a background session needs input, finishes or fails. It covers Claude Code sessions only, and it is a research preview. - Codex. The TUI notifications (
approval-requested,agent-turn-complete) tell you when a Codex session waits or finishes. In Windows Terminal they arrive as the terminal bell. - Terminal tabs. Windows Terminal can flash the taskbar icon on a bell (
"bellStyle": "all"), but not which tab rang.
Each of these sees only its own agent. Nothing built in shows Claude Code, Codex and Cursor side by side.
One status light per agent, one queue for approvals
Pet Island is a status and approval layer for Windows and macOS. It does not start agents or split work, so it is not an orchestrator or an IDE. It sits on top of the agents you already run, as an always-on-top island at the top of the screen.
- Every agent gets a status light: working, waiting for you, or done. With several running, the island shows a count like “3 agents working”.
- Permission requests from different agents line up in one queue. Each card shows the command or the diff; allow with Ctrl Y, deny with Ctrl N.
- Jump opens the exact terminal tab or editor window of an agent.
- Works with 22 coding agents, including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Grok CLI and Antigravity. Many of them can be approved right from the island; the agents page lists what works with each one.
Which one do you need?
- One agent at a time: the agent’s own notifications are enough.
- Several Claude Code sessions in the background: start with
claude agents. - Different agents in parallel (Claude Code, Codex, Cursor): a cross-agent status monitor saves the most time.
- You want a tool to plan and hand out tasks: look at an orchestrator, and keep a status monitor for the agents you run by hand.
FAQ
How do I monitor multiple coding agents at once?
Run each agent in its own folder or git worktree, then use one status view for all of them. Claude Code’s agent view covers Claude Code sessions; Pet Island shows Claude Code, Codex, Cursor and 19 other agents together.
What is the difference between an agent orchestrator and a status monitor?
An orchestrator starts agents and splits the work between them. A status monitor watches the agents you already run and shows which one is working, waiting or done.
Is Pet Island an orchestrator?
No. It does not start agents or split work. It shows their status and lets you answer their permission requests from one place.
Can I approve Claude Code and Codex requests from one place?
Yes. In Pet Island their requests line up in one queue on the island, with the command or diff, and you allow or deny each one there.
Is Pet Island open source?
No. It is a paid app: $8.99 once for the first computer, with a 7-day free trial.
Sources
Every setting and behavior on this page was checked against these pages on October 26, 2026. Agents change fast; if something no longer matches, the official page wins.
- Claude Code docs: Agent view
- OpenAI: Codex notifications
- Git documentation: git-worktree
- Microsoft Learn: Windows Terminal profile settings (bellStyle)