Claude Code Agent View: Watch and Steer Your Agents
Claude Code agent view puts every background session on one screen. How to open it with claude agents, dispatch, peek, attach, the shortcuts and the limits.
Josh Pocock7 min read
AI Coding
Agent view
Stride AI Academy
Claude Code agent view is one terminal screen for running many Claude Code sessions in the background. You open it with claude agents, type a prompt to start each session, and the list shows which ones are working, which need your input and which are done. You peek at a session to answer it, or attach to it for the full conversation. On November 22, 2026, agent view is a research preview, so the interface and shortcuts may change.
This guide covers how to open it, the dispatch, peek and attach loop, the keyboard shortcuts, running it from your shell, and how it fits with subagents and agent teams. Everything here was checked against Claude Code's agent view docs and the run agents in parallel comparison on November 22, 2026.
What agent view is for
The docs' comparison page puts it simply: use agent view when you have several independent tasks and want to hand them off, check status at a glance, and step in only when one needs you. Each row is a full Claude Code session running on your machine. You dispatch every one yourself, and each reports only to you.
That's different from the other ways to run Claude in parallel:
| Approach | Who coordinates | Where it runs |
|---|---|---|
| Subagents | Claude, inside one conversation | Your machine |
| Agent view | You, across many sessions | Your machine |
| Agent teams | A lead session that assigns work to teammates | Your machine. Experimental and off by default |
| Projects | Claude, from one conversation at claude.ai/code or the desktop app | The cloud, or your computer through Remote Control |
Subagents and teammates a session spawns don't show as separate rows in agent view. Our subagents tutorial and agent teams guide cover those two.
How to open agent view
claude agentsPress Esc to go back to your shell. Sessions keep running after you leave and reappear next time you open it. To show only sessions started under one folder:
claude agents --cwd ~/code/my-appIf you want plain claude to open agent view, run this in a normal session:
/config defaultToAgentsView=trueWith that on, pass a prompt to start a regular session instead, such as claude "fix the login test".
Dispatch, peek, attach
The whole tool is a loop of five moves.
- Dispatch. Type a prompt in the input at the bottom and press
Enter. Each prompt starts a new background session. It doesn't send a follow-up to an existing one. - Peek. Move to a row with the arrow keys and press
Space. The panel shows recent output, or the question the session is waiting on. Type a reply and pressEnter. Start the reply with!to send a Bash command. - Attach. Press
Enteror→on a row to open the full interactive conversation. - Detach. Press
←on an empty prompt, run/exit, or pressCtrl+Z. Detaching never stops the session. - Bring a session in. In a regular session, run
/bgor press←on an empty prompt to move it to the background.
To copy a session and try another direction, run /fork inside it. The copy runs as a new background session while the original keeps going.
What the list shows
Rows are grouped under Pinned, Ready for review, Needs input, Working and Completed. Each row also has a one-line summary, an age, and a pull request label such as #1234 when one is linked.
| State | Meaning |
|---|---|
| Working | Running tools or writing a response |
| Needs input | Waiting on you: a question, a permission decision, or a sandbox or MCP prompt |
| Idle | Ready for your next prompt |
| Completed | Finished successfully |
| Failed | Ended with an error |
| Stopped | Stopped by you, or ended while the background service was off |
You can filter by typing in the dispatch input. s:blocked shows sessions waiting on you, a:reviewer shows sessions running a named agent, and #1234 finds the session working on that pull request. Filters combine, such as s:blocked a:reviewer.
Agent view keyboard shortcuts
| Key | Action |
|---|---|
↑ / ↓ |
Move between rows |
Space |
Open or close the peek panel |
Enter / → |
Attach to the session |
← on an empty prompt |
Detach, or background the current session |
Ctrl+T |
Pin or unpin a session |
Ctrl+R |
Rename a session |
Ctrl+S |
Group by state or by directory |
Ctrl+F |
Find sessions by name |
Ctrl+G |
Open the dispatch prompt in your editor |
Ctrl+X |
Stop a session. Press again within two seconds to delete it |
? |
Show shortcuts in context |
Pinning matters more than it looks. Finished or waiting sessions left unattached for about an hour are stopped to free resources. Their conversations stay on disk and resume when you attach or reply, but a pinned session keeps its process running.
Run agent view from your shell
You can start background sessions without opening the screen:
claude --bg "investigate the flaky SettingsChangeDetector test"
claude --bg --name "flaky-test-fix" "investigate the flaky SettingsChangeDetector test"
claude --agent code-reviewer --bg "address review comments on PR 1234"--bg can't be combined with -p. For one-shot scripted runs, see our headless mode guide.
To set defaults for every session you dispatch from a view:
claude agents --permission-mode plan --model opus --effort highInside the dispatch input you can also type /model <name>, /effort <level> or /fast on for the next sessions.
Managing sessions from the shell:
| Command | What it does |
|---|---|
claude agents --json |
Prints sessions as JSON. Add --all for completed ones |
claude attach <id or name> |
Attaches in this terminal |
claude logs <id or name> |
Prints recent output |
claude stop <id> |
Stops a session |
claude rm <id> |
Removes a session from the list |
claude daemon status |
Shows the background service's state |
For scripts, the docs say to read state, status and waitingFor from claude agents --json --all, because the files under ~/.claude/jobs/ aren't a stable interface.
How parallel sessions avoid clashing
Dispatched sessions move into their own git worktree under .claude/worktrees/ before editing, so two sessions don't overwrite each other's files. Sessions that were already open when you backgrounded them keep editing in place, and outside a git repo sessions aren't isolated at all. Our worktrees guide explains how worktrees work.
To turn isolation off for a repository:
{ "worktree": { "bgIsolation": "none" } }Commit before you delete a session. The docs warn that worktrees Claude created are deleted with the session.
Notifications, sleep and shutdown
- While agent view is open, Claude Code sends a terminal notification when a session needs input, finishes or fails.
- Sessions survive sleep and reconnect when the machine wakes.
- Shutting down stops running sessions. Ones that stopped within 48 hours show as failed and restart when you reply.
- Auto-updates don't interrupt working, waiting or attached sessions.
To turn agent view off entirely, use the disableAgentView setting or CLAUDE_CODE_DISABLE_AGENT_VIEW. Admins can enforce it through managed settings.
Limits and cost
The docs list three limitations for the research preview:
- Rate limits apply. Background sessions use your subscription usage the same as interactive ones, so ten agents in parallel use quota about ten times as fast as one.
- Sessions are local. They run on your machine and stop if it shuts down.
- Worktrees go with the session. Claude-created worktrees are deleted when you delete the session in agent view.
That first point is the one to plan around. Check /usage while you work, and read our Claude usage limits guide before you dispatch ten sessions on Pro. For work that should keep running with your laptop closed, use Claude Code on the web instead.
Learn to build with Claude Code
The Stride AI Coding course teaches Claude Code and Codex for building client tools, written for non-programmers, with weekly live calls to review your builds. Many of the free guides cover Claude Code setups, and the Claude Code community is where members compare theirs.
FAQ
What is agent view in Claude Code?
It's a single terminal screen for managing many Claude Code background sessions. It shows what each one is doing, which need your input and which are done, and lets you start, answer, attach to and stop them. It's a research preview.
How do I open agent view?
Run claude agents in your terminal. Press Esc to return to your shell. To make plain claude open it, run /config defaultToAgentsView=true in a session.
What's the difference between agent view and subagents?
Subagents are workers Claude starts inside one conversation, and they report back to it. Agent view sessions are full sessions you dispatch yourself, and they report only to you.
Is agent view the same as agent teams?
No. In an agent team, a lead session splits the work, assigns it and keeps teammates in sync, and the teammates message each other. Agent teams are experimental and off by default. In agent view, you dispatch and track each session.
Do background sessions use more usage?
Yes. Each background session uses your subscription usage like an interactive session. The docs say ten sessions in parallel use quota roughly ten times as fast as one.
Do agent view sessions keep running after I leave?
Yes. A background service runs them, so they keep going after you press Esc and reappear the next time you open agent view. They survive sleep but stop if the machine shuts down.