Claude Code agent view: run and manage parallel sessions from one screen

Last updated: November 10, 2026

TL;DR

Claude Code agent view is one screen for all your background sessions, opened with claude agents. Type a prompt and press Enter to dispatch a session, and each one appears as a row that shows whether it's working, waiting on you or done. Press Space to peek and reply, Enter to attach to the full conversation, and the left arrow on an empty prompt to step back to the list.

• 🖥️ Open it. Run claude agents in your shell, or press the left arrow on an empty prompt in any session.
• 🚀 Dispatch. Every prompt you type starts its own background session, one row per task.
• 🚦 Read the rows. Needs input and Ready for review sit at the top, so you look there first.
• 👀 Peek and reply. Space shows the latest output or the question a session is waiting on, and you answer inline.
• 🔗 Attach when you need more. Enter opens the full session, and the left arrow takes you back.
• 🌳 Let the worktrees do their job. Dispatched sessions move into their own git worktree before they edit.
• 🧹 Clean up. Commit first, because deleting a session in agent view deletes the worktree Claude made for it.

Agent view is a research preview. It arrived in Claude Code 2.1.139, which Anthropic announced on May 11, 2026.

Before you start

Agent view runs on your own machine and needs only an up-to-date Claude Code. Four things shape how a session goes:

1. A current version. If claude agents prints a count and a list of your custom subagents and then exits, agent view isn't available in your install. Run claude update. Anthropic's agent view docs list each later addition by version.
2. A plan that includes it. Anthropic's announcement lists Pro, Max, Ultra, Team, Enterprise and Claude API plans.
3. A git repository for anything that edits files. Dispatched sessions isolate their edits in git worktrees. Outside git, with no WorktreeCreate hook configured, sessions write straight into the working directory.
4. Usage headroom. Each background session draws on your subscription quota on its own, so 10 sessions use it roughly 10 times as fast as one.

Step 1: Open agent view

Run this from your shell:

claude agents

Agent view takes over the terminal with a table of sessions and an input at the bottom. If you haven't trusted the directory yet, the same workspace trust dialog claude shows comes first.

From inside a regular session, press the left arrow on an empty prompt. That moves the current conversation into the background and opens agent view with its row selected. Press Esc to go back to the conversation, or Esc from a fresh claude agents to return to your shell. Your sessions keep running either way.

The list shows every background session you've started, across all your projects. Press Ctrl+S to group it by directory instead of by state, so each project's sessions sit together.

To make agent view what claude opens with no arguments, run /config in a regular session and turn on Open agents view by default, or set it directly:

/config defaultToAgentsView=true

Step 2: Dispatch a session for each task

Type a task in the input and press Enter. A new background session starts on it and appears as a row. Type another prompt and press Enter, and a second session starts alongside the first. A new prompt never goes to an existing session as a follow-up.

Write each prompt as a task you'd hand to a colleague, with a clear finish line:

fix the failing login test in packages/auth and open a draft pull request

The input understands a few prefixes:

Sessions you dispatch from a claude agents you opened in your shell begin the way a new claude session in that directory would, permission mode included. An agent view you opened with the left arrow takes the directory's permissions.defaultMode first, then the mode of the session you came from. To change the model or effort for the sessions you dispatch next, type these in the input first (/effort needs Claude Code 2.1.299 or later):

/model opus
/effort high

Step 3: Read the list

Agent view groups sessions so the ones that need you come first: Ready for review, then Needs input, then Working and Completed. Each row shows the session's name, a one-line summary of what it's doing or asking, and its age. A session moves to Ready for review when its pull request needs review or its checks are failing.

The icon at the start of each row carries the state:

A Haiku-class model writes each row's summary at the end of every turn, and rewrites it every few minutes during a long turn, through your normal provider. When a session opens or works on a pull request, a #1234 label appears at the right edge of the row. It turns green when checks pass and nothing blocks the review, and purple once it's merged.

Step 4: Peek and reply without leaving the list

Select a row with the arrow keys and press Space. The peek panel shows the question a waiting session is asking, a finished session's result, or a working session's latest status, plus any linked pull requests.

Type a reply and press Enter to send it. Prefix it with ! to send a shell command instead. Choices a session gives you are numbered, so press the number, then Enter. A reply of exactly /stop stops the session at once.

A reply doesn't answer a permission prompt or another dialog, such as a sandbox prompt. It waits in the queue instead. To answer the dialog, attach.

Use the up and down arrows to move to the next session's peek without closing the panel.

Step 5: Attach for the full conversation, then detach

Press Enter or the right arrow on a row to attach. The session takes over the terminal as a full interactive Claude Code session, and Claude posts a short recap of what happened while you were away. Commands and shortcuts work as they do anywhere else.

Press the left arrow on an empty prompt, or run /exit, to detach and return to the list. Detaching never stops a background session. To end one from inside it, run /stop.

Step 6: Bring sessions you already have into agent view

Three ways move work you already started into the list:

• /bg or /background sends the current conversation to the background and frees your terminal. Add a prompt, such as /bg run the test suite and fix any failures, to give one more instruction first.
• The left arrow on an empty prompt does the same in one step and opens agent view.
• /fork copies the conversation into a new background session while you keep working in the original. After the fork, the two conversations are independent.

A session you send with /bg or the left arrow keeps editing files where it already was. A /fork copy works like a dispatched session and makes its code changes in a worktree of its own.

Step 7: Review the results and clean up

Before a dispatched session edits files, Claude moves it into a git worktree under .claude/worktrees/, so parallel sessions can read the same checkout while each writes to its own. When a session has made changes in a worktree it entered, Claude commits them, pushes the branch when the repository has a remote, and opens a draft pull request when the task calls for one. It never pushes to main or master, force-pushes or merges, and your own instructions in the task, CLAUDE.md or memory take precedence.

For a task that ends in a pull request, watch the row's label and review the pull request when it turns green. Then delete the session: press Ctrl+X to stop it and Ctrl+X again within two seconds to delete it.

Deleting in agent view removes the worktree Claude created for that session, uncommitted changes included, so commit anything you want first. Claude Code refuses the delete when the worktree has commits it can't confirm are saved elsewhere, names the branch, and lets you push or delete again to discard them. The conversation itself stays on your machine, and typing /resume in the agent view input brings it back as a background session.

Manage background sessions from the shell

Every background session has a short ID, which Claude prints when you start one from your shell with Claude Code's background flag. Anthropic's agent view docs show that command and its options. The first five commands below take that ID, and the last checks the background service itself:

For a script or another program, the JSON output of claude agents is the supported way to read each session's state, status and waitingFor fields. Read that rather than the files under ~/.claude/jobs/, which can change.

Settings that change agent view

When agent view is the wrong tool

Agent view is for independent tasks you hand off and check on later, and its sessions report only to you. Anthropic's guide to running agents in parallel sets it beside four other options:

Our guide to how to run Claude Code agents in parallel compares all five. For a single change spread across a codebase, our Claude Code /batch guide covers the steps, and our Claude Code worktrees guide covers worktrees for sessions you run yourself.

When something goes wrong

• Sessions show as failed after a restart. Shutting down stops running sessions, while sleep doesn't. A session that last made progress within 48 hours shows as failed: attach or reply and it restarts from its saved conversation. Past 48 hours it shows as stopped, and pressing Enter twice on the row resumes it.
• Sessions stop when the background service stops. A separate supervisor process hosts every session. If it stops, your sessions stop about a minute later on macOS and Linux unless a new one starts, and running claude agents starts it.
• .claude/worktrees/ keeps growing. Some deletes keep a worktree on purpose. List leftovers with git worktree list and remove each with git worktree remove <path>.
• Operation not permitted on Desktop, Documents or Downloads (macOS). The background host is its own process and asks for folder access separately from your terminal. Grant it under System Settings > Privacy & Security > Files and Folders.

Frequently asked questions

What is Claude Code agent view?

Agent view is a screen that lists every Claude Code session running in the background, grouped by whether it's working, needs your input or is done. You open it with claude agents, dispatch a new session by typing a prompt, and peek, reply or attach to any row. It's a research preview.

How do I open agent view in Claude Code?

Run claude agents in your shell, or press the left arrow on an empty prompt inside any Claude Code session. To have plain claude open it every time, turn on Open agents view by default in /config or set defaultToAgentsView to true.

Do agent view sessions keep running when I close the terminal?

Yes. A separate supervisor process runs background sessions, so you can close agent view or your shell and the work continues. Sessions survive sleep but stop if the machine shuts down. Attach or reply to a stopped session to restart it from its saved conversation.

Does agent view use more of my usage limit?

Each background session uses your subscription quota the same way an interactive session does, so 10 sessions running at once use it roughly 10 times as fast as one. Row summaries add one short Haiku-class request at the end of each turn and every few minutes during a long one.

Do agent view sessions get their own git worktree?

In a git repository, sessions you dispatch from agent view, or start in the background from your shell, move into a worktree under .claude/worktrees/ before editing. A session you background yourself with /bg or the left arrow keeps editing where it was. Set worktree.bgIsolation to "none" to turn isolation off.

What's the difference between claude agents and /agents?

claude agents, run in your shell, opens agent view, the list of background sessions. /agents, typed inside a session, is a different command: it prints a reminder about creating and managing subagents and where their files live.

How do I start a background session without opening agent view?

Run /bg inside any session to send it to the background, or start one from your shell with Claude Code's background flag, shown in Anthropic's agent view docs. Either way the session appears as a row, and the shell command prints a short ID for claude attach, claude logs and claude stop.

Can I turn off agent view?

Yes. Set disableAgentView to true in your settings, or set the CLAUDE_CODE_DISABLE_AGENT_VIEW environment variable. Administrators can enforce it through managed settings.

Give every session the same rules

Agent view makes it cheap to run 10 sessions at once. Every session you dispatch starts in your working directory and reads the same CLAUDE.md at launch, so the rules you've written down decide how consistent 10 parallel results come out. Our guide to compound engineering for Claude Code shows how to build those rules up one fix at a time.