Skip to content

Agent View: claude agents

Research preview v2.1.139+

Agent view, opened with claude agents, is a single dashboard for every background Claude Code session on your machine. Each row is a full conversation that keeps running without a terminal attached. You dispatch new sessions from the input at the bottom, peek at any row to see its last output, attach for the full transcript, and step in only when one needs you.

It debuted on May 11, 2026 alongside Claude Code v2.1.139 as a research preview on Pro, Max, Team, Enterprise, and Claude API plans. Verify your version with claude --version before relying on it.

  1. Open agent view. From your shell:

    Terminal window
    claude agents

    The view takes over the terminal. Press Esc at any time to return to your shell — sessions keep running and reappear next time you open the view.

  2. Dispatch a session. Type a task in the input at the bottom and press Enter. A new background session starts and shows up as a row indicating whether it’s working, waiting on you, or done. Every prompt creates its own new session, so typing two prompts launches two parallel sessions.

  3. Peek and reply. Select a row with the arrow keys and press Space to open the peek panel — most recent output, or the question the session is blocked on. Type a reply and press Enter to send it without leaving agent view.

  4. Attach and detach. Press Enter or on a row to attach to the full conversation. Press on an empty prompt to detach back to the table. Detaching never stops a background session.

  5. Bring an existing session in. Inside any open session, run /bg to background it, or press on an empty prompt to background and open agent view in one step.

You can use claude agents as your main entry point instead of claude: dispatch everything from the dashboard, attach when you want the full transcript, press to come back.

Each row starts with an icon whose color shows the session state and shape shows whether the underlying process is alive.

StateIconMeaning
WorkingAnimatedClaude is actively running tools or generating a response
Needs inputYellowClaude is waiting on a question or permission decision from you
IdleDimmedNothing to do; ready for your next prompt
CompletedGreenThe task finished successfully
FailedRedThe task ended with an error
StoppedGreyStopped with Ctrl+X or claude stop
ShapeMeaning
/ Session process is alive; replies immediately
Process has exited; peek/reply/attach still work and restart from where it left off
A /loop session sleeping between iterations. Row shows run count and countdown

Background sessions don’t need any terminal open. A separate supervisor process runs them, so you can close agent view, close your shell, or start a new interactive session and the dispatched work keeps going.

When a session opens a pull request, a status dot appears at the right edge of the row, linked to the PR in terminals that support hyperlinks.

Dot colorPull request status
YellowWaiting on checks or review, or checks failed
GreenChecks passed and no review is blocking
PurpleMerged
GreyDraft or closed

For most autonomous tasks this row is the entire interface — review and merge the PR when the dot turns green.

Press Space on a selected row to open the peek panel. It shows the session’s most recent output, any blocking question, and any open pull requests.

Type in the peek panel and press Enter to send a reply. When the session asks a multiple-choice question, press a number key to pick. For other blocked sessions, press Tab to fill the input with a suggested reply you can edit. Prefix a reply with ! to send a Bash command instead.

Use and to peek at adjacent sessions without closing the panel; attaches.

Press Enter or on a row to attach. Agent view is replaced by the full interactive session, exactly as if you had run claude in that directory. Claude posts a short recap of what happened while you were away.

While attached, the session behaves like any other Claude Code session — every command, shortcut, and feature works. Press on an empty prompt to detach. If a dialog has focus and isn’t responding to , press Ctrl+Z to detach immediately.

, Ctrl+C, Ctrl+D, Ctrl+Z, and /exit all leave the background session running. To end a session from inside it, run /stop. You can disable the shortcut in /config if you keep hitting it accidentally.

Agent view groups sessions so the ones that need input are at the top — Ready for review and Needs input above Working and Completed. Press Ctrl+S to switch grouping to by-directory and back. Your choice persists across runs.

Within a group:

  • Ctrl+T — pin a session to the top
  • Shift+↑ / Shift+↓ — reorder sessions
  • Ctrl+R — rename a session
  • Enter on a group header — collapse it
  • Ctrl+X — stop a session; press again within 2 seconds to delete it (including its worktree)

Type in the dispatch input to filter the list instead of dispatching:

FilterShows
a:<name>Sessions running the named agent
s:<state>Sessions in the given state, e.g. s:working, s:blocked
#<number> or PR URLThe session working on that pull request

Press ? in agent view to see every shortcut in context.

ShortcutAction
/ Move between rows
EnterAttach to selected session, or dispatch if there’s text in the input
SpaceOpen / close the peek panel
Shift+EnterDispatch and attach immediately
Attach to the selected session
Alt+1..Alt+9Attach to session 1–9 in the current group
TabOn an empty input: browse all subagents. Otherwise: apply highlighted suggestion
Ctrl+SSwitch grouping between state and directory
Ctrl+TPin or unpin the selected session
Ctrl+RRename the selected session
Ctrl+GOpen the dispatch prompt in $EDITOR
Ctrl+XStop the session; press again within 2 seconds to delete it
Shift+↑ / Shift+↓Reorder the selected session
EscClose peek panel, clear input, or exit
Ctrl+CClear input; press twice to exit
?Show all shortcuts

Three places to start a background session:

Type a prompt in the input and press Enter. The session is named automatically from the prompt; rename it later with Ctrl+R. Paste an image into the prompt to include a screenshot or diagram.

InputEffect
<agent-name> <prompt>If the first word matches a subagent name, that subagent runs as main agent
@<agent-name>Mention a subagent anywhere in the prompt to run it as main agent
@<repo>Mention a sibling repo to run the session there
/<skill>Suggest skills to dispatch as the prompt
#<number> or PR URLIf a session is already working on that PR, select it instead of dispatching
Shift+EnterDispatch and immediately attach
/bg

Move the current conversation into a background session. Pass a final instruction in the same line — /bg run the test suite and fix any failures. Backgrounding starts a fresh process that resumes from the saved conversation; Claude asks for confirmation if subagents or background commands are still running.

Terminal window
claude --bg "investigate the flaky SettingsChangeDetector test"

To run a specific subagent as the session’s main agent:

Terminal window
claude --agent code-reviewer --bg "address review comments on PR 1234"

After backgrounding, Claude prints the session’s short ID and the commands for managing it.

Every background session has a short ID (printed by claude --bg, also the directory name under ~/.claude/jobs/).

CommandPurpose
claude agentsOpen agent view
claude attach <id>Attach to a session in this terminal
claude logs <id>Print the session’s recent output
claude stop <id>Stop a session. Also accepts claude kill
claude respawn <id>Restart a stopped session with its conversation intact
claude respawn --allRestart every stopped session
claude rm <id>Remove a session. Cleans up its worktree if there are no uncommitted changes

These are how you script agent view — for example, a cron job that respawns sessions after the laptop wakes from sleep.

Every background session — agent view, /bg, or claude --bg — starts in your working directory. Before editing files, Claude moves the session into an isolated git worktree under .claude/worktrees/, so parallel sessions read the same checkout but each writes to its own. Outside a git repository, sessions write directly and aren’t isolated from each other.

The worktree is removed when you delete the session, so merge or push the changes you want to keep before deleting. To make a subagent always run in its own worktree regardless of how it was started:

---
name: code-reviewer
isolation: worktree
---

A dispatched session reads its settings and permission mode from the directory it runs in — the same as claude started there. To override from the shell:

Terminal window
claude --bg --permission-mode plan "draft the implementation plan for the new pricing page"

Using bypassPermissions or auto this way is refused until you’ve accepted that mode by running claude with it once interactively, since those modes let a session you aren’t watching act without approval.

Background sessions are hosted by a per-user supervisor process, separate from your terminal and from agent view. It starts automatically the first time you background a session or open agent view; you don’t manage it directly.

A session that’s working, waiting for input, or has a terminal attached keeps its process running. Once a session finishes and sits unattached for about an hour, the supervisor stops its process to free resources. The transcript and state stay on disk; the next time you peek, reply, or attach, the supervisor starts a fresh process from where it left off.

The supervisor watches the installed Claude Code binary on disk and restarts into the new version after the auto-updater replaces it. Background sessions are detached processes — they keep running through the restart.

If you set CLAUDE_CONFIG_DIR, the supervisor uses that instead of ~/.claude and runs as a separate instance.

PathContents
~/.claude/daemon.logSupervisor log
~/.claude/daemon/roster.jsonList of running sessions (used to reconnect after restart)
~/.claude/jobs/<id>/state.jsonPer-session state shown in agent view

To turn off background agents and agent view entirely:

~/.claude/settings.json
{ "disableAgentView": true }

Or set CLAUDE_CODE_DISABLE_AGENT_VIEW=1. Admins can enforce this through managed settings.

claude agents lists subagents instead of opening the view

Section titled “claude agents lists subagents instead of opening the view”

Earlier versions didn’t open agent view in every environment (Bedrock, Vertex AI, Foundry). Run claude update to install the latest version. If it still doesn’t open, check whether it’s been disabled by a setting or env var.

”Cannot open agents — N background task(s) running”

Section titled “”Cannot open agents — N background task(s) running””

The session you’re trying to background has in-flight work — a subagent, a workflow, or a background shell command. Run /tasks to see what’s running, then /bg to confirm abandoning them.

The dispatch input expects a task description, not a one-word query. Prompts under four characters are rejected with Too short.

Sessions show as failed after waking your machine

Section titled “Sessions show as failed after waking your machine”

Background sessions don’t survive sleep or shutdown. Attach, peek, or reply to any of them and they restart from where they left off — or claude respawn --all to restart them all at once.

Worktrees are removed when you delete the session that created them. If a session ended without cleaning up, list leftovers with git worktree list in the project directory and remove each with git worktree remove <path>.

  • Rate limits apply. Background sessions consume your subscription usage the same as interactive ones — ten parallel agents burn quota roughly ten times as fast.
  • Sessions are local. They run on your machine and stop if it sleeps or shuts down.
  • Worktrees die with the session. Merge or push changes before deleting a session that edited files.