Managing All Claude Code AI Sessions with the New Agent View

Problem addressed

Running multiple parallel Claude Code tasks often requires opening many terminal tabs, remembering which pane runs which command, handling permission prompts, and preserving context when a terminal is closed. Agent View consolidates all background sessions into a single fullscreen UI, eliminating these issues.

Getting started

Prerequisites

Claude Code version ≥ v2.1.139

Supported plans: Pro, Max, Team, Enterprise, API

Check version with claude --version and update with claude update if needed

Open Agent View

claude agents

The terminal switches to a fullscreen view with an input box at the bottom and a session list at the top. Press Esc to exit; background sessions keep running.

Dispatch a task

Type a command in the input box and press Enter:

Refactor the src/utils directory, extracting duplicate code into a shared function

The new session appears instantly and starts processing. Additional tasks can be queued, for example:

Review code changes in PR #42

Add unit tests for the user module

Each task runs independently.

Peek – quick replies

Select a session with ↑ / ↓ and press Space to open the Peek preview panel. The panel shows what the session is doing and allows answering prompts (e.g., “Choose option A or B?”) without leaving Agent View.

Attach – full interaction

Press Enter or → to attach to a session, opening a regular claude session identical to the standalone CLI. To return, press ← in an empty input box; the session detaches but continues running.

Backgrounding tasks

From an existing session: /bg # or /bg run tests then auto‑fix failures Directly from the terminal: claude --bg "investigate flaky test cause" With a specific agent:

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

After launching, Claude prints the session ID and management commands, e.g.:

backgrounded · 7c5dcf5d   claude agents   list sessions   claude attach 7c5dcf5d   open in this terminal   claude logs 7c5dcf5d   show recent output   claude stop 7c5dcf5d   stop this session

Status icons

✽ (animated) – Working: Claude is executing a tool or generating a reply.

✻ (yellow) – Needs Input: Waiting for user action (permission, answer).

✻ (dim) – Idle: Waiting for input but no specific question.

✻ (green) – Completed: Task finished.

✻ (red) – Failed: Task errored.

∙ (gray) – Stopped: Process exited but can be resumed.

Session persistence

Closing the terminal does not terminate sessions; a supervisor process keeps them alive. Re‑run claude agents to view them, and after machine sleep use claude respawn --all to restore all sessions.

Shortcut reference

Space

– Toggle Peek preview panel Enter – Dispatch typed task or enter selected session ← (empty input) – Detach from current session back to the list Shift+Enter – Dispatch and immediately enter the new session Alt+1 … Alt+9 – Jump directly to the N‑th session Ctrl+S – Toggle grouping (by status or by directory) Ctrl+T – Pin / unpin a session Ctrl+R – Rename a session Ctrl+X – Stop a session (press Delete again within 2 s to confirm) ? – Show all shortcuts

File‑edit isolation with git worktree

When a background session needs to modify files, Claude automatically creates a git worktree under .claude/worktrees/. Each session gets its own working copy, preventing conflicts between parallel edits. Merge or push changes before deleting a session.

Managing sessions from the terminal

claude agents                # list all background sessions
claude attach <id>          # enter a specific session
claude logs <id>            # view recent output
claude stop <id>            # stop a session
claude respawn --all        # restore all sessions after a reboot

Practical usage patterns

Batch dispatch + consolidated PR review: launch multiple independent tasks, each with its own sub‑agent, then review all PRs together.

Long‑running agents: continuous PR monitoring, dashboard updates, etc., with a countdown shown in the list.

Temporary context switch: detach with ←, ask a quick question, then re‑attach with →.

Rapid outcome scan: status icons and Peek titles let you instantly identify sessions that have already produced PRs.

Considerations

Each background session consumes its own token quota; running many simultaneously can exhaust tokens quickly.

Session summaries are generated by the Haiku model (~15 s refresh) and add a small token cost.

The feature is currently in Research Preview; shortcuts and UI may change.

Administrators can disable Agent View with the disableAgentView setting.