Agent View — manage multiple Claude sessions on one screen

Dispatch a bug fix, a PR review, and a test investigation at the same time — step in only when needed

⏱ Est. ~6 min

01 · Read

Before, when you opened Claude Code, you could only talk to one session at a time. Want to do a bunch of things at once? You had to open a new terminal, a new session, keep track of which is where, and tab back and forth to see who needs you.

Agent View turns this into one screen. You open it by typing claude agents, then dispatch a bunch of sessions to the background at once: fix this bug, review that PR, look into a flaky test. Each session is an independent Claude Code conversation, runs in the background, and leaves results when it's done. You keep working in another window and only come back when you need to see who needs your decision and who's already finished.

💡 Picture thisLike a Kanban board in an office — each card is a task with its own progress and its own owner. You don't stare at every card; you only walk over when something lands in the "needs a decision" column.

Key points

claude agents opens Agent View
Each session runs independently in the background, no terminal attached
Sessions are grouped by status: needs input, running, completed
If you close Agent View, sessions keep running; they're still there when you open it again

02 · Read

Agent View opens as a table, one row per session, showing its name, what it's working on, and how long ago it last updated.

Status is shown by icon color:

- Animated icon = working - Yellow = needs your input (a permission decision or a question) - Green = done - Red = failed - Gray = stopped

The main controls: arrow keys select a row, Space opens a preview panel to see recent output, Enter or → attaches into the full conversation, ← on an empty input returns to Agent View, Ctrl+X twice deletes a session, Esc exits.

Background sessions are hosted by a supervisor process separate from your terminal. Closing the terminal or letting your computer sleep is fine. But shutting down the whole machine stops all sessions — when you wake up, attaching will auto-restart them.

Key points

Icon colors tell you which session needs you
Space previews recent output without attaching to the full conversation
Enter attaches in, ← exits back to the table
Sessions are detached from the terminal — closing the terminal is fine, only shutdown stops them

03 · Code example

Dispatch a background session from the shell without opening Agent View. Good for scripts or quick dispatches:

shell — dispatch a task and fetch the result

# 派一個任務到背景
claude --bg "查 SettingsChangeDetector 那個不穩定測試"
# → backgrounded · 7c5dcf5d · session-name

# 之後想看做到哪
claude logs 7c5dcf5d

# 想介入就附加進去
claude attach 7c5dcf5d

# 要停掉就 stop
claude stop 7c5dcf5d

# 列出所有背景 session
claude agents

--bg is a shortcut that sends the session straight to the background. It returns a short ID (like 7c5dcf5d) you'll use later to interact with it. claude agents opens Agent View to see all sessions at a glance.

04 · Code example

A real-world scenario: three things Claude can run on its own. You dispatch in 30 seconds and move on:

Dispatch three things at once

claude --bg --name fix-login \
  "修 #1234 登入按鈕在 Safari 沒反應的 bug"

claude --bg --name review-pr \
  "審 PR #1240，重點看 auth middleware 改動"

claude --bg --name flaky-test \
  "查 e2e/checkout.spec.ts 為什麼隨機失敗"

# 開 Agent View 看誰先完成
claude agents

The three sessions run in parallel, each editing files in its own git worktree (under .claude/worktrees/), so they don't interfere with each other. --name lets you recognize each one later at a glance — leave it out and Claude names it for you. It takes 30 seconds to dispatch, and then you can go grab coffee, run a meeting, or write other code. When you come back, open Agent View and see which is yellow (needs you) and which is green (done, ready for PR review).

05 · Read

Agent View vs subagents vs worktrees — all three let you "do many things at once," but they're used differently:- Subagent: You're still in one session, but Claude dispatches helpers to do subtasks (like code review), and they report back. You watch the main conversation. - Worktree: Multiple working directories on the same repo. You manually switch sessions. You manage multiple terminals. - Agent View: Dispatch multiple independent sessions to the background, each a full conversation with its own results. You see them all on one screen.

Short version: small task tied to your main flow → subagent; big task that takes ten minutes and is independent → Agent View.

Key points

Subagent: main session dispatches helpers for subtasks, results report back
Worktree: multiple working directories, sessions still switched manually
Agent View: multiple independent sessions running in the background, one overview screen
Independent big tasks (bug fixes, PR reviews) are easiest with Agent View

06 · Quiz

You open Agent View, dispatch three sessions, and press Esc to return to the shell. Ten minutes later you open claude agents again. What do you see?

The sessions all stopped because you closed Agent View
The sessions are still running and making progress — a supervisor process manages them in the background
The sessions are paused and need a resume to continue
The sessions restarted from the beginning

07 · Match

Match each scenario with the best tool.

(This section is interactive — please enable JavaScript.)

Other lessons in this chapter

⚠ The full interactive experience needs JavaScript. Please enable it and reload.

※ This is an independent Traditional Chinese teaching project — not an official Anthropic product. Claude™ is a trademark of Anthropic, PBC.