22  Agent View in Claude Code

Originally published May 11, 2026.

22.1 One Place for All Your Agents

By mid-2026, Claude Code users were routinely running multiple agent sessions simultaneously. A developer might have a feature-building agent in one terminal, a code review agent in another, a test-writing agent in a third, and a documentation agent in a fourth. Managing these sessions — knowing which ones needed attention, which were blocked, which had completed — became a real challenge.

Agent View solves this. It provides one unified place to see and manage all your Claude Code sessions.

22.2 What Agent View Does

Agent View is a dashboard that shows every active and recently completed agent session, their status, and what they’re working on. It lets you:

  • See all sessions at a glance — what’s running, what’s waiting, what’s done
  • Kick off new agents without opening a new terminal
  • Send sessions to background so they continue running while you focus elsewhere
  • Jump into any session when it needs your input
  • Peek at session activity without leaving the dashboard
  • Reply inline to questions from agents

22.2.1 Opening Agent View

There are two ways to open Agent View:

# From the command line
claude agents

# From within a Claude Code session
# Press the left arrow key ← at the prompt

When you press the left arrow at an empty prompt, Claude Code transitions from the session view to Agent View. Press the right arrow (or Enter) on any session to dive back in.

22.2.2 The Agent View Layout

┌─────────────────────────────────────────────────────────────┐
│  AGENT VIEW                              [+ New Agent]      │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  ● auth-refactor               [Running]     12 min         │
│    "Refactoring the authentication module to use            │
│     the new token service..."                               │
│    Last activity: Running tests...                          │
│                                                             │
│  ● pr-review-1234              [Waiting]     3 min          │
│    "Reviewing PR #1234: Add webhook retry logic"            │
│    ⚠ Needs input: "Should I flag the missing error         │
│      handler as a blocker?"                                 │
│    [Reply inline...] ___________________________ [Send]     │
│                                                             │
│  ○ deploy-staging              [Completed]   8 min          │
│    "Deploy v2.3.1 to staging and run smoke tests"           │
│    ✓ Deployed successfully, all tests passed                │
│                                                             │
│  ○ db-migration                [Completed]   1 hr           │
│    "Migrate user settings to new schema"                    │
│    ✓ 14,239 rows migrated, verified                         │
│                                                             │
└─────────────────────────────────────────────────────────────┘

Each session card shows:

  • Status indicator: Running (●), waiting for input (●), completed (○), or failed (○)
  • Session name: Auto-generated from the task or user-defined
  • Duration: How long the session has been running
  • Description: What the agent is working on
  • Last activity: The most recent action or result
  • Alerts: Warnings or questions that need your attention

22.3 Sending Sessions to Background

One of the most powerful features of Agent View is the ability to send a session to background. This means the session continues running — executing commands, making tool calls, doing its work — but it’s no longer occupying your active terminal or editor view.

22.3.1 The /bg Command

From within any session, type /bg to send it to the background:

> /bg
Session "auth-refactor" sent to background.
It will continue running and notify you when it needs attention.

You can also background a session directly from Agent View by selecting it and pressing b.

Once backgrounded:

  • The session continues executing on Claude Code’s infrastructure
  • You receive a notification when the session needs input or completes
  • You can peek at its progress at any time from Agent View
  • You can jump back in with a single click or keystroke
Tip

Use /bg liberally. If an agent is running tests and you know it’ll take 10 minutes, background it and start a new task. The notification system ensures you won’t miss anything important. This is the key to running multiple agents productively.

22.3.2 Background Session Lifecycle

State What’s Happening Your Action
Running Agent is actively working Nothing needed — check back later
Waiting Agent has a question or needs approval Reply inline or jump in
Completed Agent finished the task successfully Review results
Failed Agent encountered an error it couldn’t resolve Jump in to diagnose
Idle Agent is paused, waiting for a trigger Resume when ready

22.4 Peek and Reply Inline

Peek lets you see what a session is doing without opening it fully. From Agent View, select any session and press p (or click the peek icon). A panel slides out showing the session’s recent activity — the last several messages, tool calls, and results.

If the session is waiting for input, you can reply inline directly from the peek panel:

┌─ Peek: pr-review-1234 ──────────────────────────────────┐
│                                                         │
│  Agent: I found a potential issue in the webhook        │
│  retry logic. The retry counter is never reset after    │
│  a successful delivery, which means after 3 retries     │
│  to any endpoint, all future deliveries will fail.      │
│                                                         │
│  Should I flag this as a blocking issue in the PR       │
│  review, or just leave a comment?                       │
│                                                         │
│  ┌─────────────────────────────────────────────────┐    │
│  │ Flag it as blocking. That's a real bug.         │    │
│  └─────────────────────────────────────────────────┘    │
│                                            [Send] [Open] │
└─────────────────────────────────────────────────────────┘

This is a huge productivity win. Instead of opening each session, scrolling to find the question, typing a response, and then navigating back, you handle the interaction in seconds without leaving Agent View.

22.5 Adding Existing Sessions with /bg

If you have a session running in a terminal that you want to manage from Agent View, you can add it:

# In your terminal session
> /bg

# The session is now visible in Agent View
# and continues running in the background

This works even for sessions that were started outside Agent View. Once backgrounded, they appear in the dashboard and can be managed like any other session.

22.6 Use Cases

22.6.1 Scaling Concurrent Sessions

The primary use case for Agent View is running many sessions simultaneously. A typical workflow for a senior engineer might look like:

Agent View:
  ● session-1: Building feature X (Running, 15 min)
  ● session-2: Writing tests for feature Y (Running, 8 min)
  ● session-3: Reviewing PR #1234 (Waiting for input)
  ○ session-4: Generating API docs (Completed)
  ● session-5: Monitoring deployment (Running, 3 min)

Without Agent View, managing five concurrent sessions in five terminals is chaos. With Agent View, you can see all five at a glance, handle the one that needs input, and let the rest continue.

22.6.2 Managing Long-Running Agents

Some tasks take a long time. A migration agent might run for hours. A deep research agent might take 30 minutes. A code review on a large PR might take 20 minutes.

Agent View lets you start these long-running tasks and forget about them until they need you. The notification system ensures you’re alerted when:

  • The agent has a question
  • The agent encounters an error
  • The task is complete
  • The session has been running longer than expected

22.6.4 Seeing What Shipped

Completed sessions in Agent View serve as a log of what your agents have accomplished. Each completed session shows:

  • What was requested
  • What the agent did
  • How long it took
  • The final result

This is valuable for:

  • Reviewing productivity at the end of the day
  • Auditing agent actions for security or compliance
  • Sharing results with team members
  • Learning from failures — understanding why an agent couldn’t complete a task

22.7 Notification System

Agent View includes a notification system that alerts you when sessions need attention. Notifications appear as:

  • Desktop notifications (when the app is in the background)
  • In-app banners (when you’re using Claude Code)
  • Badge counts on the Agent View icon

Notification types:

Type Trigger Priority
Needs input Agent is waiting for your response High
Error Agent encountered an unrecoverable error High
Completed Session finished successfully Medium
Long-running Session exceeded expected duration Low
Idle Session has been waiting for a long time Low
Note

You can configure notification preferences per session type. For example, you might want immediate notifications for deployment sessions but only summary notifications for documentation sessions. Configure these in Settings → Notifications.

22.8 Keyboard Shortcuts

Agent View is designed to be fully navigable by keyboard:

Shortcut Action
← Open Agent View (from session prompt)
→ or Enter Jump into selected session
↑ / ↓ Navigate between sessions
b Send selected session to background
p Peek at selected session
r Reply inline to selected session
n Start new agent
Esc Close peek panel / return to view
/ Focus search/filter
? Show all shortcuts

22.9 Best Practices

22.9.1 Name Your Sessions

Agent View auto-generates session names, but custom names are more useful:

Bad:  "Session 3"
OK:   "Refactoring auth"
Good: "auth-refactor: token service migration"

22.9.2 Background Aggressively

If you’re not actively interacting with a session, background it. The cost is zero — the session continues running — and the benefit is a cleaner workspace.

22.9.3 Handle Waiting Sessions First

When you open Agent View, prioritize sessions with “Waiting” status. These are blocked on your input and can’t make progress until you respond. Clearing them first keeps all your agents moving.

22.9.4 Review Completed Sessions

Don’t just dismiss completed sessions. Take a moment to review what the agent did, verify the results, and learn from any issues. This builds your understanding of what agents do well and where they struggle.

22.9.5 Integration with Routines and Scheduled Tasks

Agent View integrates seamlessly with Routines (Chapter 21) and scheduled tasks. Scheduled routines appear in Agent View alongside interactive sessions, giving you a unified view of all agent activity:

Agent View:
  ● Interactive: Building feature X (Running)
  ● Interactive: Reviewing PR #1234 (Waiting)
  ○ Routine: nightly-tests (Completed, 6am)
  ○ Routine: dep-update (Completed, 9am)
  ● Triggered: incident-response (Running, triggered by PagerDuty)

This means you can monitor both your on-demand work and your automated workflows in one place. If a routine encounters an issue that needs human judgment, it appears in Agent View with a “Waiting” status, just like an interactive session.

Tip

Use Agent View as your morning dashboard. Before starting your day’s work, open Agent View to see what your overnight routines accomplished, what needs attention, and what’s still running. This gives you a complete picture of your agent ecosystem in seconds.

22.10 Key Takeaways

  • Agent View provides one unified place to manage all Claude Code sessions — active, waiting, and completed.
  • Open it with the left arrow key at any prompt, or run claude agents from the command line.
  • Send sessions to background with /bg — they continue running and notify you when they need attention.
  • Peek and reply inline lets you handle agent questions without fully opening each session.
  • The notification system alerts you to sessions that need input, encountered errors, or completed.
  • Agent View is essential for scaling concurrent sessions, managing long-running agents, and navigating between tasks.
  • Completed sessions serve as a log of what your agents accomplished, useful for review, auditing, and learning.
  • Full keyboard navigation makes it fast to move between sessions without reaching for the mouse.