2026. 05. 13. / TECH · 5 min read
A Control Tower in the Terminal — Complete Guide to Claude Code Agent View
Dispatching and managing multiple Claude Code sessions from one screen — from installation to keyboard shortcuts
claude agents.
One command opens the agent view — a single screen showing all background Claude Code sessions.
What's running, what's waiting for input, what's finished — no more jumping between multiple terminals.
What Is the Agent View?
In one line: a CLI control tower for dispatching and managing multiple Claude Code agents simultaneously.
It's a research preview feature that works on Claude Code v2.1.139 and above.
# Check version
claude --version
# Open agent view
claude agents
Use it when you have multiple independent tasks — bug fixes, PR reviews, log analysis. If it's something you need to work through together, just connect to a session and work conversationally.
What You See on Screen
Opening the agent view shows a session list grouped by status.
Needs Input
✻ power-up design needs input: double jump or wall climb? 1m
Working
✽ collision detection Edit src/physics/CollisionSystem.ts 2m
Completed
∙ sound effects result: 14 SFX exported to assets/audio 4h
The icon on each row tells you two things.
Indicator color = session status:
| Indicator | Status | Meaning |
|---|---|---|
| Animated | Working | Claude is running tools or generating a response |
| Yellow | Needs input | Waiting on a permission decision or answer |
| Dim | Idle | Waiting for input without being blocked |
| Green | Completed | Task successful |
| Red | Failed | Exited with an error |
| Gray | Stopped | Stopped via Ctrl+X or claude stop |
Icon shape = process status:
✻/ animated✽→ Session active, can answer immediately∙→ Process exited, but connecting resumes from where it left off✢→ A/loopsession sleeping between iterations
Peeking and Quick Replies
Press Space to open the peek panel.
You can see what the session is doing and what it needs — right there. Type a reply and press Enter to send it to the session without entering the session directly.
It feels like replying to a Slack message to manage your agents.
For multiple-choice questions, press number keys to select; Tab fills in the suggested answer. Prefix your reply with ! to send it as a Bash command.
Connecting and Detaching
Press Enter or → on a row to connect to that session. The agent view disappears and the terminal is fully taken over by a normal claude session.
Press ← from an empty prompt to return to the agent view.
Detaching keeps the session running. ←, Ctrl+C, Ctrl+D, and /exit all keep the session alive. To end a session, use /stop.
Dispatching New Agents
Type a prompt in the input at the bottom of the agent view and press Enter.
# Start a background session directly from the shell
claude --bg "investigate the flaky SettingsChangeDetector test"
# Run as a specific subagent
claude --agent code-reviewer --bg "address review comments on PR 1234"
To send an existing interactive session to the background, run /bg inside the session.
Prefixes you can use in the dispatch input:
| Input | Effect |
|---|---|
<agent-name> <prompt> | Run with a custom subagent |
@<repo> | Run a session in a specific repository |
/<skill> | Dispatch a skill as the prompt |
#<number> or PR URL | Select a session working on that PR |
Shift+Enter | Dispatch and immediately connect |
File Edit Isolation
When a background session edits files, Claude automatically moves to an isolated git worktree under .claude/worktrees/.
Parallel sessions can read the same checkout but each writes to its own worktree.
Deleting a session also removes its worktree. Merge or push your changes before deleting.
Full Keyboard Shortcut Reference
| Shortcut | Action |
|---|---|
↑ / ↓ | Navigate between rows |
Enter | Connect to session or dispatch |
Space | Open/close peek panel |
Shift+Enter | Dispatch and immediately connect |
→ | Connect to session |
← | Detach from session (from empty prompt) |
Alt+1..Alt+9 | Connect directly to the Nth session |
Tab | Search subagents or apply suggestion |
Ctrl+S | Toggle grouping by status/directory |
Ctrl+T | Pin/unpin session |
Ctrl+R | Rename session |
Ctrl+G | Edit prompt in $EDITOR |
Ctrl+X | Stop session (re-press within 2s to delete) |
Esc | Close peek, clear input, exit |
? | Full shortcut list |
Managing Sessions from the Shell
claude attach <id> # Connect to session
claude logs <id> # View recent output
claude stop <id> # Stop session
claude respawn <id> # Restart a stopped session
claude respawn --all # Restart all stopped sessions
claude rm <id> # Remove session from list
When sessions stall after sleep or shutdown, claude respawn --all brings them all back at once.
Good to Know
Token usage scales with session count. Background sessions are each independent — running 10 in parallel burns through your quota 10x faster.
Sessions are local. If your machine sleeps, running sessions stop. Wake it and run claude respawn --all.
Transcripts persist after process exit. About an hour after completion, the supervisor process stops the process to free resources, but connecting again resumes from where you left off.
The workflow of jumping between multiple terminals changes.
Agent view is still a research preview so the interface may change, but the direction is clear — AI doesn't work one task at a time anymore. It handles multiple tasks simultaneously.
