Originally published at devtoolpicks.com
Anthropic just shipped Claude Code v2.1.139. Two features in one release: agent view, which gives you a single dashboard for all your Claude Code sessions, and the /goal command, which lets Claude work autonomously across turns until a task is done.
How to update
Before anything else:
claude update
claude --version # confirm you're on v2.1.139 or later
Both features are live immediately after updating. No additional configuration needed.
What is Claude Code agent view?
Agent view is a session roster. One screen. Every Claude Code session you have running, blocked, or finished, in a single list.
Without it, running multiple Claude Code sessions means multiple terminal tabs, multiple tmux panes, or just losing track of what each agent is doing. Thariq from the Claude Code team at Anthropic described it as "like tmux built for CC", and that is the right mental model. You open one screen, see everything, and navigate without losing context.
How to open it
claude agents
That's it. From inside any existing session, press the left arrow to jump to agent view. To return to a session, press Enter or the right arrow.
What you see
Each row in the table shows:
- Session name or the task it's working on
- Current state: working, waiting on you, done, failed, or stopped
- Last activity timestamp
Three keyboard shortcuts cover most of what you need:
| Key | Action |
|---|---|
| Spacebar | Peek at the latest turn without attaching |
| Enter / → | Attach to the full session |
| Esc | Exit agent view (sessions keep running) |
Starting and managing sessions
From agent view, type a prompt and press Enter to start a new background session. From inside an existing session, run /bg to send it to the background. From the shell directly:
claude --bg "write tests for the auth module"
The session starts immediately in the background and appears in the agent view roster.
The rate limit caveat: read this first
This is the most important thing to understand before you start spawning parallel agents.
Background sessions draw down your subscription quota the same as interactive sessions. If you open agent view and run 5 sessions in parallel, you are using your Claude Pro or Max quota 5 times faster than normal. Ten sessions in parallel means 10x quota consumption.
For an indie hacker on Claude Pro ($20/month) or Max ($100/month), this limits how aggressively you can parallelize. Test with 2-3 concurrent sessions first and see how your quota holds up before scaling further.
The second limitation: sessions are local. They run on your machine. If your laptop goes to sleep, all background sessions stop. For tasks that need to run overnight, run Claude Code on a Hetzner or Railway VPS over SSH. The session stays alive regardless of your local machine state.
The /goal command
The second headline feature in v2.1.139 is the /goal command. Set a completion condition and Claude keeps working autonomously across turns until that condition is met or it gets stuck.
/goal deploy the staging environment with passing tests
A live overlay panel shows elapsed time, turn count, and current token usage while Claude works. It runs in interactive mode, with the -p flag, and in Remote Control.
This is separate from agent view. You can use /goal in a single session without opening agent view at all. The combination is where it gets interesting: start a session with /goal, send it to the background with /bg, then monitor progress from agent view while you work on something else.
Who benefits most
Multiple feature branches at once. Run a separate Claude Code session for each branch, monitor from agent view, peek when one needs input, attach when it's ready for review.
Parallelising independent tasks. Writing tests for one module while Claude refactors another, running a code review session while a new feature session is building. Tasks that don't block each other are the right fit.
Long-running autonomous work with /goal. Tasks that take 20-30 turns to complete without human input are now viable. Set the goal, send to background, check back when done.
Where it does not help yet: Tasks that depend on each other. Agent view is a session manager, not an orchestrator. Claude Code does not yet automatically coordinate multiple sessions. You do that manually. For automated multi-agent coordination, that's in a separate product path (Claude Managed Agents, currently in beta).
What this release means for Claude Code's direction
Boris Cherny from Anthropic called this "a shift to multi-agent workflows." That framing is right. Claude Code started as a single-session terminal tool. Agent view is the first step toward using it as a genuine agent operations layer where you run and monitor multiple parallel workstreams from one place.
The research preview label means the interface will change. Keyboard shortcuts, session display, and background session behaviour are all subject to revision based on feedback. Use it now, but expect the details to evolve over the coming weeks.
For context on how agent view fits into the broader Claude Code setup, the CLAUDE.md guide covers the project configuration that applies across all sessions. For the comparison between Claude Code and Codex agents, the Codex /goal vs Claude Code comparison is now worth revisiting. Both tools ship /goal, but the implementation differs.
Top comments (0)