Claude Code Agent View - Run multiple Claude Code sessions from one screen
by•
Agent View lets developers dispatch, monitor, and manage multiple Claude Code sessions from a single CLI screen without switching terminals.
For engineers running parallel coding tasks in Claude Code.
Replies
The bottleneck in running multiple Claude Code sessions was never the agents, it was the overhead of watching them.
What it is: Agent View is a new view in Claude Code that consolidates all background sessions into a single terminal screen, showing what each session is doing and surfacing the ones that need your input.
Before this, running parallel Claude Code sessions meant managing several terminal tabs or a tmux grid, with no unified view of which sessions were blocked, which were still working, and which had finished. You had to attach to each one individually to find out. Agent View replaces that with a state-grouped list: working, needs input, completed, failed. Each row has the session name, a Haiku-class generated summary, and the time since last activity.
What makes it different: Background sessions run independently of any terminal. A supervisor process keeps them alive whether you have agent view open or not, and they persist across terminal closes and auto-updates. Sessions that open pull requests show the PR link and CI status directly in the row. Most of the time, you never need to attach to the full transcript. The peek panel lets you read output and reply inline.
Key features:
State-grouped session table with sessions needing input pinned to the top
Peek panel for reading output and replying without attaching
Shell commands: claude --bg, claude attach, claude logs, claude stop, claude respawn --all
/bg inside an interactive session backgrounds it mid-conversation
Per-session worktree isolation to prevent file conflicts across parallel sessions
Filter by state (s:blocked), agent name (a:<name>), or PR
Research preview; requires Claude Code v2.1.139 or later
Benefits:
Collapses multi-terminal overhead into one screen
Sessions stay alive independently of terminal state
PR and CI visibility means review work is collectible at the row level
Inline reply handles most unblock actions without entering a full session
Who it's for: Developers and engineering teams using Claude Code to run several independent tasks in parallel, such as bug fixes, PR reviews, and test investigations, who need a unified view rather than separate terminals per task.
My read is that agent view moves Claude Code from a tool you interact with to a system you supervise. That distinction matters for how teams will scale agentic workflows going forward.