[FEATURE] Single-agent, multi-terminal sessions (daemon + client workflows)

Summary

Claude Code today supports multi-pane terminals via Agent Teams (teammateMode: "tmux" / iTerm2 split panes), but Agent Teams is designed for multiple agents — one teammate per pane. This request is for the complementary case: a single Claude agent that owns multiple terminals so it can run a persistent daemon in one and drive it from another, without spawning sub-agents or losing live output.

User context

I recently switched to the Claude Desktop app after running Claude Code inside the VS Code integrated terminal. One of the things I immediately missed is the ability to have multiple terminals open in the same working context — something VS Code supports by default and which made daemon + client workflows effortless. Reproducing that inside Claude Desktop would remove one of the biggest friction points of the migration.

Motivating workflows

• Daemon + client. Run a dev server, API, database, or WebSocket daemon in one terminal and hit it with curl, a test client, or a REPL in another. Today the options are: background the server and lose live logs, or step outside Claude Code entirely.
• Log tailing alongside active work. tail -f, docker logs -f, or kubectl logs -f in one pane while running commands in another.
• Watchers and builders. npm run dev, vite, webpack --watch, cargo watch, or pytest --watch running continuously while one-off commands execute elsewhere.
• Multi-service local stacks. Frontend + backend + worker, each needing its own terminal for live output.
• Debugger + target process. gdb / lldb / language-specific debuggers in one terminal, program under test in another.
• Git ops during long-running tasks. git status / diffs / commits while a build or test suite runs.

How this differs from Agent Teams

| Agent Teams | This request |
|---|---|
| One agent per pane (teammates) | One agent, many panes |
| Panes coordinate as collaborators | Panes are tools the agent uses |
| Optimized for parallel agent work | Optimized for daemon/client-style workflows within a single task |

Agent Teams can't currently be repurposed for this because each pane runs its own Claude instance with its own context.

Proposed behavior

• Allow a single session to open multiple named terminals (tabs, panes, or a terminal list).
• Claude addresses terminals by name/ID: send input to a specific terminal, read output from any of them.
• Long-running processes (servers, watchers) persist in their own terminal while Claude continues working in others.
• Ideally reuses the existing pane-backend abstraction (tmux, iTerm2, and the in-flight zellij / WezTerm / Windows Terminal backends) that Agent Teams already uses.
• Optionally: Claude can watch output from a background terminal and react when something meaningful happens (test fails, server logs an error).

Why it matters

Most non-trivial development is not a single linear command stream — it is a server plus clients, a build plus tests, a watcher plus edits. A single terminal forces Claude to stop and restart processes, or hand control back to the user mid-task. Adding single-agent multi-terminal support — on top of the pane infrastructure that already exists — would unlock a large class of real-world workflows without the overhead of spawning multiple agents.

Related

• teammateMode: "tmux" / Agent Teams (existing, complementary feature)
• Active pane-backend work: zellij, WezTerm, Windows Terminal issues