[FEATURE] Agent teams should support visible execution in terminal panes with real-time status dashboard

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

When using CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS, spawned agents run as hidden background processes. There's no way to
watch what they're doing in real-time — you only see results after they finish (or fail silently).

Proposed Solution

What I'd love to see:

• Visible agent panes — An option like visible: true on the Agent tool that launches each agent in a visible tmux pane

(or split terminal). You'd see all agents coding side-by-side in real-time, not just waiting for messages to come back.

• Built-in status dashboard — A lightweight web UI (or terminal UI) showing agent roster, task progress (pending → in

progress → completed), and message feed. Right now there's no way to see which agent is doing what without reading inbox
JSON files manually.

• Terminal-aware — Should detect the user's current terminal (Warp, iTerm2, Terminal.app, etc.) and open agent panes

there, not hardcode a specific app.

Why this matters:

Multi-agent sessions are powerful but feel like a black box. When an agent fails or gets stuck, you don't know until it's
too late. Being able to watch agents work gives you confidence that things are on track, lets you catch issues early,
and honestly — it's just cool to watch 4 agents building features in parallel.

What I built as a workaround:

I ended up building two custom skills to fill this gap:

1. A /tmux-team skill that parses prompt files, creates tmux sessions with labeled panes per agent, handles dependency

ordering (backend finishes before frontend starts), and pipes prompts to claude -p in each pane

1. A surveillance dashboard (Node.js + SSE) that watches ~/.claude/teams/ and ~/.claude/tasks/ directories and shows a

live web dashboard with agent roster, message feed, and kanban task board. To be honest I copied this idea from a youtube video.

It works, but it bypasses the native agent system entirely. These capabilities feel like they belong in Claude Code
itself.

Alternative Solutions

_No response_

Priority

Medium - Would be very helpful

Feature Category

Developer tools/SDK

Use Case Example

_No response_

Additional Context

_No response_