no1msd/seance — reverse-engineered prompt

Reverse engineered prompt

GitHub

Build me a Linux desktop terminal app for managing multiple AI coding agents at once. I want it to feel like a terminal multiplexer, but instead of a fixed grid, panes should sit in a horizontal scrolling strip so I can move through long Claude Code, Codex, or similar sessions naturally.

It should automatically notice when an agent is running inside a pane and show its status in a sidebar, like working, idle, finished, or waiting for my approval. When an agent needs permission or finishes a task, send a desktop notification and mark it as unread until I check it.

Make it native, not Electron. Use GTK style integration, support Wayland and X11, and make the terminal rendering fast. Include workspaces, tabs inside columns, session restore after restart, a command palette, focus follows mouse, and no telemetry.

Also add a command line control tool so scripts and agents can create panes, switch workspaces, send input, read output, and return JSON. Keep setup simple with zero config.

Want more depth? Deep Reverse