v0.2.4

Added

• Another agent can drive the agents running in muxel — turn on "Allow outside
  tools to control muxel" and a program on the same computer — xAI's Grok Bot, a
  script, or another coding agent — can work your agents through the new muxel ctl
  command. It can list your projects, the pane layout and every agent with its
  status, read an agent's last prompt and reply and the question it is stuck on, send
  it a prompt, answer that question, press keys, and wait for its turn to end. Every
  command answers in JSON. It is off by default and only listens on this computer,
  behind a fresh token kept in a file only you can read. It won't type into an agent
  that is busy or waiting on a question, only picks an answer the question actually
  offers, and leaves shell panes alone unless you separately allow them, since typing
  into a shell is running commands.
• A Grok Bot page in Settings walks through the setup — four steps: turn
  outside control on, with a live status showing whether it is listening; allow Grok
  Bot to run commands on this computer; copy the skill that teaches it the commands
  and the rules; and Test, which runs the same muxel ctl command Grok Bot will
  and shows what muxel answered, or why it couldn't. One copied skill works on every
  computer you use: it says how to find muxel on each, and every answer names the
  computer that gave it. On Linux it works from the .deb, the .rpm and the AppImage
  alike.
• Agents shared between computers take turns — a remote project opened on two
  computers, or a local tmux project another computer has attached to, can be
  reached by more than one muxel. Before typing into such an agent, muxel checks
  whether another muxel is partway through a turn with it and refuses if so, naming
  the computer that has it, so two machines never type into the same agent at once.
• Pause, resume and start over reading aloud, pane by pane — every agent pane
  now has its own read-aloud controls in its header: the speaker starts reading,
  then becomes pause and resume, with start over and stop beside it. Each pane keeps
  its own place, so reading another pane pauses the one that was speaking, and coming
  back to it picks up at the start of the sentence it stopped in. A tab that is being
  read shows a badge that pauses or resumes it without switching to it.
  Ctrl+Shift+R reads, pauses and resumes the focused pane, Ctrl+Alt+R starts it
  over, and Ctrl+Alt+S stops whatever is speaking; all three are in the Ctrl+P
  palette.

Changed

• Projects on the same host share one connection from the start — opening
  several remote projects on one host together, as a workspace does at launch, used
  to open a separate SSH connection for each and announce "Connected to …" once per
  project. The first project now connects and the rest wait for it and share its
  connection, and the notice appears once per host.

Fixed

• Closing a remote pane on a Mac ends the agent running there — on a host whose
  login shell is zsh, which every Mac uses by default, the command muxel sends to end
  a closed pane's tmux session failed before it could run. The agent kept running on
  the host after you closed its pane or quit muxel. The session name is now quoted so
  zsh leaves it alone, and the session really ends.
• muxel no longer mistakes another app's AppImage for its own — on Linux, a
  muxel installed from the .deb or .rpm but started from inside another AppImage
  app, such as a shell in a terminal emulator shipped as an AppImage, picked up that
  app's AppImage details and believed it was that file. Updating muxel then
  overwrote the other app with muxel. muxel now treats itself as an AppImage only
  when it is actually running from one.

Full Changelog: v0.2.3...v0.2.4