Install agent hooks
Let your coding agent report its own state to Muxie for accurate working / blocked / done tracking.
Hooks let a coding agent report its state changes directly to Muxie. This is far more accurate than detecting state by reading the screen. See Agent awareness for how Muxie uses this information.
Install hooks with the CLI
Use the muxie CLI to install hooks:
muxie hooks setup claude— install Claude Code’s lifecycle hooks.muxie hooks setup <agent>— install another agent’s hooks.muxie hooks setup all— install hooks for all recognized agents.
Recognized agents are: claude, codex, aider, gemini, cursor-agent, opencode, amp, and droid. See Supported agents.
How hooks work
Hooks translate the agent’s lifecycle events into Muxie state: working, idle, needsInput, and done. They are keyed to the pane through the MUXIE_SESSION_ID environment variable that Muxie sets in each pane, so each agent’s state lands on the correct session.
Note: Because hooks rely on
MUXIE_SESSION_ID, run the agent in the same pane Muxie opened for it, rather than detaching it into an unrelated shell.
Improve cwd, git, and prompt reporting
Shell integration adds working-directory, git, and prompt information. Enable it by adding the appropriate line to your shell startup file:
eval "$(muxie shell-integration bash)"Use zsh in place of bash for the Z shell.
Heads up: Shell integration is POSIX-only. On shells that don’t support it, the command simply does nothing and the rest of Muxie keeps working.
For the full set of CLI commands, see the CLI reference.