tmuxctl 0.2.0

Small tmux session controller with recurring sends

tmuxctl

tmuxctl is a small tmux workflow helper for three things:

• finding the session you want
• jumping to it quickly
• sending recurring follow-ups to long-running agent or worker sessions

It installs two executables:

• tmuxctl
• t

t is just the shorter alias for the same CLI.

Install

Primary install:

uv tool install tmuxctl

Then use either:

tmuxctl --help
t --help

Run without arguments to see the 10 most recent sessions plus shortcut hints:

tmuxctl
t

Core Workflow

1. Find the session you want

Show all sessions, sorted by recency, with numeric IDs:

t list

Short form:

t l
tl

Typical output:

IDX  SESSION               CREATED
1    codex                 2026-04-03 15:56:59
2    backend-worker        2026-04-03 15:22:10
3    docs                  2026-04-03 14:10:31

If you just want the recent view:

t
t r
t recent --limit 10

2. Jump into a session

Attach by name:

t codex

That is equivalent to:

t attach codex

Ask tmux to resize the window after attaching:

t git-llm-zoomcamp --resize-window
t 1 -r

Attach by recency index:

t 1
t 2
t 10

Those resolve to attach-recent N.

Attach to the newest session directly:

t attach-last

3. Create a session if it does not exist

Use a leading colon when you want create-or-attach behavior:

t :codex

That resolves to:

t create-or-attach codex

Rule of thumb:

• t codex means attach only
• t :codex means create or attach

Use t - to derive the session name from the current directory and create-or-attach it:

cd ~/git/workshops
t -

That resolves to:

t create-or-attach git-workshops

Pass a command after t - to run it only when a new session is created:

t - cy

If you want another session for the same folder, add any suffix:

cd ~/git/workshops
t -asd

That resolves to:

t create-or-attach git-workshops-asd

4. Send a one-off message

Send text directly:

t send codex --message "check status and continue"

Or send from a file:

t send rk-codex --message-file prompts/rk-codex-progress.txt

By default, send waits 200ms before pressing Enter. You can change that:

t send codex --message "status?" --enter-delay-ms 500
t send codex --message "status?" --no-enter

Automation Workflow

1. Add a recurring job

Inline message:

t jobs add codex --every 15m --message "check status and continue"

If you are already inside tmux, use :current to target the active session without typing its name:

t jobs add :current --every 15m --message \
  "Check project status and continue. Help any blocked agents, review CI, and \
  keep the pipeline moving. If nothing in the current batch needs attention, \
  pick the next two ready issues per _docs/PROCESS.md and run the full workflow."

Shared prompt file:

t jobs add rk-codex --every 30m --message-file prompts/rk-codex-progress.txt

When a job uses --message-file, tmuxctl stores the file path and reads the file at send time. Updating the file updates future scheduled runs.

2. Run the scheduler

t jobs daemon

Recurring jobs only run while the daemon is running.

3. Inspect and edit jobs

t jobs
t jobs list
t jobs show 2
t jobs logs --limit 20
t jobs edit 2 --every 45m
t jobs edit 2 --message "check status and continue"
t jobs edit 2 --session :current
t jobs edit 3 --message-file prompts/rk-codex-progress.txt

Useful job controls:

t jobs pause 3
t jobs pause-current
t jobs resume 3
t jobs resume-current
t jobs remove 3

If a scheduled job fails 3 runs in a row, tmuxctl jobs daemon removes it automatically.

Session Cleanup

Kill a session by name:

t kill codex

Kill a session by the numeric ID shown in t list:

t kill 2

Skip confirmation:

t k 2 --yes

Rename a session and retarget any scheduled jobs bound to it:

t rename codex codex-main
t rename 2 archived-worker

Inspect a Session

describe shows what is actually running inside a session — the process in each pane, its working directory, the cgroup the session lives in, and (for sessions started by t with a memory cap) live RAM and CPU usage read straight from that cgroup. Target it by name, by the numeric ID from t list, or :current:

t describe codex            # by name
t describe 2                # by the numeric ID from `t list`
t describe :current         # the session you are in

For a capped session it reads memory and CPU from the session's tmuxctl-<name>.scope cgroup, so the numbers cover the whole process tree, not just the shell:

Session:  git-myproj  (1 window(s), 1 pane(s))

WIN.PANE  PID      COMMAND          DIRECTORY
0.0*      3564734  claude           /home/you/git/myproj

Scope:    tmuxctl-git-myproj.scope  (active)
Cgroup:   /user.slice/.../robust.slice/tmuxctl-git-myproj.scope
Memory:   3.4G / 12.0G  (peak 5.2G, swap 0B)
CPU time: 7m25s
Tasks:    222

A session you did not start through t has no memory cap. describe says so and prints the real cgroup it found (e.g. a plain session-NN.scope), so you can tell at a glance which sessions are protected and which can still take the whole tmux server down under memory pressure:

Scope:    none — session is uncapped (not started by tmuxctl)
Cgroup:   /user.slice/.../session-7.scope
          No per-session RAM/CPU cap; the box-wide OOM-killer can
          take the whole tmux server. Start a capped one with:
          t :git-myproj --mem 24G

Shell Setup

Bash completion

Install completion:

t --install-completion

Preview the script:

t --show-completion bash

Completion works for:

• commands
• plain session names
• :session shortcuts

Local checkout helper

If you are working from this repository and want its virtualenv binaries on your PATH, run:

./install.sh

That appends this repo's .venv/bin and alias tl='t l' to ~/.bashrc, skipping any line that is already present.

How Scheduling Works

Recurring jobs are stored in:

~/.config/tmuxctl/tmuxctl.db

The scheduler is database-driven:

• jobs add creates jobs
• jobs edit, jobs pause, jobs resume, and jobs remove modify jobs
• jobs daemon polls for due jobs and runs them

If you want recurring jobs to survive logout or reboot, keep t jobs daemon running with something like:

• systemd --user
• launchd
• cron @reboot

Running as a systemd user service (Linux)

Create ~/.config/systemd/user/tmuxctl.service:

[Unit]
Description=tmuxctl scheduler daemon
After=default.target

[Service]
Type=simple
ExecStart=%h/.local/bin/tmuxctl jobs daemon
Restart=on-failure
RestartSec=5

[Install]
WantedBy=default.target

Adjust ExecStart to wherever tmuxctl is installed (for a local editable checkout, point at .venv/bin/tmuxctl). Then enable and start it:

systemctl --user daemon-reload
systemctl --user enable --now tmuxctl.service
systemctl --user status tmuxctl.service

To keep the daemon running after you log out, enable lingering for your user (needs sudo, one-time):

sudo loginctl enable-linger "$USER"

Logs are available via journalctl --user -u tmuxctl -f.

Known Problems

An orphaned scope can block recreating a session of the same name

Normally t kill tears a session's memory-capped scope down (systemctl --user stop tmuxctl-<name>.scope), so the unit name is free for next time. Two things have to go wrong together to defeat that:

1. the tmux server dies uncleanly (a crash or machine-wide OOM), so the normal kill path — and its scope teardown — never runs, and
2. a disowned background process (e.g. an Xvfb, a dev server, anything nohup/&-launched) is still running inside that session's scope.

The dead session's shell is gone, but the stray process keeps the tmuxctl-<name>.scope cgroup alive. The next time you try to create a session with the same derived name (e.g. t - from the same folder), tmuxctl asks systemd-run for that unit name and it fails with "Unit tmuxctl-.scope was already loaded". The new tmux pane's command dies on launch, and instead of a clear error you usually see terminal escape codes (a Device-Attributes reply such as ^[[?61;...c) leak onto your prompt as the aborted tmux client exits.

Diagnose — look for a scope whose tmux session no longer exists:

systemctl --user list-units --type=scope --all 'tmuxctl-*'
systemctl --user status tmuxctl-<name>.scope   # shows the stray process holding it open

Fix — stop the orphan scope (this also kills the stray process inside it), then create the session again:

systemctl --user stop tmuxctl-<name>.scope

This is rare (it needs an unclean crash and a backgrounded process), so it is documented rather than worked around. A future version may auto-recover by clearing a stale scope whose session is gone before creating a new one.

Alternatives

Install with pip:

pip install tmuxctl

Install directly from GitHub:

uv tool install git+https://github.com/alexeygrigorev/tmuxctl.git

Install from a local checkout in editable mode:

git clone https://github.com/alexeygrigorev/tmuxctl.git
cd tmuxctl
uv tool install -e .

If you use the local checkout install, also run:

./install.sh

Reinstall the local checkout after updates:

uv tool install -e . --force

For development:

uv sync --dev
uv run pytest
uv build