[idea] repeatable keybindings (tmux bind -r): press prefix once, then tap to step through workspaces/agents

[peinan]

idea / problem

I move between workspaces and agents constantly, keyboard-only.

There's already a workspace switcher (prefix+w), but it's a separate mode I have to open each time — great for jumping to a specific or distant workspace, but heavier than I want for quickly hopping to the next/previous one.

For that I'd rather use the direct stepper actions, which already exist and I can bind, e.g.:

[keys]
next_workspace     = "prefix+shift+j"
previous_workspace = "prefix+shift+k"

But each step is one-shot — prefix mode exits the moment the action fires. So moving across 4 workspaces means prefix shift+k, prefix shift+k, prefix shift+k, prefix shift+k, re-pressing the prefix every single time. With many workspaces/agents open this gets tedious fast.

requested change

A tmux bind -r style repeatable binding: press the prefix once, then tap the key repeatedly to keep stepping — prefix then K K K K — until I press something else or a short timeout elapses.

Ideally an opt-in per-binding flag + a global repeat timeout:

[keys]
next_workspace     = { key = "prefix+shift+j", repeat = true }
previous_workspace = { key = "prefix+shift+k", repeat = true }
next_agent         = { key = "prefix+ctrl+j", repeat = true }
previous_agent     = { key = "prefix+ctrl+k", repeat = true }

# plus a global repeat window (tmux's repeat-time), e.g.
# repeat_timeout = 500   # ms the prefix stays armed after a repeatable key

(a tmux-style -r shorthand, or a repeatable = [...] list, would work too — whatever fits the schema.)

This seems to generalize something herdr already does: in navigate mode the four pane-focus directions stay repeatable while other actions leave the mode after one press. Extending that same behavior to the next_* / previous_* / cycle_* actions — behind the opt-in flag, off by default — would cover it without changing any current defaults.

why you want this

It turns navigating across many workspaces/agents from a prefix-spam grind into something fast and fluid, and it matches the muscle memory tmux users already have from bind -r. As one general mechanism it also naturally covers tab cycling, pane cycling, and resize, so it's not a one-off addition.

[amirt01]

This was really bothering me so I implemented it on a fork: https://github.com/amirt01/herdr