Stable
Shell scheduler
Install the CLI, add a loops.yaml, run it once or leave the watcher open. Any command with an exit code — and it keeps running after the agent closes.
bun add -g github:acoyfellow/loops-yaml Where loops run
One idea.
Every runtime.
A loop is an interval and a thing to do. That works the same whether the “thing” is a shell command on a schedule or a prompt to a coding agent while you work. Everything here ships today — install from the public repository, nothing hosted, no account.
Stable
Pi
Install the Pi package, reload the session, then describe the interval and the condition that ends the loop.
pi install git:github.com/acoyfellow/loops-yaml /loop every 30s check MR !38955; stop when it merges Stable
omp
Pi-family CLI, so the same extension loads directly. Install it as a plugin, then use /loop exactly as in Pi.
omp install github:acoyfellow/loops-yaml /loop every 30s check MR !38955; stop when it merges 
StableOpenCode
Point OpenCode at the plugin in opencode.json, restart, then describe the interval and the stop condition.
{ "plugin": ["/ABSOLUTE/PATH/TO/loops-yaml/extensions/opencode/index.ts"] } every 30s check MR !38955; stop when it merges Any other agent?
Most agents can already loop themselves — the shell scheduler runs any command, and any agent with a shell can re-prompt on a timer. The native extensions above just make the loop session-aware: idle-only, with a real stop condition, instead of a blunt cron tick.
See how the two modes differ →