| name | loopany |
| description | Create, update, and evolve scheduled Loopany agent loops from a coding session. Use when the user wants to turn a task they just did into a recurring/scheduled loop, edit an existing loop's schedule or instructions, or asks to build a Loopany loop. A loop can carry a goal (a finish line) and completes itself when the goal is met; without one it runs indefinitely as a monitor. |
Loopany — build and maintain scheduled loops
Loopany turns a task into a loop that runs automatically on a schedule on this
machine. A loop with a goal is closed — each run judges the goal and the loop
finishes itself once it's met; without one it's an open monitor that runs
indefinitely. Work end to end; keep questions to quick check-ins, don't run a full
interview. Two check-ins are right: if the session has no real task to loop yet, ask
what to build rather than inventing one (references/create.md §1); and when the
user hasn't specified the loop's cadence or per-run output, propose a sensible
default and confirm it before creating (references/create.md §2).
Read the reference for the job (they live on disk next to this file, under
references/):
- Creating a loop (the common case — you just did a task and want it scheduled):
references/create.md. It decides what to build, authors the loop's folder +
task file and an inline config (with an optional goal), and runs loopany new.
- Editing an existing loop (reschedule, rename, pause, set/clear a goal, or
change what it does):
references/update.md.
- How a loop stays coherent and improves over time (the evolution pass that
sharpens its task and workflow from its own run history, then fits its
dashboard to the data):
references/evolve.md.
- How a loop behaves each time it runs (the runtime protocol: the task file as
memory, surfacing only what changed, the report/finish grammar and finish bar, the
schedule levers, and front-matter product conventions):
references/run.md.
The machine is already connected — this skill was installed at user scope for each
coding agent loopany knows about (Claude Code ~/.claude/skills/loopany/, Codex
~/.agents/skills/loopany/) when it connected via loopany up. Just author the
loop and run the loopany CLI; the references cover the exact commands.
