| name | schedule |
| description | Turn work into a scheduled task that runs unattended — "every morning", "each Friday", "remind me in an hour", "run this weekly" — or adjust an existing schedule. Use whenever the operator wants something recurring, delayed, or re-timed. |
Scheduling recurring work
protoAgent's scheduler runs full agent turns on a cadence — any cron
expression, or a one-shot at an ISO timestamp. The tools: schedule_task,
list_schedules, cancel_schedule.
Changing an existing task
Rescheduling, rewording, pausing: find it with list_schedules, then cancel
and re-create with the revised prompt/cadence. Confirm which task before
touching anything if more than one could match.
Creating a new task — distill the session
You are compressing this conversation into a prompt that will run cold,
with no access to anything said here. The discipline:
- Identify the repeatable core. Not "what did we do today" but "what
should happen every time".
- Write a self-contained prompt. Second-person imperative ("Check …",
"Compile …", "Save the report to …"). Include every file path, URL,
account, tool name, and output convention it needs; convert every
relative reference ("this folder", "like last time") into an absolute
one. Include what done looks like and any preferences the operator
stated.
- Name it short and kebab-case:
daily-inbox-digest,
friday-metrics-rollup.
- Pick the cadence. Cron for recurring (
0 8 * * 1-5 = weekday
mornings); an ISO datetime for one-shots ("remind me in an hour" =
now + 1h). If the operator didn't give a clear cadence, propose one and
confirm before creating.
Then call schedule_task. Read the created task back to the operator —
name, cadence in plain words, and one-line summary of what it will do —
so a mis-distilled prompt gets caught now, not at 6am.
protoAgent advantages worth using
Scheduled tasks here run on this machine: they can touch local folders,
use any cron cadence, and their results land in the activity thread. Big
recurring deliverables can also pair with a workflow recipe — mention it
when a task outgrows a single prompt.
