Automation Quickstart
Use recurring-tasks for assistant work that should run repeatedly without an
interactive request each time. Invoke the underlying skill directly for a
one-off run, and use llm-wakeup rather than recurring automation for resuming
one session after a usage reset or timeout.
Before enabling a job
- Configure the assistant launchers with
install-launchersif the job needs unattended agent execution. - Configure and test the skill the job will invoke. For personal planning and inbox automation, complete the Personal Assistance Quickstart first.
- Review the unattended execution boundary. A scheduled job cannot stop for questions or approvals.
Recurring jobs are experimental in the first public release and are never enabled merely because their underlying skill is installed.
Platform support and help
Recurring scheduling is implemented for Linux, macOS, and Windows, but it has been thoroughly tested only on Linux. Scheduler setup, triggering, status, or health checks may therefore fail in platform-specific ways on macOS or Windows.
If automation produces an error, show the exact error to the assistant and ask
it to diagnose and help resolve the problem. If the issue remains unresolved or
would be useful to the Famulus maintainer, invoke send-feedback in that same
session. It will prepare a redacted report from the established evidence and
send it only after you review and approve the complete message.
What to use when
| Need | Skill |
|---|---|
| Configure assistant commands and background execution support | install-launchers |
| Connect Google services needed by a job | connect-google |
| Enable, disable, test, inspect, or repair a recurring assistant job | recurring-tasks |
| Run the work once right now | Invoke the underlying skill directly |
| Resume one assistant session after a reset or timeout | llm-wakeup |
| Follow what an unattended run actually did, during or after it | milestone-logging |
Enable and verify
Ask recurring-tasks to enable the named job and schedule. After setup, test
the job once and confirm that it produced the expected result. If it fails or
stops running, use recurring-tasks to inspect the failure and help repair it.
Common recurring workflows include inbox triage and daily planning:
email-triage can add obligations to todo and opportunities to triage,
while daily-plan prepares the day's plan from the current lists and calendar.
These jobs operate on the same persistent personal-assistance state as their
interactive versions; scheduling changes when they run, not what data they
use.
Operate and troubleshoot
Use recurring-tasks to inspect a failed job, change its schedule, or disable
it when you no longer want it to run.
A failed job leaves a transcript of tool calls but no account of what the
assistant was trying to do, which is rarely enough to tell a broken job from a
job that ran and found nothing. milestone-logging is the other half: a job
that records its milestones under a run id can be read back through its
generated timeline interface via the shared famulus_dispatcher MCP server, even after
the session that started it has ended. The milestone-logging skill owns the
recording and timeline protocol.