GuidesA nightly job
A nightly job with run history
Send an agent a prompt every night, read what each run did, and clear a run that failed.
This guide sets up a scheduleSchedule. A prompt sent on a cron time, once at a set time, or on demand.: a prompt that goes to an agent every night, with a record of every runRun. One occurrence of a schedule. Every run is recorded, and a failed or uncertain one stops its schedule until you check it.. A run that fails stops the schedule until you have looked, so a broken job never repeats unseen.
The examples follow Mara, who wants a dependency check in atlas-api at 02:30.
Set it up
Section titled “Set it up”Add the schedule
Section titled “Add the schedule”From a shell:
vendetta schedule add --id deps-nightly --cron '30 2 * * *' \
--timezone Europe/Berlin --reuse --provider claude \
--cwd ~/projects/atlas-api --tmux-session atlas-api \
--prompt 'Check for outdated dependencies. Report what changed; change no files.'--crontakes five numeric fields: minute, hour, day, month, weekday. Without--timezoneit reads UTC.--tmux-sessionnames the project whose Agents tab the agent opens in.--reusesends every run to the same agent: the first run starts it, later runs prompt it again and resume it if it has exited.--newstarts another agent each night instead.
Or in the web client or the iPhone app, with Advanced on: open Schedules, choose New schedule, and fill in Name, Runs (Repeatedly, Once, or On demand), Conversation (New each time, Same agent each time, or Resume one), the project, the folder, and the Prompt.
In the chooser, s starts a new schedule for the selected project.
Check the scheduler runs
Section titled “Check the scheduler runs”install.sh sets the schedulerScheduler. The process that runs schedules at their time. install.sh sets it to run at login.Scheduler. The process that runs schedules at their time. install.sh sets it to run at login. to run at login. Confirm it, and see every schedule
with its next time:
vendetta schedule statusTry it once
Section titled “Try it once”vendetta schedule run-now --id deps-nightlyThis is a real run. The agent starts in the atlas-api Agents tab and gets the
prompt; its chat shows in the apps like any other, with the schedule that started it
on its overview card. In the apps the button is Run now.
Read the history
Section titled “Read the history”vendetta schedule history --id deps-nightlyHistory keeps the latest 1,000 runs, oldest first, and stays after a schedule is removed. In the apps, a schedule’s History lists each run as one of:
| Outcome | Meaning |
|---|---|
| Prompt sent | The agent accepted the prompt |
| Skipped | The run was held back and nothing was sent |
| Failed | The run did not go through |
| Outcome unknown | Vendetta can’t tell whether the agent got it |
Prompt sent means the prompt was submitted. Whether the job got done is in the agent’s chat.
Clear a failed run
Section titled “Clear a failed run”A failed run, or one with an unknown outcome, pauses that schedule. Nothing runs again until you say so:
-
In the apps the schedule moves to Needs attention and opens with Paused for your check. Look at the agent, then choose I checked. Acknowledge and confirm.
-
In the chooser’s schedules screen it sits under Needs your check; a acknowledges and enables it.
-
From a shell:
vendetta schedule enable --id deps-nightly --acknowledge-uncertain
Acknowledging never retries the old run. The next one goes out at its time.
A resume that can’t be verified, or a start that runs into a dialog or a question, fails this way. It never falls back to a fresh agent.
Pause, change, remove
Section titled “Pause, change, remove”vendetta schedule disable --id deps-nightly # pause; enable resumes
vendetta schedule update --id deps-nightly --cron '0 3 * * *' ... # same options as add
vendetta schedule remove --id deps-nightly # its history staysIn the apps, the Enabled switch pauses a schedule, and Edit and Delete do the rest.
What to know about timing
Section titled “What to know about timing”- A zoned schedule follows local clock time through daylight saving. A time that repeats when clocks go back runs once; a time that is skipped runs right after the jump.
- Cron minutes missed while the scheduler was down are not caught up. A one-time
schedule (
--at) that was missed runs once when the scheduler returns.
To keep the schedule in the repository, declare it in the project’s
vendetta.toml. To run it when another service calls a
URL, see Webhooks. The full reference is
Scheduling.