Skip to main content
Routines fire prompts on a schedule, without you starting the conversation. This guide walks through scheduling a daily morning briefing on the orchestrator, then confirming it works. The same shape applies to any routine on any agent. Read Routines first if you have not already. There are limits worth knowing before you build a workflow on top of routines.

​
Before you start

You need:
  • An agent already created, ideally the one you want to attach the routine to. The orchestrator is the natural home for a daily briefing because its body already knows how to triage.
  • A skill for the routine to run. Routines schedule a skill, not a freeform prompt: there is no text box in the form to type an instruction into. If the agent does not have the skill you want yet, the form’s own empty state offers to build one before you go further.

​
1. Open the Add routine form

From the orchestrator’s own profile page, press Add routine. This opens a two-step editor scoped to that agent, so the first step only lists its own skills.
The profile page only offers Add routine before an agent has its first one. To add a second routine to an agent that already has one, open Routines in the rail and press + in its sidebar header instead. If the sidebar is not scoped to one agent, the form’s first step also asks which agent, not just which skill.

​
2. Step 1: pick a skill

The first screen lists the skills this agent already has. Pick one. The instruction the routine will send is derived entirely from the skill you pick: there is nothing else to fill in on this step. If the agent has no skills yet, the screen offers to build one instead of showing an empty list.

​
3. Step 2: say when

The second screen is a plain-terms schedule builder: a frequency (day, or a specific weekday) and a time, offered on the hour and half hour. It also asks where the routine runs: This computer, or Your Agent Computer, which is shown as not set up yet and cannot be chosen, so every routine runs on this computer today. As you choose, the screen shows the sentence it will save, for example:
“On this computer” names where the routine runs today: while Rundock is open, on this machine. Confirm the sentence and press Save routine.
Only the routines of the open workspace run. If you use more than one workspace, a routine you make now runs while this workspace is the one open, and is caught up the same day if you were elsewhere when its slot came due. See Routines for the full explanation.

​
4. Find it in the list

Saving returns you to the Routines list, which is ordered by next run. The new routine’s row reads as a sentence, such as Every day at 5:00am, run: Morning briefing, with the agent’s name beneath it.

​
5. Confirm the routine fires

You do not have to wait until 5am tomorrow to see it work. Every row on the Routines list has a Run now control, the play button, which runs the routine once, straight away. A manual run moves no schedule: the next scheduled run stays exactly where it was.
One routine row in the Routines list: its schedule as a sentence, the agent's name, when it last ran and when it runs next, and the Run now, pause, edit and delete controls on the right.

Run now is the play button on the routine's row.

1

Press Run now

On the routine’s row, press Run now. The row switches to Running now (started manually), and the control is disabled until the run ends. Open the run from here with View run to watch it while it is still in progress.
2

Check what it produced

Rundock does not surface the model’s own response in a conversation. What tells you the routine did its job is either the file it wrote, the task it created, or the run’s own file list (open View run or View last run to see it: every file the run touched, or an honest note if that can’t be established).
3

Check the outcome

Open the run’s record to see how it ended. If it failed, the record shows any reason it gave. Once the routine has run on its schedule, its row shows one of Ran, Caught up, Missed, Failed or Stopped; a manual run’s outcome stays on its own record and does not change the row.
If a run is taking longer than expected, or you started it purely to test and don’t want to wait for it, the run’s own screen carries a Stop control for as long as it is going. Stopping it does not cancel the routine: it runs again at its next real slot.

​
Limits worth knowing

The same limits from the Routines concept page apply here. They are worth restating because they shape what kinds of routine actually work.
Routines fire only while Rundock is running, but they catch up the same day. Close Rundock overnight and open it in the morning, and an overnight routine runs shortly after launch. The daily note is written by the time you have made coffee. What you do not get is the run happening at the scheduled hour, or happening at all on a day you never open Rundock. If the hour itself matters, or a day might pass without you opening it, host Rundock somewhere always-on: a small VPS, or a machine you leave running. Anthropic’s managed Claude Code Routines are an option for work that can run on their cloud rather than against your local workspace.
Routine output itself is not captured to a visible conversation. Rundock records the routine’s lastRun timestamp, status, duration, and a separate list of the files the run changed, but it does not surface the model’s response or its running commentary in a Rundock conversation. Any routine that needs to leave a trace must write that trace itself, through tools (Write to a file, or a tasks or notes connector via MCP).
No notification when a routine completes. You see the row update in the Routines list. To know whether the run did the right thing, check the file or task it was meant to produce, or open the run’s own record.

​
Patterns that work

Three routines that work well in practice, shown here as you might write them by hand in the agent’s frontmatter (see Routine format for the format; a routine made in the form also records its skill, where it runs and its approval):
The shape that ties them together: the prompt is short. It assumes the agent already knows what “the morning briefing” or “the weekly research pipeline” means, whether that lives in the agent’s own instructions or in a skill it has. A routine made through the form always resolves to a one-line instruction naming the skill it schedules; a routine written by hand can say anything, as long as it is one clear instruction. Stagger times if you have multiple routines on the same agent. Two routines firing in the same minute spawn two concurrent Claude Code subprocesses that do not share context.

​
Where to next

Routines concept

What routines are and the limits worth knowing before you build on them.

Routine format

The underlying frontmatter format, for power users and scripting.