A playbook is a written procedure that your agent follows. unicorn supplies the steps and the data, your agent does the thinking, and the result lands in your briefs or plans. unicorn runs no model itself, so a playbook only happens when your agent runs it. Each playbook needs only unicorn. If you also have Ed or Moodle connected live, your agent may use them to read a spec or a thread in full, but the playbook works without them.

​
The four playbooks

How each one behaves, in terms you can check:
  • weekly-plan puts multi-day work on the earlier days, not the day before it is due. When two deadlines collide, or a state is unclear, it adds a “Needs your input” section and asks one question. A quiet week still produces a plan that says nothing is due, so your other devices find a plan for it.
  • decompose-assignment never overwrites a plan that exists. If it cannot find the assessment’s description, it says so instead of inventing tasks. It sizes the work to the assessment’s weight and plans lighter weekends unless you have told it otherwise.
  • forum-brief remembers where it stopped, so each run covers only what is new. It always writes a brief, saying “Nothing new on the forums” when that is the case, so the next run does not re-read the same period.
  • triage puts something in life/other rather than guess a course, and never changes a label you or your agent set on purpose. See buckets.
All four follow the notes you saved with remember. A note such as “no work on Sundays” changes the plans.

​
Run one now

Ask for it by name, in any client:
To narrow it, say so in the request:
How your client offers them:
  • Claude Code with the plugin installs each playbook as a skill. Claude picks the right one from a request such as “plan my week” or “what did staff post”. Type / and search for the playbook name to choose one yourself.
  • claude.ai serves them as prompts you can pick from your connector.
  • ChatGPT and any client that only calls tools use the run_playbook tool. Say “run the weekly-plan playbook” and the client calls it.
run_playbook returns the procedure, the data it needs, already fetched, and your saved corrections. Your agent then follows the steps, filing briefs and plans as it goes.

​
Schedule them

A scheduled run is a routine: your client’s own scheduler starts a fresh session on a timetable and sends one line, such as “Run the unicorn weekly-plan playbook.” unicorn has no scheduler for playbooks, because unicorn never starts your agent.
A routine runs in the cloud, so it cannot use a plugin or local connection on your computer. It reaches unicorn only through the connector you added to your Claude account. Complete the claude.ai steps in Connect your agent first.

​
Claude Code

The easiest way is to ask. With the plugin installed, say:
The plugin’s setup-routines skill walks you through adding the connection to your account, then creates each routine. To create one yourself:
When it asks which connectors the routine can use, make sure unicorn is included. A routine that calls only unicorn does not use files, so any repository works if it asks for one. You can also create routines in the browser at claude.ai/code/routines, or from the desktop app’s Code tab.

​
Cowork

Open the Schedule tab, click New Task, name it, paste the one-line prompt and choose a cadence. Cowork offers hourly, daily, weekly and weekdays, so pick the nearest match. A time may drift from the table above. Check that unicorn is enabled for the task.

​
ChatGPT

unicorn’s documentation does not cover scheduling in ChatGPT. Run playbooks there on demand with run_playbook.

​
The prompts

​
Check that it worked

After a routine has run, ask your agent “What’s new in unicorn?”. The brief appears with its title, for example “Weekly plan — 2026-W39”. Use Run now in the scheduler to try one before its first real run.