Recipes are WHAT to run. Routines are WHEN. Skills are HOW.
“Every weekday at 7, brief me on my competitors.” That’s a recipe (the brief) on a routine (7:00 weekdays), and the researcher may pull a skill (how to source claims properly) while doing it.
Open WORK › ▣ RECIPES. There are 101 built-in recipes in eight shelves — developer, research, creator, ops, business, money, data, general. Pick one, fill in its two or three fields, and it lands on TASKS as a real job for your crew.
＋ SAVE A RECIPE turns any job you keep typing into a form. Edit, tweak, or fork a built-in the same way.
Every recipe card has ◷ MAKE ROUTINE. One click turns the WHAT into a WHEN.
The recommender ranks all 101 against what you actually do and offers one at a time — with the reason, or not at all.
Open WORK › ∞ AUTOMATION and hit + ADD ROUTINE. Write the job (or start from a recipe), pick who runs it, and answer WHEN SHOULD IT RUN? — a morning brief, a nightly summary, a recurring check.
Plain words. “Summarize what changed in my inbox since yesterday and flag anything that needs a reply.” Choose the agent.
Every weekday at 7. Every Sunday night. Every 2 hours. The picker shows exactly when the next run will be.
Routines only fire while scheduling is on: ▶ ENABLE SCHEDULING. Off, a routine is saved but dormant — the panel says ○ SCHEDULING IS OFF so nothing is ambiguous.
Where does the job land? At the INBOX. A routine is exactly the “outside work” the front door exists for — it drops onto your belts and flows through your line like any other job. No belts? It goes straight to the agent’s bay.
A skill is a procedure document. Agents don’t launch skills — they consult them when a task matches. Open BUILD › ⇄ ABILITIES:
| Section | What’s in it |
|---|---|
| SKILL LIBRARY | 73 pre-installed procedures, grouped by kind. A skill stays locked until the agent has the gear it needs — a browsing skill needs the DISH placed. |
| AGENT SKILLS | Procedures your agents wrote or learned themselves. They appear as a compact index in future runs; the full body loads only when a task matches. |
| SKILL EXCHANGE | Paste an HTTPS or GitHub link to a SKILL.md and hit INSPECT — you see provenance and a guard review before anything installs. Or import an exported package. |
There is no blank “write a skill” form. Skills enter the station two ways: an agent writes one from experience, or you install one you’ve inspected. A routine can name up to 8 skills its agent should have on hand.
A ROUTINE fires at 7:00 and drops a job at the INBOX.
The job is a RECIPE — “competitor brief” with your three fields filled in.
The researcher pulls a matching SKILL mid-run — how to source every claim.
The brief lands in the OUTBOX. You rate it; the next run is built from what you valued.
No. The station is local-first — the scheduler lives in the app on your machine. Leave it running (it’s happy minimized) and routines fire on time.
Yes. It drops in at the INBOX and follows your belts through every desk, filter, and joiner like any other job.
A routine is a job you defined on a clock you set. Night Shift is the station choosing its own small job while you’re away, inside a leash. Routines run whether you’re there or not.
On your machine, with the rest of your station state. Nothing you save is uploaded anywhere.