--- title: "Claude routines and scheduled tasks for a project board" description: "A Claude Code routine is a cloud session that keeps working while your laptop is closed. Wire it to the board and it wakes on two events, with a grace period, a cap and a record of every run." url: https://projectify.si/blog/claude-routines-for-a-project-board updated: 2026-10-17 --- # Claude routines and scheduled tasks for a project board Published 2026-10-17 by the Projectify team · Topics: Claude Code, Routines A Claude Code **routine** is a session that runs on claude.ai's side, on a schedule or when something fires it, and keeps working while your laptop is closed. That is the scheduled-task half of the story. The other half is what the routine should do when it wakes, and where its work goes. On a Projectify board, both halves are set up once, per person, per project, in Setup › Routines. This is how routine triggers work on our own board, with the limits as they are built, not as we would like them to be. ## What wakes a routine Projectify does not run a scheduler of its own. The routine's schedule is the routine's; Projectify only fires it when something on the board makes a wake-up worth it. Two events do: - **answer.** You answer or dismiss a question your agent asked. The routine fires after a three-minute grace period, and the fire is skipped if the chat that asked already heard the answer on its next call. So a question answered while the session is still running costs nothing. - **task_done.** Your agent finishes a task, and the woken session takes `next_task`. This is autopilot: the routine picks up the next free, unblocked task in the queue, capped at a number of tasks in a row that you set. Everything else, a timer, a nightly run, is the routine's own schedule. Projectify adds the two events that a timer cannot know about. ## What the woken session gets The fire carries a small text block: which project, which event, which task. The session starts with `start_session(fire: …)`, which links its run to the fire for the run log and tells it why it woke. From there it is an ordinary session: it reads the briefing (the brief, the Always lines, who holds what, the last handoff), claims the task it was woken for or the next one, and ends with a handoff. The fire text is data, not instructions. A routine acts as its person, with that person's permissions, opens pull requests and never pushes to main. Every write is attributed as "Claude for Noor", the same as a session you started by hand. ## The limits, as built A routine that wakes too often is worse than none, so the dispatch is capped: - Each fire has a dedupe key, so the same event never wakes the routine twice. - At most six fires an hour per routine, by default. - At most one autopilot run at a time. - Only rate limits, outages and network errors are retried. A server error is never resent, because the routine's fire API has no idempotency key and a retried 500 could run the work twice. - A 401, 403 or 404, or a routine that reports itself paused, turns the trigger off with a reason and tells its person. Those are the numbers in the code today. They are conservative on purpose; the budget that matters is your attention when you open the app in the morning. ## Setting it up 1. In claude.ai, create a Claude Code routine for the project and copy its fire URL and token. 2. In Projectify, Setup › Routines: paste the URL and token (sealed on the server, never shown again), tick the events that should wake it, and copy the standard instructions Projectify generates into the routine. 3. Test it from the same screen. The run shows up in the project's Activity, attributed to you and your agent. Each person adds their own routine. There is no team routine, because a routine acts as a person and a question is answered by a person. ## What to use it for The routine does its best work on the parts of a project that are waiting on an answer. You ask three questions in the afternoon, answer them from your phone in the evening, and the routine picks each task back up three minutes later. The same goes for the next task after a done one: a small, well-defined queue of S-sized tasks, approved through the Define gates, runs overnight and leaves handoffs. It does not replace a fleet. A routine is one cloud session; a [fleet](/#fleet) is several local workers on your machine taking tasks from the queue, which we are still building. For a board synced from [Trello](/integrations/trello) or [Monday](/integrations/monday), the routine can also run the scheduled sync when a token is stored. ## Try it Connect a session first ([one command](/claude-code#connect)), run a task or two by hand, then add the routine. The board is [free for up to 3 people](/pricing), and routines are included.