Skip to content

Sprint loop

The sprint loop is how a coding agent works GrowthOS tickets. Say: Run the GrowthOS sprint loop for project proj_… The agent loads sprint context, ensures the sprint is ACTIVE, proposes the next unblocked task, and asks before claiming.

After you confirm, it implements one task, runs tests, posts an ## Agent handoff comment, marks DONE, then stops. Say continue for the next task. When the queue is empty it asks before advance_sprint (close active → create/start next → assign open work).

Use the in-app prompt or downloaded SKILL.md. Do not invent a bulk-claim loop.

  • Have the agent pick the next unblocked BACKLOG/TODO task (yours first, then unassigned).
  • Confirm each claim and each advance_sprint.
  • Read ## Agent handoff comments on the task.
  • Link branches, PRs, and commits back to the task.
  • An ACTIVE sprint has claimable work assigned to you or unassigned.
  • You want the agent to stop after each ticket so you can review.
  • A gos_ key and MCP connected (see API Keys & Agents).
  • An active sprint with open tasks.
  • Settings → API Keys & Agents → sprint prompt / Download SKILL.md
  • Sprint board → Connect coding agent
  1. Start or confirm the sprint is ACTIVE.
  2. In Cursor or Claude, paste the in-app prompt or say run the GrowthOS sprint loop for project proj_…
  3. Read the proposed next task and confirm (or pick another).
  4. Let the agent implement, test, comment ## Agent handoff, and mark DONE.
  5. Stop. Say continue only when you want the next task.
  6. When the queue is empty, confirm before advance_sprint.
  • Each ticket has a handoff comment and passing tests before DONE.
  • The agent never claims two tasks in one turn.
  • Leave tasks assigned to you or unassigned so they are claimable.
  • On Connected projects, do not overwrite synced status.
  • Asking the agent to finish the whole sprint in one go.
  • Skipping the confirm-before-claim gate.