Retries and assignment
When a work item fails, the board and the detail page give separate controls for retrying execution and retrying triage. Assignment works from the board card or from the detail page.
Retry execution
Retry execution appears in the card's ⋮ menu only when the work item is in the Failed state. Press it to re-dispatch the latest approved plan. The item moves straight to Executing and does not re-plan.
To retry several failed items at once, click the checkbox on each card to select them, then press Retry in the bulk action bar at the bottom of the screen. The bar applies the retry only to the selected cards that are in the Failed state.
A retry requires a previously approved plan. If the plan was rejected or does not exist, approve or create a plan first.
Retry triage
Triage failures are separate from execution failures. When the Atlas triage card shows a yellow warning, press Retry triage to replay the triage step. The button is on the work item detail page, inside the triage card, not on the board.
After a triage failure you can still write and approve a plan manually. Retry triage and manual plan creation are independent options; either can unblock the item.
Assigning work items
A work item has an owner and can have additional assignees. The owner is always the first name in the assignee list.
To change the assignee from the board, click the assignee avatar in the card footer to open the picker, then toggle members on or off. From the work item detail page, use the properties form to update the owner and assignees directly.
To assign several items at once, select the cards, press Assign in the bulk action bar, and choose a member. The chosen member becomes the owner for each selected item.
FAQ
Does retrying re-plan?
No. Retry execution re-dispatches the existing approved plan without changes. If you want different steps, reject the current plan, write a new version, and approve it before retrying.
Is there a limit on retries?
There is no built-in limit on manual work-item retries. Generation run failures have an automatic threshold after which the run is dead-lettered and will not retry on its own. See Runs for generation run behavior.