Thanks for the detailed report. This will be available in the next version.
Why it happened: while a team run was working, AgentsRoom held every message sent to one of its steps until the run stopped, so the nudge to your idle builder step could not get through. And once a run had reached its End, agents could only finish it or cancel it, never send it back to a step.
What changes: the agents in charge of a run (its callback agent, the agent that started it and the project's lead agent) and you can now reach a step while the run works, as soon as that step has received its instructions. A step that is waiting for your answer to its own question still keeps its messages, so an agent never answers in your place; to let messages through after a delay, use "Deliver mail to an agent waiting for an answer" in Settings > Agents & AI > Agent behavior. A new tool, team_reopen_run, lets an agent send a run that reached its End back to a named step with a note: the step reads the note first and the same run continues from there, instead of a cancel and a new run. It adds no extra cycle, so the team's max cycles still applies.
If it still happens after this update, please reopen this ticket with as many details as possible (a screenshot or a video helps a lot).