Coordinator stuck "Working" while only a background subagent runs — queued input not answered or shown as pending

Describe the bug

Sharing something I've hit several times in the CLI (v1.0.70). Upfront: I don't really know how Copilot's orchestration works internally — the log analysis below was done by Copilot itself at my request. The parts I can vouch for first-hand are what I saw in the UI.

• The busy indicator showed "Working", and prompts I typed were pushed to a queue — not answered, and not shown as "(pending)".
• I tried Ctrl+X B (move task to background) and it said "No running sync task to move to background" — so the CLI itself says nothing sync is running, yet it's "Working" and queueing my input.
• In /tasks , the only thing running was one subagent, marked [background] .
• The coordinator became responsive only after I manually cancelled that background subagent in /tasks . Cancelling the sole [background] task is what released my queued prompt — which makes it look like a "background" task was actually blocking the main agent.

Affected version

v1.0.70

Steps to reproduce the behavior

No response

Expected behavior

if the coordinator is free it should answer immediately; if busy, it should take my prompt as steering and show "(pending)" and act next turn. Actual: queued for minutes with nothing visibly happening except a background subagent.

Additional context

From Copilot's log analysis (may be imperfect, not internal state): it found two windows where the coordinator went quiet for ~13–17 min with only a background subagent running. In the clearest one, the subagent-completion notification actually fired fine — so this isn't just a "dropped notification"; it looks more like the coordinator being blocked by / kept "Working" behind its own background subagent, and/or how typed input is classified while a subagent runs. It also ruled out sync-blocking (all read_agent calls were wait:false , tasks were mode:background ).

Commencez par reproduire le comportement dans CLI v1.0.70 tout en observant /tasks, l’indicateur Working, les entrées mises en file d’attente et Ctrl+X B. Comparez les sessions avec un seul sous-agent [background] et inspectez les fenêtres de journal read_agent décrites dans le rapport ; c’est terminé lorsque le coordinateur répond ou marque visiblement l’entrée comme étant en attente sans nécessiter d’annulation manuelle.