You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Agent Activity (Open Agent Activity…) is a small popup built before the unified stage (#1013). With a real fleet of 30–50 agents it cannot answer the questions it exists for. The evidence and full decomposition are in #1105 (docs/decomposition/agent-activity-redesign.md), recorded against the owner's live window:
it has no "needs you" signal: no pending permission or question, no failed provider turn, no usage limit, no blocked goal loop, no stuck queued prompt;
rows are unidentifiable: 3 of 33 sessions have a title, so most rows read as the folder name (agent-code, three times);
it lists panes, not agents, so parked pool sessions (the ones most likely to be closable) are hard to reach;
it has no filter, and actions are hover-only and one at a time.
Decisions (owner, 2026-09-24)
Full screen, not a popup.
Keep Close Old Agents and Close Idle Orchestration Agents as they are. They are quick one-shot commands; Agent Activity does not replace them.
Row name when nobody set a title: the agent's Goal, then the folder.
Intended behaviour
One full-screen view that answers, for every agent in the window (lane or pool):
Needs you: live condition (permission, question, trust…), failed provider turn, provider/process error, usage limit, goal loop blocked or paused, queued prompt that is not being sent. Each row says why.
Working: live agents with their TLDR line.
Idle / Exited: oldest first, selectable in bulk.
Plus: type-to-filter, one keyboard grammar (↑/↓, Enter focus, Space select, ⌫ close, Esc), and bulk close through the approved-close executor (bulkClose.ts) with a confirmation.
Problem
Agent Activity (Open Agent Activity…) is a small popup built before the unified stage (#1013). With a real fleet of 30–50 agents it cannot answer the questions it exists for. The evidence and full decomposition are in #1105 (
docs/decomposition/agent-activity-redesign.md), recorded against the owner's live window:agent-code, three times);Decisions (owner, 2026-09-24)
Intended behaviour
One full-screen view that answers, for every agent in the window (lane or pool):
Plus: type-to-filter, one keyboard grammar (↑/↓, Enter focus, Space select, ⌫ close, Esc), and bulk close through the approved-close executor (
bulkClose.ts) with a confirmation.Acceptance criteria
closeGrantedSessions, and refuses a target that started working after approval.closedModalDerivations).Refs #1105