Advance workflow stages from the state-driven coordinator

internal/agents/manager.go

@@ -367,11 +367,6 @@ func (m *Manager) Recover(ctx context.Context) error {
 		return err
 	}
 
-	staged, err := m.recoverStages(ctx)
-	if err != nil {
-		return err
-	}
-
 	err = m.recoverPreviews(ctx)
 	if err != nil {
 		return err
@@ -399,9 +394,6 @@ func (m *Manager) Recover(ctx context.Context) error {
 	}
 
 	for _, sess := range sessions {
-		if staged[sess.Id] {
-			continue
-		}
 		task, err := m.taskFor(ctx, sess)
 		if err != nil {
 			slog.Error("resume session: load task", "session", sess.Id, "task", sess.TaskId, "err", err)
@@ -412,6 +404,23 @@ func (m *Manager) Recover(ctx context.Context) error {
 		if err != nil {
 			return err
 		}
+		started, err := m.store.SessionStarted(ctx, sess.Id)
+		if err != nil {
+			return err
+		}
+		finished, err := m.store.StageFinished(ctx, sess.Id)
+		if err != nil {
+			return err
+		}
+		if finished || !pending && !started {
+			m.endMergeTurn(sess.Id, true)
+			err = m.store.SetSessionState(ctx, sess.Id, api.AgentSessionStateIdle, "byre", nil)
+			if err != nil {
+				return err
+			}
+			slog.Info("session left idle at startup: its stage finished or it never got its first message", "session", sess.Id, "task", task.Id)
+			continue
+		}
 		if pending {
 			found, err := m.deliverRecorded(ctx, sess, task)
 			if err != nil {
@@ -1307,9 +1316,6 @@ func (m *Manager) turnEnded(ctx context.Context, l *live, res endResult) error {
 
 	if end.Discard != store.DiscardNone {
 		m.logUndelivered(ctx, l.id, "turn ended in discard")
-		if end.Discard == store.DiscardHeld && stageFinished {
-			go m.advanceStage(l)
-		}
 		return nil
 	}
 	if handoff {
@@ -1322,7 +1328,6 @@ func (m *Manager) turnEnded(ctx context.Context, l *live, res endResult) error {
 		return nil
 	}
 	if stageFinished {
-		go m.advanceStage(l)
 		return nil
 	}
 	go m.afterTurn(context.WithoutCancel(ctx), l)