Advance workflow stages from the state-driven coordinator

internal/agents/workflows.go

@@ -362,34 +362,24 @@ func (m *Manager) recordStage(ctx context.Context, l *live) error {
 	return m.store.RecordStage(ctx, l.taskID, output, head)
 }
 
-func (m *Manager) advanceStage(l *live) {
+func (m *Manager) advanceStage(sessionID string) {
 	ctx := context.Background()
-	_, err := m.nextStage(ctx, l.id)
-	if err != nil {
-		slog.Error("advance workflow stage", "session", l.id, "err", err)
-		m.recordError(l, fmt.Sprintf("The workflow didn't move on from this stage: %v. byre tries again when it next starts.", err))
-	}
-	m.afterTurn(ctx, l)
-}
-
-func (m *Manager) recoverStages(ctx context.Context) (map[string]bool, error) {
-	ids, err := m.store.FinishedStages(ctx)
-	if err != nil {
-		return nil, err
-	}
-	started := map[string]bool{}
-	for _, id := range ids {
-		next, err := m.nextStage(ctx, id)
-		if err != nil {
-			slog.Error("advance workflow stage", "session", id, "err", err)
-			continue
+	next, err := m.nextStage(ctx, sessionID)
+	if err == nil {
+		slog.Info("advanced workflow stage", "session", sessionID, "next", next)
+	} else {
+		slog.Error("advance workflow stage", "session", sessionID, "err", err)
+		sess, serr := m.store.GetSession(ctx, sessionID)
+		if serr == nil {
+			m.fileError(ctx, sess.TaskId, fmt.Sprintf("The workflow didn't move on from this stage: %v. byre tries again when it next starts.", err))
 		}
-		for _, n := range next {
-			started[n] = true
-		}
-		slog.Info("advanced workflow stage", "session", id, "next", next)
 	}
-	return started, nil
+	m.mu.Lock()
+	l := m.live[sessionID]
+	m.mu.Unlock()
+	if l != nil {
+		m.afterTurn(ctx, l)
+	}
 }
 
 func (m *Manager) nextStage(ctx context.Context, sessionID string) ([]string, error) {