show agent reasoning output

cmd/olliesrv/internal/agent/chatlog.go

@@ -34,12 +34,20 @@ func (ag *Agent) initChatHandler() {
 			ag.AppendChat(format.FormatEvent(ev.Role, ev.Name, ev.Content, ev.OutputFormat))
 
 		case "reasoning":
-			// Absorbed silently
+			if streamingRole != ev.Role {
+				if streamingRole != "" {
+					if streamingFence != "" {
+						ag.AppendChat([]byte("\n" + streamingFence + "\n"))
+						streamingFence = ""
+					}
+					ag.AppendChat([]byte("\n" + format.EndTag + "\n"))
+				}
+				ag.AppendChat([]byte(format.RoleDelim(ev.Role) + "\n"))
+				streamingRole = ev.Role
+			}
+			ag.AppendChat(format.FormatEvent(ev.Role, ev.Name, ev.Content, ev.OutputFormat))
 
 		case "tool":
-			if ev.Name == "reasoning_think" {
-				return
-			}
 			if streamingRole == "tool" {
 				if ev.Content == "" {
 					if streamingFence != "" {
@@ -134,6 +142,9 @@ func (ag *Agent) ReplayMessages(messages []backend.Message) {
 		case "user":
 			ag.AppendChat(format.FormatEvent("user", "", m.Content, ""))
 		case "assistant":
+			if m.Reasoning != "" {
+				ag.AppendChat([]byte(format.RoleDelim("reasoning") + "\n" + m.Reasoning + "\n" + format.EndTag + "\n"))
+			}
 			if m.ID != "" {
 				ag.AppendChat([]byte(format.AssistantHeaderPrefix + m.ID + "]]]\n"))
 			} else {

cmd/olliesrv/internal/agent/loop.go

@@ -231,7 +231,7 @@ func (ag *Agent) autoCompact(ctx context.Context) (int, error) {
 	}
 	ag.emit(Event{Role: "info", Content: "auto-compacting context...\n"})
 	ag.SetState("compacting")
-	n, err := ag.runCompact(ctx, "auto")
+	n, err := ag.runCompact(context.WithoutCancel(ctx), "auto")
 	if err != nil {
 		return n, fmt.Errorf("auto-compact: %w", err)
 	}

cmd/olliesrv/internal/agent/turn.go

@@ -192,7 +192,7 @@ func (ag *Agent) executeTurn(ctx context.Context, input string) string {
 		if compactLimit := ag.autoCompactLimit(actCtx); compactLimit > 0 && tokens >= compactLimit {
 			ag.emit(Event{Role: "info", Content: "auto-compacting context...\n"})
 			ag.SetState("compacting")
-			if _, err := ag.runCompact(actCtx, "auto"); err != nil {
+			if _, err := ag.runCompact(context.WithoutCancel(actCtx), "auto"); err != nil {
 				ag.log.Error("auto-compact failed: %v", err)
 			}
 			ag.SetState("thinking")
@@ -256,7 +256,11 @@ func (ag *Agent) executeTurn(ctx context.Context, input string) string {
 			ag.history.messages = ag.history.messages[:preTurnLen]
 			ag.emit(Event{Role: "info", Content: "context overflow — compacting and retrying...\n"})
 			ag.SetState("compacting")
-			if _, cerr := ag.runCompact(actCtx, "overflow"); cerr != nil {
+			if _, cerr := ag.runCompact(context.WithoutCancel(actCtx), "overflow"); cerr != nil {
+				break
+			}
+			if actCtx.Err() != nil {
+				err = actCtx.Err()
 				break
 			}
 			ag.history.appendUserMessage(input)