fix: reset MessageTool.sentInRound per processing round

ResetSentInRound() was never called by the agent loop, so once the LLM
used the message tool in any round the flag stayed true forever. This
caused all subsequent direct-answer responses to be suppressed by the
alreadySent check in llmWorker, leaving only the "Thinking..." placeholder
visible on Telegram.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

pkg/agent/loop.go

@@ -584,6 +584,16 @@ func (al *AgentLoop) llmWorker(ctx context.Context, queue <-chan bus.InboundMess
 			return
 		}
 
+		// Reset per-round message-tool state so a previous round's
+		// tool-sent flag does not suppress this round's response.
+		if defaultAgent := al.registry.GetDefaultAgent(); defaultAgent != nil {
+			if tool, ok := defaultAgent.Tools.Get("message"); ok {
+				if mt, ok := tool.(*tools.MessageTool); ok {
+					mt.ResetSentInRound()
+				}
+			}
+		}
+
 		response, err := al.processMessage(ctx, msg)
 		if err != nil {
 			response = fmt.Sprintf("Error processing message: %v", err)

pkg/tools/message_test.go

@@ -252,3 +252,29 @@ func TestMessageTool_Parameters(t *testing.T) {
 		t.Error("Expected chat_id type to be 'string'")
 	}
 }
+
+func TestMessageTool_ResetSentInRound(t *testing.T) {
+	tool := NewMessageTool()
+	tool.SetSendCallback(func(channel, chatID, content string) error {
+		return nil
+	})
+
+	ctx := WithToolContext(context.Background(), "ch", "cid")
+
+	// First round: tool sends a message
+	tool.Execute(ctx, map[string]any{"content": "hello"})
+	if !tool.HasSentInRound() {
+		t.Fatal("expected sentInRound=true after Execute")
+	}
+
+	// Reset for second round
+	tool.ResetSentInRound()
+	if tool.HasSentInRound() {
+		t.Fatal("expected sentInRound=false after ResetSentInRound")
+	}
+
+	// Second round: tool is NOT used (direct answer) → flag stays false
+	if tool.HasSentInRound() {
+		t.Error("expected sentInRound=false when tool was not used in this round")
+	}
+}