fix(studio): improve chat context, thinking tags, streaming, and tool results

- Fix cleanThinkingTags to use proper regex instead of naive ReplaceAll
- Send conversation history (last 20 messages + summary) to AI instead of single message
- Store tool results alongside tool calls so history shows complete execution info
- Stream words instead of characters for smoother SSE rendering
- Add stop button to cancel in-progress AI requests (AbortController)
- Fix markdown rendering: add h2 support, use div for bullets
- Add i18n keys for cancel/stop (EN + FR)

💘 Generated with Crush

Assisted-by: GLM-5.1 via Crush <crush@charm.land>

internal/api/handlers_chat.go

@@ -4,12 +4,15 @@ import (
 	"context"
 	"encoding/json"
 	"net/http"
+	"regexp"
 	"strings"
 
 	"github.com/muyue/muyue/internal/agent"
 	"github.com/muyue/muyue/internal/orchestrator"
 )
 
+var thinkingTagRegex = regexp.MustCompile(`(?s)<[Tt]hink[^>]*>.*?</[Tt]hink>`)
+
 const maxToolIterations = 15
 
 func (s *Server) handleChat(w http.ResponseWriter, r *http.Request) {
@@ -68,12 +71,11 @@ func (s *Server) handleStreamChat(w http.ResponseWriter, orb *orchestrator.Orche
 	}
 
 	ctx := context.Background()
-	messages := []orchestrator.Message{
-		{Role: "user", Content: userMessage},
-	}
+	messages := s.buildContextMessages(userMessage)
 
 	var finalContent string
 	var allToolCalls []map[string]interface{}
+	var allToolResults []map[string]interface{}
 
 	for i := 0; i < maxToolIterations; i++ {
 		resp, err := orb.SendWithTools(messages)
@@ -86,8 +88,13 @@ func (s *Server) handleStreamChat(w http.ResponseWriter, orb *orchestrator.Orche
 		content := cleanThinkingTags(choice.Message.Content)
 
 		if content != "" {
-			for _, ch := range strings.Split(content, "") {
-				writeSSE(map[string]interface{}{"content": ch})
+			words := strings.Fields(content)
+			for i, w := range words {
+				chunk := w
+				if i < len(words)-1 {
+					chunk += " "
+				}
+				writeSSE(map[string]interface{}{"content": chunk})
 			}
 			finalContent = content
 		}
@@ -133,6 +140,14 @@ func (s *Server) handleStreamChat(w http.ResponseWriter, orb *orchestrator.Orche
 			}
 			writeSSE(map[string]interface{}{"tool_result": resultData})
 
+			allToolResults = append(allToolResults, map[string]interface{}{
+				"tool_call_id": tc.ID,
+				"name":         tc.Function.Name,
+				"args":         tc.Function.Arguments,
+				"result":       result.Content,
+				"is_error":     result.IsError,
+			})
+
 			messages = append(messages, orchestrator.Message{
 				Role:       "tool",
 				Content:    result.Content,
@@ -146,7 +161,11 @@ func (s *Server) handleStreamChat(w http.ResponseWriter, orb *orchestrator.Orche
 
 	storeContent := finalContent
 	if len(allToolCalls) > 0 {
-		storeObj := map[string]interface{}{"content": storeContent, "tool_calls": allToolCalls}
+		storeObj := map[string]interface{}{
+			"content":      storeContent,
+			"tool_calls":   allToolCalls,
+			"tool_results": allToolResults,
+		}
 		storeJSON, _ := json.Marshal(storeObj)
 		storeContent = string(storeJSON)
 	}
@@ -157,9 +176,7 @@ func (s *Server) handleStreamChat(w http.ResponseWriter, orb *orchestrator.Orche
 
 func (s *Server) handleNonStreamChat(w http.ResponseWriter, orb *orchestrator.Orchestrator, userMessage string) {
 	ctx := context.Background()
-	messages := []orchestrator.Message{
-		{Role: "user", Content: userMessage},
-	}
+	messages := s.buildContextMessages(userMessage)
 
 	var finalContent string
 
@@ -223,7 +240,59 @@ func (s *Server) handleNonStreamChat(w http.ResponseWriter, orb *orchestrator.Or
 }
 
 func cleanThinkingTags(content string) string {
-	return strings.ReplaceAll(content, "<think", "")
+	return strings.TrimSpace(thinkingTagRegex.ReplaceAllString(content, ""))
+}
+
+const contextWindowMessages = 20
+
+func (s *Server) buildContextMessages(userMessage string) []orchestrator.Message {
+	history := s.convStore.Get()
+	start := 0
+	if len(history) > contextWindowMessages {
+		start = len(history) - contextWindowMessages
+	}
+
+	messages := make([]orchestrator.Message, 0, len(history[start:])+1)
+
+	summary := s.convStore.GetSummary()
+	if summary != "" {
+		messages = append(messages, orchestrator.Message{
+			Role:    "system",
+			Content: "Résumé de la conversation précédente:\n" + summary,
+		})
+	}
+
+	for _, m := range history[start:] {
+		content := m.Content
+		if m.Role == "assistant" {
+			var parsed struct {
+				Content   string `json:"content"`
+				ToolCalls []struct {
+					ToolCallID string `json:"tool_call_id"`
+					Name       string `json:"name"`
+					Args       string `json:"args"`
+				} `json:"tool_calls"`
+			}
+			if err := json.Unmarshal([]byte(content), &parsed); err == nil && parsed.Content != "" {
+				content = parsed.Content
+			}
+		}
+		role := m.Role
+		if role == "system" {
+			continue
+		}
+		messages = append(messages, orchestrator.Message{
+			Role:    role,
+			Content: content,
+		})
+	}
+
+	messages = append(messages, orchestrator.Message{
+		Role:    "user",
+		Content: userMessage,
+	})
+
+	return messages
 }
 
 func (s *Server) autoSummarize() {

internal/api/handlers_shell_chat.go

@@ -136,8 +136,13 @@ func (s *Server) handleShellChatStream(w http.ResponseWriter, orb *orchestrator.
 		content := cleanThinkingTags(choice.Message.Content)
 
 		if content != "" {
-			for _, ch := range strings.Split(content, "") {
-				writeSSE(map[string]interface{}{"content": ch})
+			words := strings.Fields(content)
+			for i, w := range words {
+				chunk := w
+				if i < len(words)-1 {
+					chunk += " "
+				}
+				writeSSE(map[string]interface{}{"content": chunk})
 			}
 			finalContent = content
 		}