feat: add token usage tracking with real API data and fallback estima…

internal/agents/conductor.go

@@ -751,7 +751,19 @@ func (a *ConductorAgent) Run(ctx context.Context, input string, mem *memory.Conv
 
 		if choice.Content != "" {
 			if a.Publisher != nil {
-				a.Publisher.Publish("ai_response", choice.Content, a.Name())
+				metadata := map[string]interface{}{}
+				if resp.Usage != nil {
+					metadata["usage"] = map[string]interface{}{
+						"prompt_tokens":     resp.Usage.PromptTokens,
+						"completion_tokens": resp.Usage.CompletionTokens,
+						"total_tokens":      resp.Usage.TotalTokens,
+					}
+				}
+				if len(metadata) > 0 {
+					a.Publisher.PublishWithMetadata("ai_response", choice.Content, a.Name(), metadata)
+				} else {
+					a.Publisher.Publish("ai_response", choice.Content, a.Name())
+				}
 			}
 		}
 

internal/agents/executor.go

@@ -72,7 +72,19 @@ func RunAgentLoop(ctx context.Context, cfg ExecutorConfig) (string, error) {
 
 		choice := resp.Choices[0]
 		if choice.Content != "" && cfg.Publisher != nil {
-			cfg.Publisher.Publish("ai_response", choice.Content, cfg.AgentName)
+			metadata := map[string]interface{}{}
+			if resp.Usage != nil {
+				metadata["usage"] = map[string]interface{}{
+					"prompt_tokens":     resp.Usage.PromptTokens,
+					"completion_tokens": resp.Usage.CompletionTokens,
+					"total_tokens":      resp.Usage.TotalTokens,
+				}
+			}
+			if len(metadata) > 0 {
+				cfg.Publisher.PublishWithMetadata("ai_response", choice.Content, cfg.AgentName, metadata)
+			} else {
+				cfg.Publisher.Publish("ai_response", choice.Content, cfg.AgentName)
+			}
 		}
 
 		// Build assistant message

internal/agents/meta.go

@@ -68,7 +68,19 @@ func (a *MetaAgent) Run(ctx context.Context, input string) (string, error) {
 	}
 
 	if a.Publisher != nil {
-		a.Publisher.Publish("ai_response", content, a.Name())
+		metadata := map[string]interface{}{}
+		if resp.Usage != nil {
+			metadata["usage"] = map[string]interface{}{
+				"prompt_tokens":     resp.Usage.PromptTokens,
+				"completion_tokens": resp.Usage.CompletionTokens,
+				"total_tokens":      resp.Usage.TotalTokens,
+			}
+		}
+		if len(metadata) > 0 {
+			a.Publisher.PublishWithMetadata("ai_response", content, a.Name(), metadata)
+		} else {
+			a.Publisher.Publish("ai_response", content, a.Name())
+		}
 	}
 
 	return content, nil

internal/llm/engine.go

@@ -50,9 +50,17 @@ type FunctionCall struct {
 	Arguments string `json:"arguments"`
 }
 
+// TokenUsage contains token usage information returned by the LLM API.
+type TokenUsage struct {
+	PromptTokens     int64 `json:"prompt_tokens"`
+	CompletionTokens int64 `json:"completion_tokens"`
+	TotalTokens      int64 `json:"total_tokens"`
+}
+
 // Response represents the LLM's response to a GenerateContent call.
 type Response struct {
-	Choices []Choice `json:"choices"`
+	Choices []Choice    `json:"choices"`
+	Usage   *TokenUsage `json:"usage,omitempty"`
 }
 
 // Choice represents a single choice in the LLM response.

internal/llm/engine_openai.go

@@ -129,12 +129,23 @@ func (e *OpenAIEngine) generateStreaming(ctx context.Context, params openai.Chat
 		return nil, fmt.Errorf("openai streaming: %w", err)
 	}
 
+	// Extract real token usage from streaming accumulator
+	var usage *TokenUsage
+	if acc := stream.Current(); acc.Usage.PromptTokens > 0 || acc.Usage.CompletionTokens > 0 {
+		usage = &TokenUsage{
+			PromptTokens:     acc.Usage.PromptTokens,
+			CompletionTokens: acc.Usage.CompletionTokens,
+			TotalTokens:      acc.Usage.TotalTokens,
+		}
+	}
+
 	return &Response{
 		Choices: []Choice{{
 			Content:   content,
 			Reasoning: reasoning,
 			ToolCalls: toolCalls,
 		}},
+		Usage: usage,
 	}, nil
 }
 
@@ -260,5 +271,15 @@ func (e *OpenAIEngine) toResponse(completion *openai.ChatCompletion) *Response {
 
 		resp.Choices = append(resp.Choices, c)
 	}
+
+	// Extract real token usage from API response
+	if completion.Usage.PromptTokens > 0 || completion.Usage.CompletionTokens > 0 {
+		resp.Usage = &TokenUsage{
+			PromptTokens:     completion.Usage.PromptTokens,
+			CompletionTokens: completion.Usage.CompletionTokens,
+			TotalTokens:      completion.Usage.TotalTokens,
+		}
+	}
+
 	return resp
 }

internal/tui/i18n.go

@@ -51,7 +51,6 @@ type translations struct {
 	ConfirmDialogYes     string
 	ConfirmDialogNo      string
 	// Command mode (vim-like modal editing)
-	CommandModePrompt   string
 	CommandModeTips     string
 	CommandModeIdleTips string
 	EditModeTips        string
@@ -97,7 +96,6 @@ var langMap = map[Language]translations{
 		ConfirmCancelMessage:             "确定要取消当前任务吗？",
 		ConfirmDialogYes:                 "确认 (Enter)",
 		ConfirmDialogNo:                  "取消 (Esc)",
-		CommandModePrompt:                "命令",
 		CommandModeTips:                  "gg/G:首/尾  j/k:上下  f/b:翻页  ctrl+d/u:半页  i:编辑  ctrl+e:编辑模式  ZZ:退出",
 		CommandModeIdleTips:              "gg/G:首/尾  j/k:上下  f/b:翻页  ctrl+d/u:半页  ::命令  /:搜索  ?:帮助  i:编辑  ZZ:退出",
 		EditModeTips:                     "ctrl+s:提交  ctrl+e:命令模式  ctrl+h:历史  ctrl+l:语言  ctrl+c:退出",
@@ -161,7 +159,6 @@ var langMap = map[Language]translations{
 		ConfirmCancelMessage:             "Are you sure you want to cancel the current task?",
 		ConfirmDialogYes:                 "Confirm (Enter)",
 		ConfirmDialogNo:                  "Cancel (Esc)",
-		CommandModePrompt:                "COMMAND",
 		CommandModeTips:                  "gg/G:top/btm  j/k:scroll  f/b:pgdn/up  ctrl+d/u:half  i:edit  ctrl+e:edit  ZZ:quit",
 		CommandModeIdleTips:              "gg/G:top/btm  j/k:scroll  f/b:pgdn/up  ctrl+d/u:half  ::cmd  /:search  ?:help  i:edit  ZZ:quit",
 		EditModeTips:                     "ctrl+s:submit  ctrl+e:cmd  ctrl+h:history  ctrl+l:lang  ctrl+c:quit",
@@ -288,8 +285,6 @@ func (lm *LanguageManager) GetText(key string) string {
 		return translations.ConfirmDialogYes
 	case "ConfirmDialogNo":
 		return translations.ConfirmDialogNo
-	case "CommandModePrompt":
-		return translations.CommandModePrompt
 	case "CommandModeTips":
 		return translations.CommandModeTips
 	case "CommandModeIdleTips":