@ww_nero/mini-cli 1.0.91 → 1.0.92

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ww_nero/mini-cli",
-  "version": "1.0.91",
+  "version": "1.0.92",
   "description": "极简的 AI 命令行助手",
   "bin": {
     "mini": "bin/mini.js"

package/src/chat.js

@@ -26,7 +26,7 @@ const {
   mergeReasoningContent,
   renderAssistantOutput,
   sanitizeContentWithThink,
-  printUsageTokens
+  formatUsageTokens
 } = require('./utils/output');
 const {
   createHistoryFilePath,
@@ -598,21 +598,29 @@ const startChatSession = async ({
   const DEFAULT_PROMPT = `${chalk.cyan('用户>')} `;
   rl.setPrompt(DEFAULT_PROMPT);
 
-  const LOADING_PROMPT_BASE_TEXT = '⌛ 回答中...（按 Esc 取消）';
+  const LOADING_PROMPT_BASE_TEXT = '思考中...';
   const FLOW_GRADIENT_COLORS = ['#FFF7CC', '#FFE58F', '#FFD666', '#FFC53D', '#FFA940', '#FF7A45'];
   const FLOW_DIM_COLOR = '#5F5F6A';
-  const loadingPromptCharacters = Array.from(LOADING_PROMPT_BASE_TEXT);
-  const loadingPromptCharCount = loadingPromptCharacters.length || 1;
   const supportsHexColor = typeof chalk.hex === 'function';
   const flowGradientFns = supportsHexColor
     ? FLOW_GRADIENT_COLORS.map((hex) => chalk.hex(hex))
     : [];
   const dimTextFn = supportsHexColor ? chalk.hex(FLOW_DIM_COLOR) : chalk.yellow;
 
+  const buildLoadingPromptText = () => {
+    if (!loadingPromptTokensText) {
+      return `${LOADING_PROMPT_BASE_TEXT}(按Esc取消)`;
+    }
+    return `${LOADING_PROMPT_BASE_TEXT}(${loadingPromptTokensText}，按Esc取消)`;
+  };
+
   const formatLoadingPrompt = (frame) => {
+    const promptText = buildLoadingPromptText();
     if (!supportsHexColor) {
-      return chalk.yellow(LOADING_PROMPT_BASE_TEXT);
+      return chalk.yellow(promptText);
     }
+    const loadingPromptCharacters = Array.from(promptText);
+    const loadingPromptCharCount = loadingPromptCharacters.length || 1;
     const gradientLength = Math.min(flowGradientFns.length, loadingPromptCharCount);
     const startIndex = frame % loadingPromptCharCount;
     return loadingPromptCharacters.map((char, index) => {
@@ -628,6 +636,7 @@ const startChatSession = async ({
   let loadingPromptFrame = 0;
   let loadingPromptTimer = null;
   let loadingPromptVisible = false;
+  let loadingPromptTokensText = '';
 
   const getCurrentLoadingPromptText = () => formatLoadingPrompt(loadingPromptFrame);
 
@@ -635,7 +644,7 @@ const startChatSession = async ({
     if (!loadingPromptVisible || !supportsHexColor) {
       return;
     }
-    loadingPromptFrame = (loadingPromptFrame + 1) % loadingPromptCharCount;
+    loadingPromptFrame += 1;
     refreshLoadingPrompt();
   };
 
@@ -681,6 +690,15 @@ const startChatSession = async ({
     writeLoadingPrompt();
   };
 
+  const updateLoadingPromptTokens = (usage) => {
+    const nextText = formatUsageTokens(usage) || '';
+    if (nextText === loadingPromptTokensText) {
+      return;
+    }
+    loadingPromptTokensText = nextText;
+    refreshLoadingPrompt();
+  };
+
   const clearLoadingPrompt = () => {
     if (!loadingPromptVisible) {
       return;
@@ -908,6 +926,7 @@ const startChatSession = async ({
           tools: toolSchemas,
           onToken: handleStreamChunk,
           onReasoningToken: handleReasoningChunk,
+          onUsage: updateLoadingPromptTokens,
           onRetry: (retryCount, error) => {
             const detail = error ? `（${error.message}）` : '';
             process.stdout.write(`\n${chalk.yellow(`请求失败，${retryCount} 次重试中${detail}...`)}\n`);
@@ -915,6 +934,7 @@ const startChatSession = async ({
         });
 
         finalizeStreamState();
+        updateLoadingPromptTokens(result.usage);
 
         const rawReasoning = (reasoningFromModel || result.reasoningContent || '').trim();
         const mergedReasoning = mergeReasoningContent(
@@ -1000,7 +1020,6 @@ const startChatSession = async ({
               content: toolContent
             });
           }
-          printUsageTokens(result.usage);
           persistHistorySafely();
 
           // 检测是否需要触发 compact
@@ -1052,7 +1071,6 @@ const startChatSession = async ({
         );
         messages.push(assistantMessage);
         persistHistorySafely();
-        printUsageTokens(result.usage);
         break;
       }
     };

package/src/llm.js

@@ -68,7 +68,7 @@ const sanitizeOptions = (options = {}) => {
   }, {});
 };
 
-const chatCompletion = async ({ endpoint, messages, abortController, onToken, onReasoningToken, onRetry, onComplete, tools = [] }) => {
+const chatCompletion = async ({ endpoint, messages, abortController, onToken, onReasoningToken, onUsage, onRetry, onComplete, tools = [] }) => {
   const controller = abortController || new AbortController();
   const includeToolCalls = Array.isArray(tools) && tools.length > 0;
 
@@ -112,6 +112,11 @@ const chatCompletion = async ({ endpoint, messages, abortController, onToken, on
         onReasoningToken(chunk, aggregate);
       }
     },
+    onUsage: (usage) => {
+      if (onUsage) {
+        onUsage(usage);
+      }
+    },
     onComplete: () => {
       if (onComplete) {
         onComplete(fullContent);

package/src/request.js

@@ -88,6 +88,7 @@ const processStreamResponse = (response, options = {}) => {
   const {
     onContent = null,
     onReasoningContent = null,
+    onUsage = null,
     onComplete = null,
     abortController = null,
     includeToolCalls = false
@@ -138,6 +139,9 @@ const processStreamResponse = (response, options = {}) => {
           const payload = JSON.parse(event.data);
           if (payload.usage && typeof payload.usage === 'object') {
             usage = payload.usage;
+            if (onUsage) {
+              onUsage(usage);
+            }
           }
           const choice = payload.choices && payload.choices[0];
           if (!choice || !choice.delta) return;

package/src/utils/output.js

@@ -77,7 +77,7 @@ const printReasoningContent = (text, ensureNewline) => {
 
   if (!normalized) return;
   ensureNewline();
-  console.log(chalk.gray(`思考：\n${normalized}`));
+  console.log(chalk.gray(normalized));
 };
 
 const printAssistantContent = (text, ensureNewline) => {
@@ -116,16 +116,16 @@ const formatTokensToK = (totalTokens) => {
   return `${(totalTokens / 1000).toFixed(1)}k`;
 };
 
-const printUsageTokens = (usage) => {
+const formatUsageTokens = (usage) => {
   if (!usage || typeof usage !== 'object') {
-    return;
+    return null;
   }
   const totalTokens = usage.total_tokens;
   const formatted = formatTokensToK(totalTokens);
   if (!formatted) {
-    return;
+    return null;
   }
-  console.log(chalk.gray(`已使用tokens: ${formatted}`));
+  return `会话tokens: ${formatted}`;
 };
 
 module.exports = {
@@ -134,5 +134,5 @@ module.exports = {
   printAssistantContent,
   renderAssistantOutput,
   sanitizeContentWithThink,
-  printUsageTokens
+  formatUsageTokens
 };