feat(compress): allow forced /compress to always run and repair broken tool_call context; add compress_status UI feedback

Author: JerryliDe

packages/core/src/services/compressionService.ts

@@ -571,6 +571,9 @@ export class CompressionService {
    * @param originalTokenCount 原始token数量（可选，如果提供则跳过重复计算）
    * @param overridePreserveRatio 可选的覆盖保留比例（0-1），用于激进压缩
    * @param isModelSwitchCompression 是否是模型切换时的压缩（默认false）
+   * @param force 是否为用户主动触发的强制压缩（默认false）。强制压缩时放宽
+   *              "历史过少"与"找不到边界"两个守卫，只要有内容就尽力压缩，
+   *              并允许压缩全部历史（用于修复残缺的 tool_call 上下文）
    * @returns 压缩结果
    */
   async compressHistory(
@@ -583,7 +586,8 @@ export class CompressionService {
     abortSignal: AbortSignal,
     originalTokenCount?: number,
     overridePreserveRatio?: number,
-    isModelSwitchCompression: boolean = false
+    isModelSwitchCompression: boolean = false,
+    force: boolean = false
   ): Promise<CompressionResult> {
     try {
       // 获取或计算原始token数量
@@ -606,12 +610,20 @@ export class CompressionService {
       const conversationHistory = history.slice(this.skipEnvironmentMessages);
 
       // 如果对话历史太少，不进行压缩
-      if (conversationHistory.length <= 2) {
+      // 强制压缩（用户主动 /compress）时放宽此限制：只要有内容就允许压缩，
+      // 这样用户可以随时通过压缩来修复残缺的 tool_call 上下文。
+      if (!force && conversationHistory.length <= 2) {
         return {
           success: false,
           error: 'Insufficient conversation history to compress'
         };
       }
+      if (force && conversationHistory.length === 0) {
+        return {
+          success: false,
+          error: 'No conversation history to compress'
+        };
+      }
 
       // 确定保留比例：优先使用 override，否则使用配置默认值
       const preserveRatio = overridePreserveRatio ?? this.compressionPreserveThreshold;
@@ -627,7 +639,7 @@ export class CompressionService {
       // 寻找最近的完整工具调用对边界，统一处理主agent和subAgent场景
       compressBeforeIndex = this.findToolCallBoundary(conversationHistory, compressBeforeIndex);
 
-      // 如果没有找到合适的压缩边界，不进行压缩
+      // 如果没有找到合适的压缩边界
       if (compressBeforeIndex === -1) {
         console.warn(`[compressHistory] Could not find suitable compression boundary. Conversation history structure may prevent compression.`);
         console.log(`[compressHistory] Last 5 messages in conversationHistory:`);
@@ -641,10 +653,19 @@ export class CompressionService {
           }).join(',') || 'empty';
           console.log(`  [${i}] role=${msg.role}, parts=[${partTypes}]`);
         }
-        return {
-          success: false,
-          error: 'Could not find suitable compression boundary'
-        };
+
+        if (force) {
+          // 用户主动压缩：即使找不到"干净"的边界也尽力而为，
+          // 退化为压缩全部对话历史（保留 0 条）。validateAndCleanHistory
+          // 会在拼接后清理孤立的 tool_use/tool_result，从而修复残缺上下文。
+          console.warn(`[compressHistory] Force compression: no clean boundary found, compressing entire conversation history.`);
+          compressBeforeIndex = conversationHistory.length;
+        } else {
+          return {
+            success: false,
+            error: 'Could not find suitable compression boundary'
+          };
+        }
       }
 
       const historyToCompress = conversationHistory.slice(0, compressBeforeIndex);
@@ -1019,7 +1040,10 @@ IMPORTANT POST-COMPRESSION RULES:
           geminiClient,
           prompt_id,
           abortSignal,
-          shouldCompressResult.tokenCount
+          shouldCompressResult.tokenCount,
+          undefined,
+          false,
+          force
         );
 
         // 如果压缩失败且没有明确的跳过原因，抛出错误以触发重试

packages/vscode-ui-plugin/src/extension.ts

@@ -3068,6 +3068,12 @@ function setupSlashCommandHandlers() {
   // 通过现有 system-notification / chat_compressed 通道反馈给 webview。
   // ─────────────────────────────────────────────────────────────────
   communicationService.addMessageHandler('builtin_compress', async () => {
+    // 本次压缩对应的 session（用于 webview 把状态消息归位到正确的会话）。
+    const compressSessionId = sessionManager.getCurrentSession()?.info.id || null;
+    // 给本次压缩分配一个稳定 statusId，让 webview 能把 start → done/error 的
+    // 多条 compress_status 更新到同一条 in-chat 通知上（而不是不断追加新条目）。
+    const statusId = `compress-${Date.now()}`;
+
     try {
       // 当前会话的 AIService —— 与 setupChatHandlers 内 send_message 的
       // 取法（getCurrentInitializedAIService）保持一致。
@@ -3090,6 +3096,14 @@ function setupSlashCommandHandlers() {
         return;
       }
 
+      // ── 阶段 1：start ──────────────────────────────────────────────
+      // 立即把"正在压缩"状态推给 webview，让它在对话流里插入一条持久的
+      // in-chat 通知 + 底部进度条，用户不再面对一片空白干等。
+      communicationService.sendMessage({
+        type: 'compress_status',
+        payload: { phase: 'start', statusId, sessionId: compressSessionId },
+      });
+
       logger.info('[/compress] Manual compression triggered via slash command');
       const promptId = `slash-compress-${Date.now()}`;
       const result = await geminiClient.tryCompressChat(promptId, new AbortController().signal, true);
@@ -3099,6 +3113,20 @@ function setupSlashCommandHandlers() {
         const after = (result as any).newTokenCount;
         logger.info(`[/compress] Compressed ${before} → ${after} tokens`);
         const fmt = (n: any) => (typeof n === 'number' ? n.toLocaleString() : String(n));
+
+        // ── 阶段 2：done ─────────────────────────────────────────────
+        // 把压缩结果（前后 token 数）回写到同一条 in-chat 通知上。
+        communicationService.sendMessage({
+          type: 'compress_status',
+          payload: {
+            phase: 'done',
+            statusId,
+            sessionId: compressSessionId,
+            originalTokenCount: before,
+            newTokenCount: after,
+          },
+        });
+
         communicationService.sendMessage({
           type: 'slash_command_result',
           payload: {
@@ -3108,13 +3136,29 @@ function setupSlashCommandHandlers() {
         });
       } else {
         logger.warn('[/compress] tryCompressChat returned falsy result');
+        // ── 阶段 2'：skipped ─────────────────────────────────────────
+        // 历史本来就不大、无需压缩：也要给一个明确的结束态，避免进度条悬挂。
+        communicationService.sendMessage({
+          type: 'compress_status',
+          payload: { phase: 'skipped', statusId, sessionId: compressSessionId },
+        });
         communicationService.sendMessage({
           type: 'slash_command_result',
           payload: { success: false, error: 'Compression returned no result. The history may already be small enough.' },
         });
       }
     } catch (error) {
       logger.error('[/compress] Compression failed', error instanceof Error ? error : undefined);
+      // ── 阶段 2''：error ───────────────────────────────────────────
+      communicationService.sendMessage({
+        type: 'compress_status',
+        payload: {
+          phase: 'error',
+          statusId,
+          sessionId: compressSessionId,
+          error: error instanceof Error ? error.message : String(error),
+        },
+      });
       communicationService.sendMessage({
         type: 'slash_command_result',
         payload: {

packages/vscode-ui-plugin/src/types/messages.ts

@@ -510,6 +510,19 @@ export type ExtensionToWebViewMessage =
       info?: string;
       error?: string;
     } }
+  // 🎯 手动 /compress 的状态推送（让 webview 在对话流里展示"正在压缩 / 压缩结果"）
+  | { type: 'compress_status'; payload: {
+      /** start = 开始压缩；done = 完成；skipped = 历史太小无需压缩；error = 失败 */
+      phase: 'start' | 'done' | 'skipped' | 'error';
+      /** 同一次压缩的稳定 id，让 start → done/error 落到同一条 in-chat 通知 */
+      statusId: string;
+      sessionId: string | null;
+      /** done 时携带：压缩前后的 token 数 */
+      originalTokenCount?: number;
+      newTokenCount?: number;
+      /** error 时携带：错误信息 */
+      error?: string;
+    } }
   // 🎯 模型切换压缩确认
   | { type: 'compression_confirmation_request'; payload: { requestId: string; sessionId: string; targetModel: string; currentTokens: number; targetTokenLimit: number; compressionThreshold: number; message: string } }
   // 🎯 Token使用情况更新（压缩后）

packages/vscode-ui-plugin/webview/src/components/MessageInput.tsx

@@ -970,30 +970,26 @@ export const MessageInput = React.forwardRef<MessageInputHandle, MessageInputPro
             if (result.success && result.sideEffect === 'compress') {
               // ─────────────────────────────────────────────────────
               // 内置 /compress：不发 AI，转发副作用消息让 backend 调
-              // tryCompressChat。结果通过 show_notification 显示。
+              // tryCompressChat。
+              //
+              // 状态反馈走 backend 主动推送的 `compress_status`（start →
+              // done/error/skipped），由 MultiSessionApp 在对话流里插入并
+              // 原地更新一条持久的 in-chat 通知（带 spinner + 结果），
+              // 用户不再面对空白干等。这里**不再**弹 info/success toast，
+              // 避免与 in-chat 通知重复。
+              //
+              // 仅保留：早退失败的兜底 toast —— 当 geminiClient 尚未就绪
+              // 或已有压缩在进行时，backend 会直接返回 error 且**不会**发出
+              // compress_status:start（因此没有 in-chat 通知可承载该错误）。
               // ─────────────────────────────────────────────────────
               consumedBySlashCommand = true;
               console.log(`🎯 [SlashCommand] /${commandName} → side effect: compress`);
 
-              if (window.vscode && result.info) {
-                window.vscode.postMessage({
-                  type: 'show_notification',
-                  payload: { message: result.info, type: 'info' },
-                });
-              }
-
-              // 异步触发 backend 真正的压缩，结果再通知
+              // 异步触发 backend 真正的压缩。成功/进行中状态由 compress_status
+              // 驱动的 in-chat 通知体现；这里只在「压缩根本没启动」的早退失败时兜底提示。
               slashCommandHandler.triggerBuiltinCompress().then((compressResult) => {
                 if (!window.vscode) return;
-                if (compressResult.success) {
-                  window.vscode.postMessage({
-                    type: 'show_notification',
-                    payload: {
-                      message: compressResult.info || 'Compression completed.',
-                      type: 'info',
-                    },
-                  });
-                } else {
+                if (!compressResult.success) {
                   window.vscode.postMessage({
                     type: 'show_notification',
                     payload: {

packages/vscode-ui-plugin/webview/src/components/MultiSessionApp.tsx

@@ -923,6 +923,105 @@ export const MultiSessionApp: React.FC = () => {
       setStreamRecoveryVisible(false);
     }));
 
+    // 🎯 监听手动 /compress 的状态推送（start → done/error/skipped）。
+    // 目标：用户敲 /compress 后不再面对空白干等 —— 在对话流里插入一条
+    // 持久的 in-chat 通知（带 spinner），压缩结束后原地更新为最终结果，
+    // 同时复用底部 compression-progress-bar 作为全局进度提示。
+    cleanups.push(messageService.onExtensionMessage('compress_status', (payload: any) => {
+      console.log('🗜️ [MultiSessionApp] Received compress_status:', payload);
+      const targetSessionId: string | undefined =
+        payload?.sessionId || stateRef.current.currentSessionId || undefined;
+      if (!targetSessionId) {
+        console.warn('🗜️ [MultiSessionApp] compress_status without resolvable sessionId, ignoring');
+        return;
+      }
+      // 用 statusId 派生稳定的消息 id，保证 start/done/error 落到同一条通知。
+      const statusId: string = payload?.statusId || `compress-${Date.now()}`;
+      const notificationId = `notif-${statusId}`;
+
+      const fmt = (n: any) => (typeof n === 'number' ? n.toLocaleString() : String(n));
+
+      if (payload?.phase === 'start') {
+        // 底部进度条
+        setIsCompressing(true);
+        // in-chat 持久通知（进行中）
+        addMessage(targetSessionId, {
+          id: notificationId,
+          type: 'notification',
+          content: createTextMessageContent(''),
+          timestamp: Date.now(),
+          notificationType: 'compression',
+          notificationTitle: t('compression.manualTitle', {}, 'Context Compression'),
+          notificationDescription: t(
+            'compression.manualInProgressDesc',
+            {},
+            'Summarizing older messages while preserving recent context. This may take a moment.'
+          ),
+          severity: 'info',
+          notificationInProgress: true,
+          statusId,
+        } as any);
+        return;
+      }
+
+      // 任何结束态都要收起底部进度条
+      setIsCompressing(false);
+
+      if (payload?.phase === 'done') {
+        const original = Number(payload?.originalTokenCount);
+        const compressed = Number(payload?.newTokenCount);
+        const saved =
+          Number.isFinite(original) && Number.isFinite(compressed)
+            ? Math.max(0, original - compressed)
+            : undefined;
+        const percent =
+          Number.isFinite(original) && original > 0 && saved !== undefined
+            ? `-${Math.round((saved / original) * 100)}%`
+            : '';
+        updateMessage(targetSessionId, notificationId, {
+          notificationInProgress: false,
+          notificationTitle: t('compression.manualDone', {}, 'Context compressed'),
+          notificationDescription: t(
+            'compression.manualDoneDesc',
+            {
+              original: fmt(original),
+              compressed: fmt(compressed),
+              saved: fmt(saved),
+              percent,
+            },
+            `Reduced from ${fmt(original)} to ${fmt(compressed)} tokens.`
+          ),
+          severity: 'info',
+        } as any);
+        return;
+      }
+
+      if (payload?.phase === 'skipped') {
+        updateMessage(targetSessionId, notificationId, {
+          notificationInProgress: false,
+          notificationTitle: t('compression.manualSkipped', {}, 'Compression skipped'),
+          notificationDescription: t(
+            'compression.manualSkippedDesc',
+            {},
+            'The conversation history is already small enough — nothing to compress.'
+          ),
+          severity: 'info',
+        } as any);
+        return;
+      }
+
+      if (payload?.phase === 'error') {
+        updateMessage(targetSessionId, notificationId, {
+          notificationInProgress: false,
+          notificationType: 'warning',
+          notificationTitle: t('compression.manualFailed', {}, 'Compression failed'),
+          notificationDescription: String(payload?.error || 'Compression failed.'),
+          severity: 'error',
+        } as any);
+        return;
+      }
+    }));
+
     // 🔐 监听认证过期通知（服务端返回 HTTP 401 时由 extension 主动推送）
     cleanups.push(messageService.onAuthExpired(({ reason }) => {
       console.log('🔐 [MultiSessionApp] Auth expired notification received:', reason);

packages/vscode-ui-plugin/webview/src/components/SystemNotificationMessage.css

@@ -51,6 +51,38 @@
   flex-shrink: 0;
 }
 
+/* 进行中状态的内联 spinner（替代图标，体现"正在处理"） */
+.notification-spinner {
+  width: 12px;
+  height: 12px;
+  margin-right: 6px;
+  flex-shrink: 0;
+  border: 2px solid var(--vscode-editorInfo-foreground, #469af8);
+  border-top-color: transparent;
+  border-radius: 50%;
+  animation: notification-spin 0.8s linear infinite;
+}
+
+@keyframes notification-spin {
+  to {
+    transform: rotate(360deg);
+  }
+}
+
+/* 进行中通知整体轻微脉动，强化"还在进行"的感知 */
+.system-notification.notification-in-progress {
+  animation: notification-pulse 1.6s ease-in-out infinite;
+}
+
+@keyframes notification-pulse {
+  0%, 100% {
+    opacity: 1;
+  }
+  50% {
+    opacity: 0.7;
+  }
+}
+
 .notification-title {
   flex: 1;
 }

packages/vscode-ui-plugin/webview/src/components/SystemNotificationMessage.tsx

@@ -42,10 +42,14 @@ export const SystemNotificationMessage: React.FC<SystemNotificationMessageProps>
   };
 
   return (
-    <div className={`system-notification ${getNotificationStyle()}`}>
-      {/* 通知头部 - 图标 + 标题 */}
+    <div className={`system-notification ${getNotificationStyle()}${message.notificationInProgress ? ' notification-in-progress' : ''}`}>
+      {/* 通知头部 - 图标/spinner + 标题 */}
       <div className="notification-header">
-        <span className="notification-icon" aria-hidden="true">{getNotificationIcon()}</span>
+        {message.notificationInProgress ? (
+          <span className="notification-spinner" aria-label="loading" role="status" />
+        ) : (
+          <span className="notification-icon" aria-hidden="true">{getNotificationIcon()}</span>
+        )}
         <span className="notification-title">{message.notificationTitle}</span>
       </div>