grinev · grinev · Mar 28, 2026 · Mar 26, 2026 · Mar 28, 2026 · Mar 28, 2026
diff --git a/PRODUCT.md b/PRODUCT.md
@@ -156,7 +156,7 @@ Open tasks for upcoming iterations:
 - [ ] `/messages` command: browse session messages with fork/revert actions
 - [ ] `/skills` command: browse skills and choose one for usage
 - [ ] `/mcps` command: browse available MCP servers
-- [ ] Dynamic subagent activity display during task execution
+- [x] Dynamic subagent activity display during task execution
 - [ ] Git tree support
 - [ ] Docker runtime support and deployment guide
 - [ ] OpenCode server monitoring with automatic restart on stop/crash

diff --git a/README.md b/README.md
@@ -28,6 +28,7 @@ Languages: English (`en`), Deutsch (`de`), Español (`es`), Français (`fr`), Р
 - **Live status** — pinned message with current project, model, context usage, and changed files list, updated in real time
 - **Model switching** — pick models from OpenCode favorites and recent history directly in the chat (favorites are shown first)
 - **Agent modes** — switch between Plan and Build modes on the fly
+- **Subagent activity** — watch live subagent progress in chat, including the current task, agent, model, and active tool step
 - **Custom Commands** — run OpenCode custom commands (and built-ins like `init`/`review`) from an inline menu with confirmation
 - **Interactive Q&A** — answer agent questions and approve permissions via inline buttons
 - **Voice prompts** — send voice/audio messages, transcribe them via a Whisper-compatible API, then forward recognized text to OpenCode

diff --git a/src/bot/index.ts b/src/bot/index.ts
@@ -54,6 +54,7 @@ import {
   formatToolInfo,
   getAssistantParseMode,
 } from "../summary/formatter.js";
+import { renderSubagentCards } from "../summary/subagent-formatter.js";
 import { ToolMessageBatcher } from "../summary/tool-message-batcher.js";
 import { getCurrentSession } from "../session/manager.js";
 import { ingestSessionInfoForCache } from "../session/cache-manager.js";
@@ -89,6 +90,7 @@ const TELEGRAM_DOCUMENT_CAPTION_MAX_LENGTH = 1024;
 const RESPONSE_STREAM_THROTTLE_MS = 200;
 const RESPONSE_STREAM_TEXT_LIMIT = 3800;
 const SESSION_RETRY_PREFIX = "🔁";
+const SUBAGENT_STREAM_PREFIX = "🧩";
 const __filename = fileURLToPath(import.meta.url);
 const __dirname = path.dirname(__filename);
 const TEMP_DIR = path.join(__dirname, "..", ".tmp");
@@ -401,7 +403,7 @@ async function ensureEventSubscription(directory: string): Promise<void> {
     const chatId = chatIdInstance;
 
     try {
-      const streamedViaMessages = await finalizeAssistantResponse({
+      await finalizeAssistantResponse({
         responseStreaming: config.bot.responseStreaming,
         sessionId,
         messageId,
@@ -425,16 +427,16 @@ async function ensureEventSubscription(directory: string): Promise<void> {
             format,
           });
         },
+        deleteMessages: async (messageIds) => {
+          for (const msgId of messageIds) {
+            try {
+              await botApi.deleteMessage(chatId, msgId);
+            } catch (err) {
+              logger.warn(`[Bot] Failed to delete streamed message ${msgId}:`, err);
+            }
+          }
+        },
       });
-
-      if (streamedViaMessages) {
-        logger.debug(
-          `[Bot] Final assistant message already streamed (session=${sessionId}, message=${messageId})`,
-        );
-        foregroundSessionState.markIdle(sessionId);
-        await scheduledTaskRuntime.flushDeferredDeliveries();
-        return;
-      }
     } catch (err) {
       logger.error("Failed to send message to Telegram:", err);
       // Stop processing events after critical error to prevent infinite loop
@@ -461,7 +463,11 @@ async function ensureEventSubscription(directory: string): Promise<void> {
       toolInfo.hasFileAttachment &&
       (toolInfo.tool === "write" || toolInfo.tool === "edit" || toolInfo.tool === "apply_patch");
 
-    if (config.bot.hideToolCallMessages || shouldIncludeToolInfoInFileCaption) {
+    if (
+      config.bot.hideToolCallMessages ||
+      shouldIncludeToolInfoInFileCaption ||
+      toolInfo.tool === "task"
+    ) {
       return;
     }
 
@@ -475,6 +481,32 @@ async function ensureEventSubscription(directory: string): Promise<void> {
     }
   });
 
+  summaryAggregator.setOnSubagent(async (sessionId, subagents) => {
+    if (!botInstance || !chatIdInstance) {
+      return;
+    }
+
+    if (config.bot.hideToolCallMessages) {
+      return;
+    }
+
+    const currentSession = getCurrentSession();
+    if (!currentSession || currentSession.id !== sessionId) {
+      return;
+    }
+
+    try {
+      const renderedCards = await renderSubagentCards(subagents);
+      if (!renderedCards) {
+        return;
+      }
+
+      toolCallStreamer.replaceByPrefix(sessionId, SUBAGENT_STREAM_PREFIX, renderedCards);
+    } catch (err) {
+      logger.error("Failed to render subagent activity for Telegram:", err);
+    }
+  });
+
   summaryAggregator.setOnToolFile(async (fileInfo) => {
     if (!botInstance || !chatIdInstance) {
       logger.error("Bot or chat ID not available for sending file");
@@ -582,25 +614,45 @@ async function ensureEventSubscription(directory: string): Promise<void> {
       responseStreaming: config.bot.responseStreaming,
       hideThinkingMessages: config.bot.hideThinkingMessages,
     });
+
+    // Refresh pinned message so it shows the latest in-memory context
+    // (accumulated from silent token updates). 1 API call per thinking event.
+    if (pinnedMessageManager.isInitialized()) {
+      await pinnedMessageManager.refresh();
+    }
   });
 
-  summaryAggregator.setOnTokens(async (tokens) => {
+  summaryAggregator.setOnTokens(async (tokens, isCompleted) => {
     if (!pinnedMessageManager.isInitialized()) {
       return;
     }
 
     try {
-      logger.debug(`[Bot] Received tokens: input=${tokens.input}, output=${tokens.output}`);
+      logger.debug(
+        `[Bot] Received tokens: input=${tokens.input}, output=${tokens.output}, completed=${isCompleted}`,
+      );
 
-      // Update keyboardManager SYNCHRONOUSLY before any await
-      // This ensures keyboard has correct context when onComplete sends the reply
       const contextSize = tokens.input + tokens.cacheRead;
       const contextLimit = pinnedMessageManager.getContextLimit();
+
+      // Skip non-completed messages with zero context: a new assistant message
+      // starts with tokens={input:0, ...} which would overwrite valid context
+      // from the previous step. Only accept zeros from completed messages.
+      if (!isCompleted && contextSize === 0) {
+        logger.debug("[Bot] Skipping zero-token intermediate update");
+        return;
+      }
+
+      // Update both keyboard and pinned state in memory (keeps them in sync)
       if (contextLimit > 0) {
         keyboardManager.updateContext(contextSize, contextLimit);
       }
+      pinnedMessageManager.updateTokensSilent(tokens);
 
-      await pinnedMessageManager.onMessageComplete(tokens);
+      // Full pinned message update (API call) only on completed messages
+      if (isCompleted) {
+        await pinnedMessageManager.onMessageComplete(tokens);
+      }
     } catch (err) {
       logger.error("[Bot] Error updating pinned message with tokens:", err);
     }

diff --git a/src/bot/streaming/response-streamer.ts b/src/bot/streaming/response-streamer.ts
@@ -16,6 +16,11 @@ export interface StreamingMessagePayload {
   editOptions?: TelegramEditMessageOptions;
 }
 
+export interface StreamCompleteResult {
+  streamed: boolean;
+  telegramMessageIds: number[];
+}
+
 interface ResponseStreamerCompleteOptions {
   flushFinal?: boolean;
 }
@@ -142,10 +147,12 @@ export class ResponseStreamer {
     messageId: string,
     payload?: StreamingMessagePayload,
     options?: ResponseStreamerCompleteOptions,
-  ): Promise<boolean> {
+  ): Promise<StreamCompleteResult> {
+    const notStreamed: StreamCompleteResult = { streamed: false, telegramMessageIds: [] };
+
     const state = this.states.get(buildStateKey(sessionId, messageId));
     if (!state) {
-      return false;
+      return notStreamed;
     }
 
     if (payload) {
@@ -163,13 +170,13 @@ export class ResponseStreamer {
       await this.cleanupBrokenStream(state, "complete_broken_stream");
       this.cancelState(state);
       this.states.delete(state.key);
-      return false;
+      return notStreamed;
     }
 
     if (state.telegramMessageIds.length === 0) {
       this.cancelState(state);
       this.states.delete(state.key);
-      return false;
+      return notStreamed;
     }
 
     let synced = true;
@@ -180,9 +187,10 @@ export class ResponseStreamer {
       }
     }
 
+    const messageIds = [...state.telegramMessageIds];
     this.cancelState(state);
     this.states.delete(state.key);
-    return synced;
+    return { streamed: synced, telegramMessageIds: messageIds };
   }
 
   clearMessage(sessionId: string, messageId: string, reason: string): void {

diff --git a/src/bot/utils/finalize-assistant-response.ts b/src/bot/utils/finalize-assistant-response.ts
@@ -1,5 +1,6 @@
 import type { StreamingMessagePayload, ResponseStreamer } from "../streaming/response-streamer.js";
 import type { TelegramTextFormat } from "./telegram-text.js";
+import { logger } from "../../utils/logger.js";
 
 interface FinalizeAssistantResponseOptions {
   responseStreaming: boolean;
@@ -17,6 +18,7 @@ interface FinalizeAssistantResponseOptions {
     options: { reply_markup: unknown } | undefined,
     format: TelegramTextFormat,
   ) => Promise<void>;
+  deleteMessages: (messageIds: number[]) => Promise<void>;
 }
 
 export async function finalizeAssistantResponse({
@@ -31,8 +33,9 @@ export async function finalizeAssistantResponse({
   resolveFormat,
   getReplyKeyboard,
   sendText,
+  deleteMessages,
 }: FinalizeAssistantResponseOptions): Promise<boolean> {
-  let streamedViaMessages = false;
+  let streamedMessageIds: number[] = [];
 
   if (responseStreaming) {
     const preparedStreamPayload = prepareStreamingPayload(messageText);
@@ -41,17 +44,27 @@ export async function finalizeAssistantResponse({
       preparedStreamPayload.editOptions = undefined;
     }
 
-    streamedViaMessages = await responseStreamer.complete(
+    const result = await responseStreamer.complete(
       sessionId,
       messageId,
       preparedStreamPayload ?? undefined,
     );
+
+    if (result.streamed) {
+      streamedMessageIds = result.telegramMessageIds;
+    }
   }
 
   await flushPendingServiceMessages();
 
-  if (streamedViaMessages) {
-    return true;
+  // When the response was streamed, delete the streamed messages and re-send
+  // via the non-streamed path so the reply keyboard carries the latest context.
+  if (streamedMessageIds.length > 0) {
+    try {
+      await deleteMessages(streamedMessageIds);
+    } catch (err) {
+      logger.warn("[FinalizeResponse] Failed to delete streamed messages, sending with keyboard anyway:", err);
+    }
   }
 
   const parts = formatSummary(messageText);

diff --git a/src/i18n/de.ts b/src/i18n/de.ts
@@ -292,6 +292,18 @@ export const de: I18nDictionary = {
   "pinned.line.model": "Modell: {model}",
   "pinned.line.context": "Kontext: {used} / {limit} ({percent}%)",
   "pinned.line.cost": "Kosten: {cost} ausgegeben",
+  "subagent.header": "Subagent {agent}: {description}",
+  "subagent.line.status": "Status: {status}",
+  "subagent.line.task": "Aufgabe: {task}",
+  "subagent.line.agent": "Agent: {agent}",
+  "subagent.working": "Arbeitet...",
+  "subagent.working_with_details": "Arbeitet: {details}",
+  "subagent.completed": "Abgeschlossen",
+  "subagent.failed": "Aufgabe fehlgeschlagen",
+  "subagent.status.pending": "ausstehend",
+  "subagent.status.running": "laeuft",
+  "subagent.status.completed": "abgeschlossen",
+  "subagent.status.error": "Fehler",
   "pinned.files.title": "Dateien ({count}):",
   "pinned.files.item": "  {path}{diff}",
   "pinned.files.more": "  ... und {count} mehr",

diff --git a/src/i18n/en.ts b/src/i18n/en.ts
@@ -280,6 +280,18 @@ export const en = {
   "pinned.line.model": "Model: {model}",
   "pinned.line.context": "Context: {used} / {limit} ({percent}%)",
   "pinned.line.cost": "Cost: {cost} spent",
+  "subagent.header": "Subagent {agent}: {description}",
+  "subagent.line.status": "Status: {status}",
+  "subagent.line.task": "Task: {task}",
+  "subagent.line.agent": "Agent: {agent}",
+  "subagent.working": "Working...",
+  "subagent.working_with_details": "Working: {details}",
+  "subagent.completed": "Completed",
+  "subagent.failed": "Task failed",
+  "subagent.status.pending": "pending",
+  "subagent.status.running": "running",
+  "subagent.status.completed": "completed",
+  "subagent.status.error": "error",
   "pinned.files.title": "Files ({count}):",
   "pinned.files.item": "  {path}{diff}",
   "pinned.files.more": "  ... and {count} more",

diff --git a/src/i18n/es.ts b/src/i18n/es.ts
@@ -290,6 +290,18 @@ export const es: I18nDictionary = {
   "pinned.line.model": "Modelo: {model}",
   "pinned.line.context": "Contexto: {used} / {limit} ({percent}%)",
   "pinned.line.cost": "Costo: {cost} gastado",
+  "subagent.header": "Subagente {agent}: {description}",
+  "subagent.line.status": "Estado: {status}",
+  "subagent.line.task": "Tarea: {task}",
+  "subagent.line.agent": "Agente: {agent}",
+  "subagent.working": "Trabajando...",
+  "subagent.working_with_details": "Trabajando: {details}",
+  "subagent.completed": "Completada",
+  "subagent.failed": "Error de tarea",
+  "subagent.status.pending": "pendiente",
+  "subagent.status.running": "en ejecucion",
+  "subagent.status.completed": "completado",
+  "subagent.status.error": "error",
   "pinned.files.title": "Archivos ({count}):",
   "pinned.files.item": "  {path}{diff}",
   "pinned.files.more": "  ... y {count} más",

diff --git a/src/i18n/fr.ts b/src/i18n/fr.ts
@@ -292,6 +292,18 @@ export const fr: I18nDictionary = {
   "pinned.line.model": "Modèle : {model}",
   "pinned.line.context": "Contexte : {used} / {limit} ({percent}%)",
   "pinned.line.cost": "Coût : {cost} dépensé",
+  "subagent.header": "Sous-agent {agent} : {description}",
+  "subagent.line.status": "Statut : {status}",
+  "subagent.line.task": "Tache : {task}",
+  "subagent.line.agent": "Agent : {agent}",
+  "subagent.working": "En cours...",
+  "subagent.working_with_details": "En cours : {details}",
+  "subagent.completed": "Terminee",
+  "subagent.failed": "Echec de la tache",
+  "subagent.status.pending": "en attente",
+  "subagent.status.running": "en cours",
+  "subagent.status.completed": "termine",
+  "subagent.status.error": "erreur",
   "pinned.files.title": "Fichiers ({count}) :",
   "pinned.files.item": "  {path}{diff}",
   "pinned.files.more": "  ... et encore {count}",

diff --git a/src/i18n/ru.ts b/src/i18n/ru.ts
@@ -281,6 +281,18 @@ export const ru: I18nDictionary = {
   "pinned.line.model": "Модель: {model}",
   "pinned.line.context": "Контекст: {used} / {limit} ({percent}%)",
   "pinned.line.cost": "Стоимость: {cost} потрачено",
+  "subagent.header": "Сабагент {agent}: {description}",
+  "subagent.line.status": "Статус: {status}",
+  "subagent.line.task": "Задача: {task}",
+  "subagent.line.agent": "Агент: {agent}",
+  "subagent.working": "В работе...",
+  "subagent.working_with_details": "В работе: {details}",
+  "subagent.completed": "Завершена",
+  "subagent.failed": "Ошибка задачи",
+  "subagent.status.pending": "ожидание",
+  "subagent.status.running": "в работе",
+  "subagent.status.completed": "завершен",
+  "subagent.status.error": "ошибка",
   "pinned.files.title": "Файлы ({count}):",
   "pinned.files.item": "  {path}{diff}",
   "pinned.files.more": "  ... и еще {count}",

diff --git a/src/i18n/zh.ts b/src/i18n/zh.ts
@@ -254,6 +254,18 @@ export const zh: I18nDictionary = {
   "pinned.line.model": "模型: {model}",
   "pinned.line.context": "上下文: {used} / {limit} ({percent}%)",
   "pinned.line.cost": "费用: {cost}",
+  "subagent.header": "子代理 {agent}: {description}",
+  "subagent.line.status": "状态: {status}",
+  "subagent.line.task": "任务: {task}",
+  "subagent.line.agent": "代理: {agent}",
+  "subagent.working": "执行中...",
+  "subagent.working_with_details": "执行中: {details}",
+  "subagent.completed": "已完成",
+  "subagent.failed": "任务失败",
+  "subagent.status.pending": "等待中",
+  "subagent.status.running": "运行中",
+  "subagent.status.completed": "已完成",
+  "subagent.status.error": "错误",
   "pinned.files.title": "文件（{count}）：",
   "pinned.files.item": "  {path}{diff}",
   "pinned.files.more": "  ... 还有 {count} 个",