diff --git a/README.md b/README.md
index 6c5c6689..152e6625 100644
--- a/README.md
+++ b/README.md
@@ -169,6 +169,10 @@ DMs do not require an @ mention. Groups and topic groups require `@bot` by defau
 - **Tool-call display**: controls whether tool blocks appear in the final card / markdown reply.
 - **COT process message**: `off` sends only the final reply; `brief` first sends a COT message with agent progress text and tool summaries; `detailed` also includes tool args and truncated output.
 
+Fresh Codex profiles default to final-only delivery: plain text, hidden tool
+calls, and COT off. Existing profiles with explicit display preferences keep
+their choices. Claude keeps the streaming message-card default.
+
 When COT is enabled, the bridge splits the process view and final answer into two messages. The COT message is for tracing what the agent did; the final answer is still generated from the agent's raw text, without heuristic bridge-side filtering. If an agent emits final-answer text as ordinary stream text, that text can also appear in the COT process message.
 
 ## lark-cli identity policy
diff --git a/README.zh.md b/README.zh.md
index fae27156..fc263158 100644
--- a/README.zh.md
+++ b/README.zh.md
@@ -169,6 +169,8 @@ lark-channel-bridge profile export <name> --include-secrets --yes
 - **工具调用显示**：控制最终回复卡片 / markdown 中是否展示工具块。
 - **COT 过程消息**：`关闭` 只发送最终回复；`简略` 先用 COT 消息展示 agent 的过程文本和工具摘要；`详细` 还会展示工具参数和截断后的输出。
 
+新建 Codex profile 默认只交付最终结果：纯文本一次性发送、隐藏工具调用、关闭 COT。已有 profile 如果明确设置过展示偏好，会继续保留原选择；Claude 仍默认使用流式消息卡片。
+
 开启 COT 后，bridge 会把过程消息和最终答案拆成两条消息。过程消息用于追踪 agent 做了什么；最终答案仍由 agent 原始文本生成，bridge 不做启发式过滤。若 agent 把最终答案也作为普通流式文本输出，COT 过程消息中可能会出现对应片段。
 
 ## lark-cli 身份策略
diff --git a/src/card/config-card.ts b/src/card/config-card.ts
index a20b8af9..d84c4713 100644
--- a/src/card/config-card.ts
+++ b/src/card/config-card.ts
@@ -174,7 +174,7 @@ export function configFormCard(opts: ConfigFormOpts): object {
               initial_option: opts.messageReply === 'card' ? 'markdown' : opts.messageReply,
               options: [
                 { text: { tag: 'plain_text', content: '纯文本' }, value: 'text' },
-                { text: { tag: 'plain_text', content: '消息卡片(默认)' }, value: 'markdown' },
+                { text: { tag: 'plain_text', content: '消息卡片' }, value: 'markdown' },
               ],
             },
             {
@@ -189,7 +189,7 @@ export function configFormCard(opts: ConfigFormOpts): object {
               name: 'show_tool_calls',
               initial_option: opts.showToolCalls ? 'show' : 'hide',
               options: [
-                { text: { tag: 'plain_text', content: '显示(默认)' }, value: 'show' },
+                { text: { tag: 'plain_text', content: '显示' }, value: 'show' },
                 { text: { tag: 'plain_text', content: '隐藏' }, value: 'hide' },
               ],
             },
diff --git a/src/config/schema.ts b/src/config/schema.ts
index ed00c178..e231d4dc 100644
--- a/src/config/schema.ts
+++ b/src/config/schema.ts
@@ -87,7 +87,7 @@ export interface AppAccess {
 }
 
 export interface AppPreferences {
-  /** Reply rendering mode for IM (group/p2p) messages. Default 'card'. */
+  /** Reply rendering mode for IM (group/p2p) messages. */
   messageReply?: MessageReplyMode;
   /**
    * Internal marker: pre-0.1.27 the value `'text'` meant "lightweight
@@ -100,8 +100,9 @@ export interface AppPreferences {
   messageReplyMigrated?: boolean;
   /**
    * Whether to render tool-call blocks (Bash / Read / Edit / ...) in the
-   * output. Default true. Turn off if you only care about Claude's final
-   * text answer and want to hide the "工具调用过程".
+   * output. Defaults to false for Codex and true for Claude. Turn off if you
+   * only care about the agent's final text answer and want to hide the
+   * "工具调用过程".
    */
   showToolCalls?: boolean;
   /**
@@ -163,6 +164,8 @@ export interface AppPreferences {
  * belong at this top level alongside them.
  */
 export interface AppConfig {
+  /** Present on v2 profile configs; absent on legacy single-profile configs. */
+  agentKind?: 'claude' | 'codex';
   accounts: {
     app: AppCredentials;
   };
@@ -200,7 +203,8 @@ export function secretKeyForApp(appId: string): string {
  * (which sets `messageReplyMigrated: true`), we map their `text` →
  * `markdown` so the behavior stays the same after upgrade.
  *
- * Default for fresh configs (no `messageReply` set) is `'markdown'`.
+ * Fresh Codex profiles default to `'text'` so verbose progress and command
+ * execution stay out of Lark. Claude and legacy configs keep `'markdown'`.
  */
 export function getMessageReplyMode(cfg: AppConfig): MessageReplyMode {
   const raw = cfg.preferences?.messageReply;
@@ -208,12 +212,14 @@ export function getMessageReplyMode(cfg: AppConfig): MessageReplyMode {
     return 'markdown';
   }
   if (raw === 'card' || raw === 'markdown' || raw === 'text') return raw;
-  return 'markdown';
+  return cfg.agentKind === 'codex' ? 'text' : 'markdown';
 }
 
 /** Resolve the show-tool-calls preference with default fallback. */
 export function getShowToolCalls(cfg: AppConfig): boolean {
-  return cfg.preferences?.showToolCalls !== false;
+  const configured = cfg.preferences?.showToolCalls;
+  if (configured !== undefined) return configured;
+  return cfg.agentKind !== 'codex';
 }
 
 export function getCotMessages(cfg: AppConfig): CotMessagesMode {
diff --git a/tests/integration/bot/claude-regression.test.ts b/tests/integration/bot/claude-regression.test.ts
index eab1edd5..df234290 100644
--- a/tests/integration/bot/claude-regression.test.ts
+++ b/tests/integration/bot/claude-regression.test.ts
@@ -1,7 +1,11 @@
 import { readFile } from 'node:fs/promises';
 import { join } from 'node:path';
 import { afterEach, describe, expect, it, vi } from 'vitest';
-import { getMessageReplyMode, getRequireMentionInGroup } from '../../../src/config/schema.js';
+import {
+  getMessageReplyMode,
+  getRequireMentionInGroup,
+  getShowToolCalls,
+} from '../../../src/config/schema.js';
 import { PendingQueue } from '../../../src/bot/pending-queue.js';
 import type { NormalizedMessage } from '@larksuite/channel';
 
@@ -31,6 +35,25 @@ describe('Claude IM regression boundaries', () => {
     expect(getMessageReplyMode(cardCfg)).toBe('card');
   });
 
+  it('defaults Codex to a final-only reply while preserving explicit display choices', () => {
+    const codexCfg = {
+      agentKind: 'codex' as const,
+      accounts: { app: { id: 'app-id', secret: 'secret', tenant: 'feishu' as const } },
+    };
+
+    expect(getMessageReplyMode(codexCfg)).toBe('text');
+    expect(getShowToolCalls(codexCfg)).toBe(false);
+    expect(
+      getMessageReplyMode({
+        ...codexCfg,
+        preferences: { messageReply: 'markdown' as const },
+      }),
+    ).toBe('markdown');
+    expect(
+      getShowToolCalls({ ...codexCfg, preferences: { showToolCalls: true } }),
+    ).toBe(true);
+  });
+
   it('queues messages that arrive while a run is active and flushes them as the next batch', () => {
     vi.useFakeTimers();
     const flushed: Array<{ scope: string; batch: NormalizedMessage[] }> = [];
diff --git a/tests/integration/bot/markdown-stream-startup-failure.test.ts b/tests/integration/bot/markdown-stream-startup-failure.test.ts
index 7769523f..70a4faaf 100644
--- a/tests/integration/bot/markdown-stream-startup-failure.test.ts
+++ b/tests/integration/bot/markdown-stream-startup-failure.test.ts
@@ -324,7 +324,12 @@ async function createHarness(options: {
     codex: {
       binaryPath: '/usr/local/bin/codex',
     },
-    ...(options.messageReply ? { preferences: { messageReply: options.messageReply } } : {}),
+    preferences: {
+      // This suite exercises stream behavior explicitly. Codex profiles now
+      // default to final-only text delivery, so opt into a streaming mode.
+      messageReply: options.messageReply ?? 'markdown',
+      messageReplyMigrated: true,
+    },
   });
   const profileConfig = {
     ...baseProfileConfig,