fix(aibox): sync compressor and snapshots on client swap via set_client (M1)

run_full_subagent replaced sub_chat.client by direct assignment after
construction, leaving ConversationCompressor (bound to the original
client in DeepSeekChat.__init__) and the api_key/name snapshots
pointing at the default model — compression requests silently kept
using the default endpoint/model while the dialogue itself used the
requested one.

DeepSeekChat gains set_client(): swaps self.client and refreshes
api_key, name, and compressor.llm_client in one place. run_full_subagent
now routes model overrides through it.

Tests: new TestM1SetClient (2 cases: full reference refresh on a real
DeepSeekChat instance, and run_full_subagent calling set_client with
the new client); full suite 295 passed.

aibox/README.md

@@ -92,7 +92,7 @@ aibox/
 │   ├── prompt_tools.py     # B 批：UpdateIdentity/UpdateSoul/SaveMemoryAndEndConversation
 │   ├── proxy_tools.py      # B 批：SetProxy/GetProxy
 │   └── misc_tools.py       # B 批：AskUserTool
-├── tests/             # 单元测试套件（293 项全过）
+├── tests/             # 单元测试套件（295 项全过）
 │   ├── conftest.py                    # sys.path 配置 + websockets 桩
 │   ├── test_basic_tools.py            # A 批工具 43 项
 │   ├── test_b_tools.py                # B 批工具 58 项
@@ -885,12 +885,13 @@ elif command == "/mycommand":
 ## 📄 更新日志
 
 ### v3.4
-- ✅ **多模型档案加固（H1/H2/H3/M2 四项修复）**：
+- ✅ **多模型档案加固（H1/H2/H3/M1/M2 五项修复）**：
   - **H1（凭据隔离）**：档案端点环境变量取不到时直接报错，**绝不回退读 `~/.cert/deepseek.key`**（该文件为 DeepSeek 专属凭据，跨厂商复用属于泄漏）；key 文件回退仅对无档案的旧 `api` 扁平节生效
   - **H2（schema 校验前移）**：档案缺 `api_url`/`model` 或类型错误时，解析期即抛出指明档案名与缺失字段的 `ValueError`——不再把档案名当模型名发出、不再延迟到 `requests InvalidURL` 或 API 400
   - **H3（--model 服务模式生效）**：`--model` 升级为进程级覆盖，WebSocket 服务模式（主对话/每个会话实例/定时任务实例）与 `--once` 均生效，此前服务模式静默忽略
   - **M2（防模型名幻觉）**：`create_subagent` 工具描述与 JSON Schema 动态携带 `models` 节可用档案名列表（默认档案标注），LLM 无从编造名称；`models` 为空时描述保持原样（零破坏）
-- ✅ **测试**：293 项全过（H1 隔离 4 项 + H2 校验 6 项 + H3 传导 6 项 + M2 提示 5 项新增）
+  - **M1（换模型同步派生引用）**：新增 `DeepSeekChat.set_client()` 统一替换 LLM 客户端——同步刷新 `compressor.llm_client` 与 `api_key`/`name` 快照；`create_subagent` 子代理换模型经此入口，压缩链不再残留默认模型
+- ✅ **测试**：295 项全过（H1 隔离 4 项 + H2 校验 6 项 + H3 传导 6 项 + M2 提示 5 项 + M1 替换 2 项新增）
 
 ### v3.3
 - ✅ **安全审计与修复**（两轮审计报告见 [docs/reports/aibox_audit_20260817.md](../docs/reports/aibox_audit_20260817.md)）：

aibox/aibox.py

@@ -651,8 +651,10 @@ def run_full_subagent(role, task, config, logger=None, model=None,
         )
 
         # 覆盖模型（如果指定了不同模型）
+        # M1：经 set_client 统一替换——同步刷新 compressor.llm_client 与
+        # api_key/name 快照，避免压缩链仍用默认模型发请求
         if model:
-            sub_chat.client = create_llm_client(config, api_key=None, logger=logger, model=model)
+            sub_chat.set_client(create_llm_client(config, api_key=None, logger=logger, model=model))
 
         # 轮次限制：防止工具调用无限循环
         # 注意：超限时不置 conversation_active=False——那会触发 think_and_respond
@@ -944,6 +946,18 @@ class DeepSeekChat:
         self.tool_registry = ToolRegistry(self.logger)
         self._register_tools(exit_callback, command_executor)
 
+    def set_client(self, client) -> None:
+        """M1：统一替换 LLM 客户端（换模型/换档案），同步刷新派生引用。
+
+        直接赋值 self.client 会让 ConversationCompressor 等持有旧 client 的
+        组件继续用默认模型发请求（压缩链与实际对话模型不一致）。本方法
+        一并更新 compressor.llm_client 与 api_key/name 快照。
+        """
+        self.client = client
+        self.api_key = client.api_key
+        self.name = client.name
+        self.compressor.llm_client = client
+    
     def cancel(self):
         """中断当前正在进行的AI思考/工具执行。线程安全，可从任意线程调用。"""
         self.logger.info("收到中断信号，正在取消当前操作...")

aibox/tests/test_models_profile.py

@@ -439,6 +439,46 @@ class TestM2SubAgentModelHint:
         assert "可用模型档案" not in tool.get_description()
 
 
+# ==================== M1：set_client 统一替换 LLM 客户端 ====================
+
+class TestM1SetClient:
+    """M1：run_full_subagent 换模型后 compressor 仍持旧 client（压缩链用默认模型）。
+
+    验证 DeepSeekChat.set_client 统一替换：client/api_key/name/compressor.llm_client
+    四处同步刷新；run_full_subagent 经 set_client 而非直接赋值。
+    """
+
+    def test_set_client_refreshes_all_references(self):
+        """set_client → client/api_key/name/compressor.llm_client 同步为新 client"""
+        from test_once_mode import _make_ai, FakeLLMClient
+        ai = _make_ai()
+        assert ai.compressor.llm_client is ai.client  # 前置：初始一致
+        new_client = FakeLLMClient()
+        new_client.api_key = "sk-new"
+        new_client.name = "NewVendor"
+        ai.set_client(new_client)
+        assert ai.client is new_client
+        assert ai.api_key == "sk-new"
+        assert ai.name == "NewVendor"
+        assert ai.compressor.llm_client is new_client
+
+    def test_run_full_subagent_uses_set_client(self):
+        """run_full_subagent(model=...) → 经 set_client 替换（compressor 不再持旧 client）"""
+        from aibox import run_full_subagent
+        fake_sub_chat = mock.MagicMock()
+        fake_sub_chat.think_and_respond = mock.MagicMock(
+            return_value="任务完成报告")
+        fake_sub_chat.stats = {}
+        fake_sub_chat.messages = []
+        fake_new_client = mock.MagicMock()
+        with mock.patch("aibox.DeepSeekChat", return_value=fake_sub_chat), \
+             mock.patch("aibox.create_llm_client", return_value=fake_new_client), \
+             mock.patch("aibox.SubprocessCommandExecutor"):
+            run_full_subagent("测试角色", "测试任务", None,
+                              model="kimi", logger=None)
+        fake_sub_chat.set_client.assert_called_once_with(fake_new_client)
+
+
 # ==================== DeepSeekChat 档案级 reasoning 读取 ====================
 
 class TestChatReasoningFlag: