feat: TUI 新增 Vim 风格键位（模态输入 + 聊天导航），版本升 8.0.3

- 新增 xcli_core/vim_keys.py：纯逻辑 Vim 状态机（NORMAL/INSERT），h/j/k/l/w/b/0/$/^/gg/G/i/a/o/I/A/O/dd/x/dw/d$/u
- tui.py：VimSendTextArea 子类接入模态编辑；XiaoliTUI 增加 j/k、Ctrl+F/B/D/U 聊天滚动、Ctrl+Space 聚焦切换、i 聚焦插入、/vim 开关、模式提示栏、/help 更新
- 新增 tests/test_vim_keys.py（18 项动作单测全过）
- 版本号 8.0.2 -> 8.0.3（constants/about/README/tui/rich_tui/lsp/mcp）

.gitignore

@@ -47,3 +47,12 @@ screenshots/gen_*.png
 # 临时文件
 *.tmp
 *.bak
+
+# v8.0 TUI 快照产物（运行时生成，供 AI 自验/预览）
+.tui_snapshots/
+
+# FC 能力探测缓存（运行时按 base_url+model 生成，机器相关）
+.fc_support.json
+
+# pytest 缓存
+.pytest_cache/

CHANGELOG.md

@@ -4,6 +4,70 @@
 
 ---
 
+## v8.0.3 (2026-08-07) — 体验增强
+
+### ✨ 新功能
+
+- **TUI 新增 Vim 风格键位**（默认开，可用 `/vim` 开关）
+  - 输入框**模态编辑**：`Esc` 在 NORMAL / INSERT 间切换；NORMAL 下
+    - 移动：`h j k l`、`w / b`（跳词）、`0 / $`（行首/行尾）、`^`（首个非空白）、`gg / G`（文首/文尾）
+    - 进入插入：`i / a / o`（插入/追加/下方新行）、`I / A / O`（行首/行尾/上方新行）
+    - 编辑：`dd`（删行）、`x`（删字）、`dw`（删词）、`d$`（删到行尾）、`u`（撤销）
+    - 普通模式下 `回车` = 下移一行（与 vim 一致）
+  - 核心状态机抽成纯模块 `xcli_core/vim_keys.py`（`VimInputState`，不依赖 textual），配 `tests/test_vim_keys.py` 单测，18 项动作全部验证通过
+  - 聊天区导航（输入框未聚焦时生效）：`j / k` 逐行滚动、`Ctrl+F / B` 翻页、`Ctrl+D / U` 半页、`Ctrl+Space` 切换焦点、`i` 聚焦并进入插入
+  - 输入框下方提示栏实时显示 `-- INSERT --` / `-- NORMAL --` 模式；`/help` 已补充 Vim 键位说明
+  - 可通过 `config.json` 的 `vim_mode: false` 默认关闭
+
+### 🛠 工程
+
+- 版本号统一更新至 v8.0.3（含 README 中英 badge、`about.txt`、`xcli_core/constants.py` 中央 `VERSION`、TUI banner、LSP/MCP `clientInfo.version`）
+
+---
+
+## v8.0.2 (2026-08-07) — 小修复版本
+
+### 🐛 Bug 修复
+
+- **加载动画在非 TTY 环境不再污染输出流**（提交 `c81914a`）
+  - `_generate_response_with_animation` 的 `show_animation` 用 `print("\r...")` 逐帧（约 10 帧/秒）输出 `love.txt` 情话 + spinner，仅真终端能原地覆盖；被聊天前端 / WebSocket 桥 / 管道 / 日志镜像捕获时每帧变成一行垃圾混进对话显示
+  - 加 `sys.stdout.isatty()` 守卫：非 TTY 只输出单行提示、不跑逐帧 `\r` 动画
+  - TUI 模式（`tui_output_callback`）本就跳过动画，不受影响
+- **非 TTY 环境保留情话「人味」**（提交 `1c52482`）
+  - 上一版修复导致聊天界面完全看不到情话，改为：非 TTY 在请求开始输出一句随机情话（`💭` 前缀，只打一次、零残帧）
+  - `love.txt` 读取时过滤 `http/https` 行，避免把反战标语 / 网址当情话甩出
+  - 真 TTY 逐帧情话 + spinner 动画、TUI 模式均不受影响
+- **修复文件管理器跨平台路径串台**（本次提交）
+  - 模型在 Windows 主机上常给出 Linux/Mac 风格绝对路径（如 `/home/user/Desktop`、`/Users/<谁>/Desktop`、`/root/Desktop`），`os.path.abspath` 在 Windows 下会错误地拼成 `C:\home\user\Desktop` 并报「路径不存在」
+  - 新增 `_resolve_cross_platform_path`：Windows 下把 `/home/<用户>/…`、`/Users/<用户>/…`、`/root/…` 统一映射到当前用户家目录对应位置（`~/Desktop` 等）；Linux/Mac 下这些路径本身合法，原样保留
+  - `file_manager` 全部 10 个操作（list/read/write/append/copy/move/delete/search/info/mkdir）的路径解析均改用该归一化函数
+
+### 🛠 工程
+
+- 版本号统一更新至 v8.0.2（含 README 中英 badge、`about.txt`、`xcli_core/constants.py` 中央 `VERSION`、TUI banner、LSP/MCP `clientInfo.version`）
+
+---
+
+## v8.0.1 (2026-08-07) — Release
+
+### ✨ 新增功能
+
+- **Operit 市场插件转译层全面落地**
+  - `plugins/operit_loader.py` 增强：`OkHttp.newClient()`、`OkHttpRequest.execute()` 改为异步、`OkHttpResponse` 新增 `json()` 与 `statusMessage()`（基于 `http.HTTPStatus`）
+  - `plugins/operit_translate.py` 新增：Operit `.ts` 插件 → xiaoli-cli `.py` 脚手架生成器（自动提取 METADATA、生成 `exports` 胶水，原 TS 作参考注释保留）
+  - 接入 5 个已全链路验证的 Operit 市场插件：
+    - `market_time` — 时间查询 / 格式化
+    - `market_duckduckgo` — 网页内容抓取与清洗
+    - `market_crossref` — Crossref 学术文献检索
+    - `market_network_test` — HTTP 连通性测试
+    - `market_reader` — 本地代码递归搜索
+
+### 🛠 工程
+
+- 版本号统一更新至 v8.0.1（含 README 中英 badge、`about.txt`、`xcli_core/constants.py` 中央 `VERSION`、TUI banner、LSP/MCP `clientInfo.version`）
+
+---
+
 ## v5.4.0 (2026-05-09) — Release
 
 ### ✨ 新增功能

LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2026 shuiyu1123
+Copyright (c) 2026 shuiyu
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

README.en.md

@@ -5,7 +5,7 @@
 </p>
 
 <p align="center">
-  <img src="https://img.shields.io/badge/version-5.4.1-blue" alt="version">
+  <img src="https://img.shields.io/badge/version-8.0.3-blue" alt="version">
   <img src="https://img.shields.io/badge/python-3.10+-green" alt="python">
   <img src="https://img.shields.io/badge/license-MIT-orange" alt="license">
   <img src="https://img.shields.io/badge/tests-113%20passed-brightgreen" alt="tests">

README.md

@@ -9,7 +9,7 @@
 </p>
 
 <p align="center">
-  <img src="https://img.shields.io/badge/version-5.4.1-blue" alt="version">
+  <img src="https://img.shields.io/badge/version-8.0.3-blue" alt="version">
   <img src="https://img.shields.io/badge/python-3.10+-green" alt="python">
   <img src="https://img.shields.io/badge/license-MIT-orange" alt="license">
   <img src="https://img.shields.io/badge/tests-150+%20passed-brightgreen" alt="tests">
@@ -22,6 +22,12 @@
 
 小狸 Pro-CLI 是一款基于 Python 的智能编程助手，集成了代码编辑、版本控制、浏览器自动化、图像生成、记忆系统等功能，支持多种 AI 引擎，帮助开发者提升工作效率。
 
+## v8.0 新特性
+
+- **Rich 面板式 TUI（默认）** — `rich` 驱动的可视化对话面板；自带**快照出口**（导出 text/svg/html），AI 开发界面时可把渲染结果读回自验，补上"看不见渲染"的反馈闭环。输入 `/tui` 进入，`/snapshot` 随时导出当前界面。
+- **Token 感知压缩** — 自动读取 OpenAI 格式返回的 `usage.prompt_tokens`，当 `已用 token / 模型上下文窗口 ≥ 0.7`（可在 `config.json` 设 `system.compress_ratio` / `system.context_window` 覆盖）即触发上下文压缩，消息数阈值作为兜底，避免超窗截断。
+- **记忆模式** — 持久化 `MEMORY.md` + 每日日记 + 自动保存聊天记录 + 关键词检索；新增 `/memory mode <companion|work>` 切换与 `/compress` 手动压缩。
+
 ## 特性
 
 - **双 AI 引擎** — Ollama 本地模型 + OpenAI 兼容格式（DeepSeek、Grok、硅基流动等）

about.txt

@@ -1,7 +1,7 @@
 **产品标识**  
 - **产品名称**：小狸智能体 增强-终端（Xiaoli Pro-CLI）  
-- **内部版本号**：20260428
+- **内部版本号**：20260801
-- **发布版本**：发行版5.1.3
+- **发布版本**：发行版8.0.3
 
 **技术规格**  
 - **主要语言**：简体中文（UTF-8）  

agents/coder.md

@@ -0,0 +1,16 @@
+---
+name: coder
+description: 代码实现专家，负责编写、修改、重构代码文件
+tools: code_editor, git_tools, file_manager, cmd_executor
+model:
+---
+你是一个资深程序员「coder」，专注于把需求变成可运行的代码。
+
+工作准则：
+1. 先读取相关文件，理解现有结构与风格，再动手。
+2. 改动保持最小且自洽，新增代码加必要注释。
+3. 涉及多文件时，先给出改动清单再逐个落地。
+4. 完成后做冒烟验证（语法 / 类型 / 单元测试）。
+
+任务完成后，用如下 JSON 报告：
+{"status": "done", "summary": "一句话总结", "details": "关键改动与验证结果"}

agents/reviewer.md

@@ -0,0 +1,20 @@
+---
+name: reviewer
+description: 代码审查专家，负责检查质量、安全与一致性
+tools: code_editor, git_tools, file_manager
+model:
+---
+你是一个严格的代码审查员「reviewer」，独立于实现者审视代码。
+
+审查维度：
+1. 正确性：逻辑漏洞、边界条件、异常处理。
+2. 安全性：注入、越权、敏感信息泄露、危险命令。
+3. 一致性：命名、风格、是否贴合项目既有约定。
+4. 可维护性：重复、复杂度、可读性。
+
+输出格式（按严重度分级）：
+- 🔴 严重：必须修改
+- 🟡 建议：建议修改
+- 🟢 亮点：值得保留
+
+最后给出总体评价与是否可合入的结论。

ai_engines/manual_engine.py

@@ -18,6 +18,17 @@ current_dir = os.path.dirname(os.path.abspath(__file__))
 project_root = os.path.dirname(current_dir)
 sys.path.insert(0, project_root)
 
+try:
+    from xcli_core.tool_result import normalize_tool_text
+except Exception:
+    def normalize_tool_text(result, default="无结果"):
+        """兜底归一化：ToolResult / dict / str 统一成字符串"""
+        if result is None:
+            return default
+        if isinstance(result, dict):
+            return str(result.get("result", default))
+        return str(result)
+
 # 通信用的临时目录
 COMM_DIR = os.path.join(tempfile.gettempdir(), "xiaoli_manual_engine")
 
@@ -33,6 +44,8 @@ class ManualAI:
         self.conversation_history = []
         self.max_history = 10
         self._subprocess = None
+        # 最近一次请求的 prompt token 数（手动引擎无真实模型，恒为 None）
+        self.last_prompt_tokens = None
         self._session_id = str(int(time.time()))
         self._started = False
 
@@ -229,6 +242,11 @@ while True:
             time.sleep(0.3)
 '''
 
+    @property
+    def max_input_tokens(self):
+        """当前引擎支持的最大输入 token 数（统一 API）。手动引擎无真实模型，给名义上限。"""
+        return 128000
+
     def generate_response(self, user_input, tool_results=None, system_prompt=None):
         """生成响应 — 转发给另一个终端的人工"""
         paths = self._get_paths()
@@ -246,7 +264,7 @@ while True:
         # 把用户消息写入文件
         msg = user_input
         if tool_results:
-            msg = f"[工具结果] {tool_results.get('result', '无结果')}\n\n{user_input}"
+            msg = f"[工具结果] {normalize_tool_text(tool_results)}\n\n{user_input}"
 
         to_human = paths["msg_to_human"]
         with open(to_human, 'w', encoding='utf-8') as f:

ai_engines/ollama_engine.py

@@ -1,4 +1,5 @@
 import os
+import re
 import sys
 import json
 import requests
@@ -6,21 +7,57 @@ import threading
 import time
 from colorama import Fore, Style
 
-# 延迟导入 ollama，避免程序启动时卡住
+# 真正的惰性导入：只有真正用到 ollama 引擎时才 import ollama，
-# 如果导入失败，ollama_client 将为 None
+# 避免 xiaoli-cli 启动 / 加载所有引擎时因 ollama SDK 初始化
-try:
+# （其内部 platform.machine() -> WMI 查询）在某些环境下卡住。
-    import ollama
+_ollama_client = None
-    ollama_client = ollama
+_ollama_import_tried = False
-except Exception as e:
+
-    print(f"{Fore.YELLOW}警告: 无法导入 ollama 库: {e}{Style.RESET_ALL}")
+
-    print(f"{Fore.YELLOW}请确保已安装 ollama 包 (pip install ollama){Style.RESET_ALL}")
+def _ollama():
-    ollama_client = None
+    """惰性获取 ollama SDK 客户端；导入失败返回 None（仅尝试一次）。"""
+    global _ollama_client, _ollama_import_tried
+    if _ollama_import_tried:
+        return _ollama_client
+    _ollama_import_tried = True
+    try:
+        import ollama
+        _ollama_client = ollama
+    except Exception as e:
+        _ollama_client = None
+        print(f"{Fore.YELLOW}警告: 无法导入 ollama 库: {e}{Style.RESET_ALL}")
+        print(f"{Fore.YELLOW}请确保已安装 ollama 包 (pip install ollama){Style.RESET_ALL}")
+    return _ollama_client
 
 # 添加项目根目录到sys.path
 current_dir = os.path.dirname(os.path.abspath(__file__))
 project_root = os.path.dirname(current_dir)
 sys.path.insert(0, project_root)
 
+try:
+    from xcli_core.verbose import vprint as _vprint
+except Exception:
+    def _vprint(*args, **kwargs):
+        pass
+
+# <input> 特殊语法解析（配置模型时可交互式询问用户）
+try:
+    from xcli_core.config import resolve_input_value
+except Exception:
+    def resolve_input_value(value, field_name="", default_prompt=None):
+        return value
+
+try:
+    from xcli_core.tool_result import normalize_tool_text
+except Exception:
+    def normalize_tool_text(result, default="无结果"):
+        """兜底归一化：ToolResult / dict / str 统一成字符串"""
+        if result is None:
+            return default
+        if isinstance(result, dict):
+            return str(result.get("result", default))
+        return str(result)
+
 
 class OllamaAI:
     """基于Ollama的本地AI引擎"""
@@ -62,6 +99,9 @@ Ollama AI引擎插件帮助信息
         self.conversation_history = []
         
         # 直接从 config.json 读取配置
+        # 模型最大 token 大小（单位 K，如 32 = 32K = 32768 token）
+        # 支持 <input> 语法：配置里写 <input:提示语> 会在配置模型时询问用户
+        max_token_k = "<input:请输入模型最大token大小(K)>"
         config_file = os.path.join(project_root, "config.json")
         try:
             with open(config_file, 'r', encoding='utf-8') as f:
@@ -71,6 +111,7 @@ Ollama AI引擎插件帮助信息
                 self.base_url = engine_config.get("base_url", "")
                 self.model = engine_config.get("model", "")
                 self.max_history = config.get("system", {}).get("max_history", 10)
+                max_token_k = engine_config.get("max_token_k", max_token_k)
         except Exception as e:
             # 如果读取失败，使用空值
             self.api_key = ""
@@ -78,35 +119,78 @@ Ollama AI引擎插件帮助信息
             self.model = ""
             self.max_history = 10
 
+        # 解析 <input> 语法（询问用户）并换算成 num_ctx（token 数）
+        resolved_k = resolve_input_value(max_token_k, field_name="模型最大token大小(K)")
+        self.num_ctx = self._parse_num_ctx(resolved_k)
+        
         # 共享对话历史引用（由CLI设置）
         self.shared_conversation_history = None
 
+        # 最近一次请求的 prompt token 数（供 token 感知压缩使用）
+        self.last_prompt_tokens = None
+        
         # 直接启用思考标记，由主程序自动处理深度思考模型的响应
         self.thinking_start_marker = "<thinking>"
         self.thinking_end_marker = "</thinking>"
 
-        # 检测Ollama服务是否运行
+        # 惰性服务检测：不在构造时急切点服务（避免无 ollama / 沙箱环境下
-        self.is_service_running = self.check_ollama_service()
+        # import ollama 卡顿）。首次真正使用时（generate_response 等）才检测并缓存。
+        self._service_checked = False
+        self._service_running = False
     
-        if self.is_service_running:
+    @staticmethod
-            print(f"{Fore.GREEN}Ollama服务检测成功,引擎已启用{Style.RESET_ALL}")
+    def _parse_num_ctx(value):
-        else:
+        """把用户输入的 K 值换算成 token 数：32 / 32K → 32768。非法/空 → 4096"""
-            print(f"{Fore.RED}Ollama服务未运行,引擎将不可用{Style.RESET_ALL}")
+        if value is None:
+            return 4096
+        s = str(value).strip().lower()
+        digits = re.sub(r"[^\d.]", "", s)
+        if not digits:
+            return 4096
+        try:
+            k = float(digits)
+        except ValueError:
+            return 4096
+        if k <= 0:
+            return 4096
+        return max(1024, int(k * 1024))
 
     def check_ollama_service(self):
         """检测Ollama服务是否运行"""
         # 检查 ollama 库是否可用
-        if ollama_client is None:
+        if _ollama() is None:
             return False
         try:
             # 尝试调用Ollama API获取版本信息
-            response = ollama_client.list()
+            response = _ollama().list()
             # 如果没有异常,说明服务运行正常
             return True
         except Exception as e:
             # 如果出现异常,说明服务未运行
             return False
 
+    @property
+    def max_input_tokens(self):
+        """当前引擎支持的最大输入 token 数（统一 API，供上下文压缩使用）。
+        本地 Ollama 由配置 max_token_k 换算的 num_ctx 决定。"""
+        return self.num_ctx
+
+    @property
+    def is_service_running(self):
+        """惰性检测 Ollama 服务是否运行：首次访问时才 import ollama 并点服务。
+
+        这样 AICLI() 构造 / 加载所有引擎时不会因 import ollama 卡住
+        （沙箱或 ollama 未安装的环境也能正常启动 CLI，仅真正选用本引擎时才校验）。
+        """
+        if not self._service_checked:
+            self._service_running = self.check_ollama_service()
+            self._service_checked = True
+            if self._service_running:
+                _vprint(f"{Fore.GREEN}Ollama服务检测成功,引擎已启用{Style.RESET_ALL}")
+            else:
+                print(f"{Fore.RED}Ollama服务未运行,引擎将不可用{Style.RESET_ALL}")
+        return self._service_running
+
     def generate_response(self, user_input, tool_results=None, system_prompt=None, tools=None):
         """
         生成AI响应 - 使用Ollama API
@@ -131,27 +215,46 @@ Ollama AI引擎插件帮助信息
 
             # 如果有工具结果,将其作为上下文返回给AI处理
             if tool_results:
-                tool_result_text = f"工具执行结果: {tool_results.get('result', '无结果')}"
+                tool_result_text = f"工具执行结果: {normalize_tool_text(tool_results)}"
                 messages = self._build_messages_with_history(system_prompt, tool_result_text)
             else:
                 messages = self._build_messages_with_history(system_prompt, user_input)
 
-            # 构建调用参数
+            # 构建调用参数（num_ctx: 模型上下文窗口大小，由配置的 max_token_k 决定）
             kwargs = {
                 "model": self.model,
                 "messages": messages,
-                "stream": False
+                "stream": False,
+                "options": {"num_ctx": self.num_ctx}
             }
 
-            # ── Function Calling: 注入工具定义 ──
+            # ── Function Calling: 注入工具定义（仅引擎支持 FC 时下发）──
-            if tools:
+            fc_enabled = bool(tools) and self._fc_supported()
+            if fc_enabled:
                 kwargs["tools"] = tools
 
-            # 使用Ollama Python库调用API
+            # 使用Ollama Python库调用API（FC 不支持时降级为无工具重试一次）
-            response = ollama_client.chat(**kwargs)
+            try:
+                response = _ollama().chat(**kwargs)
+            except Exception as _fc_err:
+                if fc_enabled and "tool" in str(_fc_err).lower():
+                    self._mark_fc_unsupported()
+                    kwargs.pop("tools", None)
+                    response = _ollama().chat(**kwargs)
+                else:
+                    raise
 
             # ── Function Calling: 检查 tool_calls ──
             message = response.get("message", {})
+            # 上报本轮 prompt token 数（Ollama 返回 prompt_eval_count），供 token 感知压缩
+            try:
+                if isinstance(response, dict):
+                    _pet = response.get("prompt_eval_count")
+                else:
+                    _pet = getattr(response, "prompt_eval_count", None)
+                self.last_prompt_tokens = int(_pet) if _pet else None
+            except Exception:
+                self.last_prompt_tokens = None
             tool_calls_raw = message.get("tool_calls")
             if tool_calls_raw:
                 # Ollama 的 tool_calls 格式: [{"function": {"name": ..., "arguments": {...}}}]
@@ -197,6 +300,22 @@ Ollama AI引擎插件帮助信息
         except Exception as e:
             return f"Ollama API调用失败: {str(e)}"
 
+    def _fc_supported(self):
+        """该 (base_url, model) 是否支持 FC。未探测过时默认 True（乐观，先试一次）"""
+        try:
+            from xcli_core.fc_tools import supports_fc
+            return supports_fc(self.base_url, self.model)
+        except Exception:
+            return False
+
+    def _mark_fc_unsupported(self):
+        """标记该 (base_url, model) 不支持 FC 并持久化"""
+        try:
+            from xcli_core.fc_tools import mark_fc_unsupported
+            mark_fc_unsupported(self.base_url, self.model)
+        except Exception:
+            pass
+
     def _build_messages_with_history(self, system_prompt, user_input):
         """构建包含历史记录的消息列表（支持 FC tool 消息）"""
         messages = []
@@ -217,13 +336,11 @@ Ollama AI引擎插件帮助信息
 
                 role = msg["role"]
 
-                # tool 角色消息（FC 工具结果）
+                # tool 角色消息（FC 工具结果）—— 官方格式不含 name 字段
                 if role == "tool":
                     tool_msg = {"role": "tool", "content": msg.get("content", "")}
                     if "tool_call_id" in msg:
                         tool_msg["tool_call_id"] = msg["tool_call_id"]
-                    if "name" in msg:
-                        tool_msg["name"] = msg["name"]
                     messages.append(tool_msg)
                     continue
 
@@ -285,7 +402,7 @@ Ollama AI引擎插件帮助信息
         })
         
         # 使用Ollama Python库调用API
-        response = ollama_client.chat(
+        response = _ollama().chat(
             model=self.model,
             messages=messages,
             stream=False
@@ -308,7 +425,7 @@ Ollama AI引擎插件帮助信息
             return ["错误:Ollama服务未运行"]
         
         try:
-            response = ollama_client.list()
+            response = _ollama().list()
             models = [model["name"] for model in response["models"]]
             return models
         except Exception as e:
@@ -388,7 +505,7 @@ Ollama AI引擎插件帮助信息
             print(f"{Fore.YELLOW}正在拉取模型 {model_name}...{Style.RESET_ALL}")
             
             # 使用Ollama Python库拉取模型
-            response = ollama_client.pull(model_name, stream=True)
+            response = _ollama().pull(model_name, stream=True)
             
             # 逐块处理响应
             for part in response:

ai_engines/openai_engine.py

@@ -13,6 +13,56 @@ import logging
 # 设置日志
 logger = logging.getLogger(__name__)
 
+try:
+    from xcli_core.config import get_system_config
+except Exception:
+    def get_system_config(key, default=None):
+        return default
+
+try:
+    from xcli_core import model_registry
+except Exception:
+    model_registry = None
+
+try:
+    from xcli_core.verbose import vprint
+except Exception:
+    def vprint(*args, **kwargs):
+        pass
+
+try:
+    from xcli_core.tool_result import normalize_tool_text
+except Exception:
+    def normalize_tool_text(result, default="无结果"):
+        """兜底归一化：ToolResult / dict / str 统一成字符串"""
+        if result is None:
+            return default
+        if isinstance(result, dict):
+            return str(result.get("result", default))
+        return str(result)
+
+# ── 常见模型的上下文窗口（token 数）──
+# OpenAI /chat/completions 协议本身不返回模型窗口，需本地维护。
+MODEL_CONTEXT_WINDOWS = {
+    # 讯飞星火（Spark，经 one-api/new-api 转发时模型名常为 spark-* / general*）
+    "spark-4.0": 128000, "spark-max": 8192, "spark-pro": 8192, "spark-lite": 8192,
+    "spark": 8192, "generalv3.5": 8192, "generalv3": 8192, "general": 4096,
+    "gpt-4o": 128000, "gpt-4o-mini": 128000, "gpt-4": 8192, "gpt-4-turbo": 128000,
+    "gpt-3.5-turbo": 16385,
+    "deepseek-chat": 64000, "deepseek-reasoner": 64000,
+    "qwen2.5": 32768, "qwen2": 32768, "qwen-max": 32768, "qwen-plus": 32768,
+    "llama3": 8192, "llama3.1": 128000, "llama3.2": 128000, "llama2": 4096,
+    "claude-3": 200000, "claude-3.5": 200000, "claude-3.7": 200000, "claude": 200000,
+    "grok": 131072, "grok-2": 131072, "grok-3": 131072,
+    "glm-4": 128000, "glm-4v": 128000,
+    "mistral": 32768, "mixtral": 32768, "mistral-large": 128000,
+    "gemini": 1000000, "gemini-1.5": 1000000, "gemini-2.0": 1000000, "gemini-2.5": 1000000,
+    "yi": 200000, "yi-large": 200000,
+    "kimi": 200000, "moonshot": 200000,
+    "abab": 200000, "minimax": 200000,
+    "baichuan": 192000, "chatglm": 32768,
+}
+
 # 添加项目根目录到sys.path
 current_dir = os.path.dirname(os.path.abspath(__file__))
 project_root = os.path.dirname(current_dir)
@@ -95,6 +145,12 @@ OpenAI 兼容格式引擎帮助信息
         self.conversation_history = []
         self.shared_conversation_history = None
 
+        # 应用注册表中的当前模型（覆盖扁平默认配置，向后兼容）
+        self._apply_registry_model()
+
+        # 最近一次请求的 token 用量（供 token 感知压缩使用）
+        self.last_prompt_tokens = None
+
         # 思考标记（兼容深度思考模型）
         self.thinking_start_marker = "<think>"
         self.thinking_end_marker = "</think>"
@@ -105,7 +161,64 @@ OpenAI 兼容格式引擎帮助信息
         if not self.base_url:
             print(f"{Fore.YELLOW}警告: openai 引擎的 base_url 未设置{Style.RESET_ALL}")
         else:
-            print(f"{Fore.GREEN}OpenAI 兼容引擎已启用 (base_url: {self.base_url}){Style.RESET_ALL}")
+            vprint(f"{Fore.GREEN}OpenAI 兼容引擎已启用 (base_url: {self.base_url}){Style.RESET_ALL}")
+
+    # ── 模型注册表（多模型在线切换）──
+
+    def _apply_registry_model(self):
+        """若注册表存在 current 模型，用它覆盖 __init__ 读到的扁平默认配置。"""
+        if not model_registry:
+            return
+        try:
+            m = model_registry.get_model()
+            if m:
+                self.api_key = m.get("api_key", self.api_key)
+                self.base_url = (m.get("base_url") or self.base_url).rstrip('/')
+                self.model = m.get("model", self.model)
+        except Exception:
+            pass
+
+    def list_registry_models(self):
+        """返回 (models, current_name)，供 CLI 列出已配置模型。"""
+        if not model_registry:
+            return [], None
+        return model_registry.list_models()
+
+    def current_model_name(self):
+        """当前激活模型在注册表中的名称（兼容单槽位时返回 model 值）。"""
+        if not model_registry:
+            return self.model
+        try:
+            models, current = model_registry.list_models()
+            return current or self.model
+        except Exception:
+            return self.model
+
+    def set_registry_model(self, name):
+        """切换到注册表中指定模型（含 base_url/api_key/model）并持久化 current。
+        成功返回 True。"""
+        if not model_registry:
+            return False
+        m = model_registry.get_model(name)
+        if not m:
+            return False
+        self.api_key = m.get("api_key", "")
+        self.base_url = (m.get("base_url") or "").rstrip('/')
+        self.model = m.get("model", "")
+        model_registry.set_current(name)
+        return True
+
+    def add_model(self, entry):
+        """新增/更新模型并持久化。entry: {name, base_url, api_key, model}。"""
+        if not model_registry:
+            return None
+        return model_registry.add_model(entry)
+
+    def remove_model(self, name):
+        """删除模型并持久化。成功返回 True。"""
+        if not model_registry:
+            return False
+        return model_registry.remove_model(name)
 
     def _get_headers(self):
         """构建请求头"""
@@ -125,6 +238,26 @@ OpenAI 兼容格式引擎帮助信息
             url = url.rstrip('/') + '/chat/completions'
         return url
 
+    # ── FC 能力探测缓存 ──
+
+    def _fc_supported(self):
+        """当前 (base_url, model) 是否支持 function calling（未探测过默认支持）"""
+        try:
+            from xcli_core.fc_tools import supports_fc
+            return supports_fc(self.base_url, self.model)
+        except Exception:
+            return True
+
+    def _mark_fc_unsupported(self):
+        """降级成功后记住该模型不支持 FC，后续请求不再带 tools"""
+        try:
+            from xcli_core.fc_tools import mark_fc_unsupported
+            if mark_fc_unsupported(self.base_url, self.model):
+                print(f"{Fore.CYAN}已记住 {self.model} 不支持 FC，后续请求将跳过工具字段"
+                      f"（/fc reset 可清除）{Style.RESET_ALL}")
+        except Exception:
+            pass
+
     def generate_response(self, user_input, tool_results=None, system_prompt=None, tools=None):
         """
         生成 AI 响应
@@ -141,7 +274,7 @@ OpenAI 兼容格式引擎帮助信息
 
             # 处理工具结果
             if tool_results:
-                tool_result_text = f"工具执行结果: {tool_results.get('result', '无结果')}"
+                tool_result_text = f"工具执行结果: {normalize_tool_text(tool_results)}"
                 messages = self._build_messages_with_history(system_prompt, tool_result_text)
             else:
                 messages = self._build_messages_with_history(system_prompt, user_input)
@@ -153,10 +286,20 @@ OpenAI 兼容格式引擎帮助信息
                 "stream": False
             }
 
-            # ── Function Calling: 注入工具定义 ──
+            # ── Function Calling: 注入工具定义（统一纯官方格式）──
-            if tools:
+            # 已探测过不支持 FC 的模型直接跳过 tools，省掉每次白撞一次 500
-                body["tools"] = tools
+            if tools and self._fc_supported():
-                body["tool_choice"] = "auto"
+                pure_tools = []
+                for t in tools:
+                    if not isinstance(t, dict):
+                        continue
+                    if t.get("type") == "function" and isinstance(t.get("function"), dict):
+                        pure_tools.append(t)
+                    elif isinstance(t.get("function"), dict):
+                        pure_tools.append({"type": "function", "function": t["function"]})
+                if pure_tools:
+                    body["tools"] = pure_tools
+                    body["tool_choice"] = "auto"
 
             # 发送请求
             api_url = self._get_api_url()
@@ -169,6 +312,13 @@ OpenAI 兼容格式引擎帮助信息
 
             if response.status_code == 200:
                 response_data = response.json()
+                # ── 记录 token 用量（供 token 感知压缩）──
+                usage = response_data.get("usage", {})
+                if isinstance(usage, dict) and usage.get("prompt_tokens"):
+                    try:
+                        self.last_prompt_tokens = int(usage["prompt_tokens"])
+                    except (TypeError, ValueError):
+                        self.last_prompt_tokens = None
                 if "choices" in response_data and len(response_data["choices"]) > 0:
                     message = response_data["choices"][0].get("message", {})
 
@@ -202,19 +352,41 @@ OpenAI 兼容格式引擎帮助信息
                 return f"API 端点不存在，请检查 base_url 是否正确: {api_url}"
             elif response.status_code == 400:
                 # 工具定义可能不兼容，降级为无工具重试
-                if tools and "tools" in str(response.text).lower():
+                if body.get("tools") and "tools" in str(response.text).lower():
                     print(f"{Fore.YELLOW}FC 不支持，降级为普通模式{Style.RESET_ALL}")
                     body.pop("tools", None)
                     body.pop("tool_choice", None)
                     retry = requests.post(url=api_url, json=body, headers=self._get_headers(), timeout=None)
                     if retry.status_code == 200:
+                        self._mark_fc_unsupported()
                         rd = retry.json()
                         if "choices" in rd and rd["choices"]:
                             ai_response = rd["choices"][0].get("message", {}).get("content", "")
                             return ai_response.encode('utf-8', 'replace').decode('utf-8') if ai_response else ""
-                return f"API 调用失败: {response.status_code} - {response.text}"
+                return self._friendly_error(response.status_code, response)
             else:
-                return f"API 调用失败: {response.status_code} - {response.text}"
+                # 500 等错误：部分服务（如讯飞星火经 one-api 转发）不支持 FC 工具，
+                # 会直接报 500 Invalid Params —— 尝试去掉 tools 降级重试一次
+                lower = str(response.text).lower()
+                if body.get("tools") and ("xunfei" in lower or "invalid params" in lower or "requestparamserror" in lower):
+                    print(f"{Fore.YELLOW}该接口可能不支持 FC 工具，降级为普通模式重试...{Style.RESET_ALL}")
+                    body.pop("tools", None)
+                    body.pop("tool_choice", None)
+                    retry = requests.post(url=api_url, json=body, headers=self._get_headers(), timeout=None)
+                    if retry.status_code == 200:
+                        self._mark_fc_unsupported()
+                        rd = retry.json()
+                        retry_usage = rd.get("usage", {})
+                        if isinstance(retry_usage, dict) and retry_usage.get("prompt_tokens"):
+                            try:
+                                self.last_prompt_tokens = int(retry_usage["prompt_tokens"])
+                            except (TypeError, ValueError):
+                                pass
+                        if "choices" in rd and rd["choices"]:
+                            ai_response = rd["choices"][0].get("message", {}).get("content", "")
+                            ai_response = self._process_thinking_content(ai_response)
+                            return ai_response.encode('utf-8', 'replace').decode('utf-8') if ai_response else ""
+                return self._friendly_error(response.status_code, response)
 
         except requests.exceptions.ConnectionError:
             return f"无法连接到 API 服务: {self.base_url}，请检查服务是否运行"
@@ -241,22 +413,20 @@ OpenAI 兼容格式引擎帮助信息
 
                 role = msg["role"]
 
-                # tool 角色消息（FC 工具结果）
+                # tool 角色消息（FC 工具结果）—— 官方 OpenAI 格式不含 name 字段
                 if role == "tool":
                     tool_msg = {"role": "tool", "content": msg.get("content", "")}
                     if "tool_call_id" in msg:
                         tool_msg["tool_call_id"] = msg["tool_call_id"]
-                    if "name" in msg:
-                        tool_msg["name"] = msg["name"]
                     messages.append(tool_msg)
                     continue
 
-                # assistant 消息可能包含 tool_calls
+                # assistant 消息可能包含 tool_calls —— 重建成纯官方结构
                 if role == "assistant" and "tool_calls" in msg:
                     assistant_msg = {
                         "role": "assistant",
-                        "content": msg.get("content"),
+                        "content": msg.get("content") or "",
-                        "tool_calls": msg["tool_calls"]
+                        "tool_calls": self._pure_tool_calls(msg["tool_calls"])
                     }
                     messages.append(assistant_msg)
                     continue
@@ -337,11 +507,86 @@ OpenAI 兼容格式引擎帮助信息
 
         return response
 
+    @staticmethod
+    def _friendly_error(status_code, response):
+        """把错误响应里的可读信息提取出来，附带常见原因提示"""
+        try:
+            data = response.json()
+            err = data.get("error", {})
+            msg = err.get("message", "") if isinstance(err, dict) else ""
+        except Exception:
+            msg = ""
+        if not msg:
+            return f"API 调用失败: {status_code} - {response.text}"
+
+        hint = ""
+        low = str(msg).lower()
+        if "invalid params" in low or "requestparamserror" in low or "xunfei" in low:
+            hint = ("\n提示: 讯飞星火接口报'参数错误'，常见原因:"
+                    "\n  1) model 名与 one-api/new-api 渠道里的模型映射不一致（如 spark-4.0-ultra / spark-max / generalv3.5）"
+                    "\n  2) 该模型/渠道不支持 function calling（引擎已尝试去掉 tools 重试）"
+                    "\n  3) base_url 需为 OpenAI 兼容转发地址（如 one-api 的 https://域名/v1）")
+        return f"API 调用失败: {status_code} - {msg}{hint}"
+
+    @staticmethod
+    def _pure_tool_calls(tool_calls):
+        """把历史里的 tool_calls 重建成官方 OpenAI 格式:
+        [{"id": ..., "type": "function", "function": {"name": ..., "arguments": "json串"}}]
+        剔除内部字段（action/tool/args 等），arguments 统一为字符串。
+        """
+        pure = []
+        for tc in (tool_calls or []):
+            if not isinstance(tc, dict):
+                continue
+            func = tc.get("function")
+            if isinstance(func, dict):
+                name = func.get("name", tc.get("name", ""))
+                args = func.get("arguments", "{}")
+            else:
+                name = tc.get("name", "")
+                args = tc.get("arguments", "{}")
+            if isinstance(args, str):
+                args_str = args
+            else:
+                try:
+                    args_str = json.dumps(args, ensure_ascii=False)
+                except (TypeError, ValueError):
+                    args_str = "{}"
+            pure.append({
+                "id": tc.get("id", f"call_{name}"),
+                "type": "function",
+                "function": {"name": name, "arguments": args_str}
+            })
+        return pure
+
     def set_model(self, model_name):
         """设置模型"""
         self.model = model_name
         print(f"{Fore.GREEN}已切换模型为: {model_name}{Style.RESET_ALL}")
 
+    def context_window(self):
+        """
+        返回当前模型的上下文窗口（token 数）。
+        优先级: config.system.context_window > 模型名查表 > 默认 128000。
+        注: OpenAI /chat/completions 协议本身不返回模型窗口，需本地维护。
+        """
+        cfg = get_system_config("context_window", None)
+        if cfg:
+            try:
+                return int(cfg)
+            except (TypeError, ValueError):
+                pass
+        model = (self.model or "").lower()
+        for key, win in MODEL_CONTEXT_WINDOWS.items():
+            if key and key in model:
+                return win
+        return 128000
+
+    @property
+    def max_input_tokens(self):
+        """当前引擎支持的最大输入 token 数（统一 API，供上下文压缩使用）。"""
+        return self.context_window()
+
     def list_models(self):
         """列出可用模型（调用 API 的 /models 端点）"""
         if not self.base_url:

config.json.example

@@ -4,7 +4,8 @@
       "ollama": {
         "api_key": "",
         "base_url": "http://localhost:11434",
-        "model": "gemma4:31b"
+        "model": "gemma4:31b",
+        "max_token_k": "<input:请输入模型最大token大小(K)>"
       },
       "openai": {
         "api_key": "your-api-key-here",
@@ -16,5 +17,13 @@
   "system": {
     "default_engine": "ollama",
     "max_history": 50
+  },
+  "lspServers": {
+    "_说明": "语言服务器（LSP）。键为语言ID或扩展名，值为启动命令。配置后重启即生效，工具 lsp__diagnostics/hover/definition/references/completion 自动可用，且 AI 写/改 .py 文件后会自动触发诊断并注入模型上下文。python 默认用自研本地服务器（jedi 语义 + py_detect 中文诊断，纯本地零安装）；换回 pylsp 则 command 改 \"pylsp\"、args 清空。",
+    "python": {
+      "command": "python",
+      "args": ["-m", "xcli_core.local_lsp_server"],
+      "disabled": false
+    }
   }
 }

launcher.py

@@ -16,6 +16,13 @@ from pathlib import Path
 from datetime import datetime
 import platform
 
+# <input> 特殊语法：配置模型时可用它交互式询问用户并获取输入
+try:
+    from xcli_core.config import resolve_input_value
+except Exception:
+    def resolve_input_value(value, field_name="", default_prompt=None):
+        return value
+
 # ── 编码设置 ──
 if sys.stdout.encoding != 'utf-8':
     sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding='utf-8')
@@ -120,6 +127,7 @@ class Launcher:
             'colorama', 'openai', 'requests', 'ollama', 'pygame',
             'PIL', 'Pillow', 'cv2', 'opencv-python', 'numpy',
             'ascii_magic', 'tkinter', 'matplotlib', 'pandas',
+            'jedi',  # py_detect L2 语义验证后端 / 自研本地 LSP（纯 Python，预装即启用）
         }
 
     # ── 清屏 ──
@@ -471,6 +479,12 @@ class Launcher:
                     entry['base_url'] = 'https://api.deepseek.com/v1'
                     entry['model'] = 'deepseek-chat'
                 config['api']['engines'][name] = entry
+            if name == 'ollama':
+                # 模型最大 token 大小（单位 K，如 32 = 32K = 32768 token）
+                # 默认用 <input> 语法：配置模型时交互式询问用户
+                config['api']['engines'][name].setdefault(
+                    'max_token_k', '<input:请输入模型最大token大小(K)>'
+                )
 
         self.config_file.write_text(json.dumps(config, ensure_ascii=False, indent=2), 'utf-8')
         return config
@@ -498,7 +512,7 @@ class Launcher:
                     info(f'当前密钥: {masked}')
                 if prompt_yn(f'配置 {name} 的 API 密钥?', 'n'):
                     key = self.secret_input(f'{name} API 密钥')
-                    cfg['api_key'] = key
+                    cfg['api_key'] = resolve_input_value(key, field_name='API 密钥')
 
             # URL
             current_url = cfg.get('base_url', '')
@@ -506,7 +520,7 @@ class Launcher:
                 info(f'当前 URL: {current_url}')
             if prompt_yn(f'配置 {name} 的 Base URL?', 'n'):
                 url = prompt('Base URL', current_url)
-                cfg['base_url'] = url
+                cfg['base_url'] = resolve_input_value(url, field_name='Base URL')
 
             # Model
             current_model = cfg.get('model', '')
@@ -514,7 +528,26 @@ class Launcher:
                 info(f'当前模型: {current_model}')
             if prompt_yn(f'配置 {name} 的模型?', 'n'):
                 model = prompt('模型名称', current_model)
-                cfg['model'] = model
+                cfg['model'] = resolve_input_value(model, field_name='模型')
+
+            # 最大 token 大小（仅 ollama；单位 K，如 32 = 32K = 32768 token；支持 <input> 语法）
+            if name == 'ollama':
+                current_ctx = cfg.get('max_token_k', '')
+                is_syntax = isinstance(current_ctx, str) and '<input' in current_ctx
+                if is_syntax:
+                    info('当前最大 token: (启动时询问)')
+                elif current_ctx:
+                    shown = str(current_ctx)
+                    info(f'当前最大 token: {shown}K' if shown.replace('.', '', 1).isdigit() else f'当前最大 token: {shown}')
+                else:
+                    info('当前最大 token: 未设置')
+                if prompt_yn(f'配置 {name} 的最大 token 大小(K)?', 'n'):
+                    default_ctx = '' if is_syntax else str(current_ctx)
+                    val = prompt('最大 token (K)', default_ctx)
+                    if val.strip():
+                        cfg['max_token_k'] = resolve_input_value(val, field_name='最大 token 大小(K)')
+                elif not current_ctx:
+                    cfg['max_token_k'] = '<input:请输入模型最大token大小(K)>'
 
             config.setdefault('api', {}).setdefault('engines', {})[name] = cfg
 

lsp_client.py

@@ -0,0 +1,617 @@
+"""
+小狸 LSP 客户端
+==============
+连接外部语言服务器（Language Server Protocol），把真实的诊断 / 悬停 /
+定义跳转 / 引用 / 补全喂给模型——这是对标 opencode 的 P2 收尾能力。
+
+- transport = stdio，使用 LSP 标准的 **Content-Length 分帧** JSON-RPC
+  （与 MCP 的逐行 JSON 不同！LSP 消息头是 `Content-Length: N\\r\\n\\r\\n` + N 字节 JSON）。
+- 同步 + 线程安全：与 mcp_client.py 一致，契合小狸「同步主循环 + ThreadPoolExecutor」。
+- 按文件扩展名路由到对应的语言服务器（python→pylsp、rust→rust-analyzer …），
+  配置驱动，不绑定具体 server；未安装/启动失败不影响主程序与其他 LSP。
+- 单个 server 失败不阻断其他，也不阻断主程序启动。
+
+后续可在此文件扩展 workspace 级多文件夹、incremental 文本同步等。
+"""
+
+import os
+import re
+import json
+import time
+import shutil
+import sys
+import threading
+import subprocess
+import logging
+from typing import Dict, List, Any, Optional, Tuple
+
+logger = logging.getLogger(__name__)
+
+
+class LspClientError(RuntimeError):
+    """LSP 客户端错误"""
+
+
+# 扩展名 → LSP languageId
+_EXT_TO_LANG = {
+    ".py": "python", ".pyi": "python",
+    ".rs": "rust",
+    ".ts": "typescript", ".tsx": "typescript",
+    ".js": "javascript", ".jsx": "javascript", ".mjs": "javascript", ".cjs": "javascript",
+    ".go": "go",
+    ".c": "c", ".h": "c",
+    ".cpp": "cpp", ".cc": "cpp", ".cxx": "cpp", ".hpp": "cpp",
+    ".java": "java",
+    ".lua": "lua",
+    ".sh": "shellscript", ".bash": "shellscript",
+    ".json": "json", ".jsonc": "json",
+    ".html": "html", ".htm": "html",
+    ".css": "css", ".scss": "scss", ".less": "less",
+    ".md": "markdown", ".markdown": "markdown",
+    ".vue": "vue",
+    ".rb": "ruby",
+    ".php": "php",
+    ".sql": "sql",
+    ".yaml": "yaml", ".yml": "yaml",
+    ".xml": "xml",
+    ".toml": "toml",
+}
+
+
+def ext_to_language(path: str) -> Optional[str]:
+    """从文件路径推断 LSP languageId"""
+    ext = os.path.splitext(path)[1].lower()
+    return _EXT_TO_LANG.get(ext)
+
+
+def path_to_uri(path: str) -> str:
+    """绝对路径 → file:// URI（Windows 下 file:///C:/...）"""
+    abspath = os.path.abspath(path)
+    # 统一为正斜杠
+    norm = abspath.replace("\\", "/")
+    if not norm.startswith("/"):
+        norm = "/" + norm
+    return "file://" + norm
+
+
+def _parse_position(pos: Any) -> Tuple[int, int]:
+    """兼容 {line,character} / [line,char] / (line,char) → (line, character)（0-based）"""
+    if isinstance(pos, dict):
+        return int(pos.get("line", 0)), int(pos.get("character", 0))
+    if isinstance(pos, (list, tuple)) and len(pos) >= 2:
+        return int(pos[0]), int(pos[1])
+    return 0, 0
+
+
+class LspClient:
+    """单个语言服务器的 stdio 客户端（同步、线程安全、Content-Length 分帧）"""
+
+    def __init__(self, name: str, command: str, args: Optional[List[str]] = None,
+                 env: Optional[Dict[str, str]] = None, cwd: Optional[str] = None,
+                 root_uri: Optional[str] = None, timeout: float = 30.0,
+                 language_id: str = ""):
+        self.name = name
+        self.command = command
+        self.args = list(args or [])
+        self.env = env or {}
+        self.cwd = cwd
+        self.root_uri = root_uri
+        self.timeout = timeout
+        self.language_id = language_id
+
+        self.proc = None
+        self._lock = threading.RLock()
+        self._next_id = 1
+        self._pending: Dict[int, Dict[str, Any]] = {}
+        self._reader_thread = None
+        self._started = False
+        self._opened: set = set()
+        self._diagnostics: Dict[str, List[Dict]] = {}  # uri -> diagnostics
+        self.capabilities: Dict[str, Any] = {}
+        self._diag_lock = threading.Lock()
+
+    # ── 生命周期 ──
+    def start(self) -> None:
+        if self._started:
+            return
+        full_env = dict(os.environ)
+        full_env.update(self.env)
+        cmd = [self.command] + self.args
+        # 鲁棒性: command 为裸 "python"/"python3" 但 PATH 找不到时, 回退当前解释器
+        if cmd and os.path.basename(cmd[0]) in ("python", "python3") and os.path.dirname(cmd[0]) == "":
+            found = None
+            for cand in (cmd[0], cmd[0] + ".exe"):
+                if found is None and shutil.which(cand):
+                    found = cand
+            if not found:
+                cmd = [sys.executable] + self.args
+        try:
+            # 二进制模式：避免 CJK 字符数与 Content-Length 字节数不一致
+            self.proc = subprocess.Popen(
+                cmd,
+                stdin=subprocess.PIPE,
+                stdout=subprocess.PIPE,
+                stderr=subprocess.DEVNULL,
+                cwd=self.cwd,
+                env=full_env,
+                bufsize=0,
+            )
+        except Exception as e:
+            raise LspClientError(f"启动 LSP server '{self.name}' 失败: {e}")
+
+        self._reader_thread = threading.Thread(target=self._reader_loop, daemon=True)
+        self._reader_thread.start()
+        self._started = True
+
+    def _reader_loop(self):
+        """后台读线程：按 Content-Length 分帧解析 JSON-RPC，响应按 id 投递；
+        通知（publishDiagnostics 等）走回调。"""
+        try:
+            while True:
+                msg = self._read_message()
+                if msg is None:
+                    break
+                self._dispatch(msg)
+        except Exception:
+            pass
+        finally:
+            with self._lock:
+                pendings = list(self._pending.values())
+                self._pending.clear()
+            for pend in pendings:
+                pend["error"] = {"code": -1, "message": "LSP server 连接断开"}
+                pend["event"].set()
+
+    def _read_message(self) -> Optional[Dict[str, Any]]:
+        """从 stdout 读取一条 Content-Length 分帧消息，返回解析后的 dict 或 None(EOF)"""
+        # 读头部，直到遇到空行（\r\n\r\n）
+        header = b""
+        while b"\r\n\r\n" not in header:
+            chunk = self.proc.stdout.read(1)
+            if not chunk:
+                return None
+            header += chunk
+            if len(header) > 65536:  # 防御：头部异常大
+                return None
+        # 解析 Content-Length
+        m = re.search(rb"Content-Length:\s*(\d+)", header)
+        if not m:
+            return None
+        length = int(m.group(1))
+        body = b""
+        while len(body) < length:
+            chunk = self.proc.stdout.read(length - len(body))
+            if not chunk:
+                return None
+            body += chunk
+        try:
+            return json.loads(body.decode("utf-8"))
+        except (json.JSONDecodeError, UnicodeDecodeError):
+            return None
+
+    def _dispatch(self, msg: Dict[str, Any]) -> None:
+        msg_id = msg.get("id")
+        method = msg.get("method")
+        if msg_id is not None and "method" not in msg:
+            # 响应
+            with self._lock:
+                pend = self._pending.pop(msg_id, None)
+            if pend is not None:
+                if "error" in msg:
+                    pend["error"] = msg["error"]
+                else:
+                    pend["value"] = msg.get("result")
+                pend["event"].set()
+            return
+        # 通知（无 id 或 method 在通知里）
+        if method == "textDocument/publishDiagnostics":
+            params = msg.get("params", {})
+            uri = params.get("uri", "")
+            diags = params.get("diagnostics", [])
+            with self._diag_lock:
+                self._diagnostics[uri] = diags
+        # 其余通知（window/logMessage 等）忽略
+
+    def _send(self, method: str, params: Any = None,
+              is_notification: bool = False) -> Optional[Any]:
+        if not self._started or self.proc is None:
+            raise LspClientError(f"LSP client '{self.name}' 未启动")
+        with self._lock:
+            pend = None
+            req_id = None
+            if not is_notification:
+                req_id = self._next_id
+                self._next_id += 1
+                pend = {"event": threading.Event(), "value": None, "error": None}
+                self._pending[req_id] = pend
+            req = {"jsonrpc": "2.0", "method": method}
+            if not is_notification:
+                req["id"] = req_id
+            if params is not None:
+                req["params"] = params
+        try:
+            data = json.dumps(req, ensure_ascii=False).encode("utf-8")
+            header = f"Content-Length: {len(data)}\r\n\r\n".encode("ascii")
+            self.proc.stdin.write(header + data)
+            self.proc.stdin.flush()
+        except (BrokenPipeError, ValueError, OSError) as e:
+            raise LspClientError(f"向 LSP server '{self.name}' 写入失败: {e}")
+        if is_notification:
+            return None
+        if not pend["event"].wait(self.timeout):
+            with self._lock:
+                self._pending.pop(req_id, None)
+            raise LspClientError(f"LSP server '{self.name}' 响应超时: {method}")
+        if pend["error"]:
+            err = pend["error"]
+            raise LspClientError(
+                f"LSP server '{self.name}' 错误 [{err.get('code')}]: {err.get('message')}")
+        return pend["value"]
+
+    # ── LSP 协议方法 ──
+    def initialize(self, root_uri: Optional[str] = None) -> Dict[str, Any]:
+        if root_uri:
+            self.root_uri = root_uri
+        result = self._send("initialize", {
+            "processId": os.getpid(),
+            "rootUri": self.root_uri,
+            "capabilities": {
+                "textDocument": {
+                    "synchronization": {"dynamicRegistration": True,
+                                          "didSave": True},
+                    "publishDiagnostics": {"relatedInformation": True},
+                    "hover": {"contentFormat": ["markdown", "plaintext"]},
+                    "definition": {"linkSupport": True},
+                    "references": {},
+                    "completion": {"completionItem": {"snippetSupport": True}},
+                },
+                "workspace": {"workspaceFolders": True,
+                               "configuration": True},
+            },
+            "clientInfo": {"name": "xiaoli-cli", "version": "8.0.3"},
+        })
+        if isinstance(result, dict):
+            self.capabilities = result.get("capabilities", {}) or {}
+        # 通知服务器初始化完成
+        self._send("initialized", {}, is_notification=True)
+        return result or {}
+
+    def _ensure_open(self, uri: str, text: str) -> None:
+        if uri in self._opened:
+            # 已打开则同步最新内容（full 同步）。didChange 是通知，无需等待响应。
+            self._send("textDocument/didChange", {
+                "textDocument": {"uri": uri},
+                "contentChanges": [{"text": text}],
+            }, is_notification=True)
+            return
+        lang = self.language_id or "plaintext"
+        # didOpen 是通知，无需等待响应
+        self._send("textDocument/didOpen", {
+            "textDocument": {
+                "uri": uri,
+                "languageId": lang,
+                "version": 1,
+                "text": text,
+            }
+        }, is_notification=True)
+        self._opened.add(uri)
+
+    def _read_file(self, file_path: str) -> str:
+        try:
+            with open(file_path, "r", encoding="utf-8", errors="replace") as f:
+                return f.read()
+        except Exception as e:
+            return f"# 无法读取文件 {file_path}: {e}"
+
+    def diagnostics(self, file_path: str, wait: float = 2.0) -> List[Dict]:
+        """返回某文件的诊断（先 didOpen 触发分析，等待并收集 publishDiagnostics）"""
+        uri = path_to_uri(file_path)
+        text = self._read_file(file_path)
+        self._ensure_open(uri, text)
+        # 等待服务器推送诊断（最多 wait 秒）
+        deadline = time.time() + wait
+        while time.time() < deadline:
+            with self._diag_lock:
+                if uri in self._diagnostics:
+                    return list(self._diagnostics[uri])
+            time.sleep(0.1)
+        with self._diag_lock:
+            return list(self._diagnostics.get(uri, []))
+
+    def hover(self, file_path: str, line: int, character: int) -> Any:
+        uri = path_to_uri(file_path)
+        text = self._read_file(file_path)
+        self._ensure_open(uri, text)
+        return self._send("textDocument/hover", {
+            "textDocument": {"uri": uri},
+            "position": {"line": line, "character": character},
+        })
+
+    def definition(self, file_path: str, line: int, character: int) -> Any:
+        uri = path_to_uri(file_path)
+        text = self._read_file(file_path)
+        self._ensure_open(uri, text)
+        return self._send("textDocument/definition", {
+            "textDocument": {"uri": uri},
+            "position": {"line": line, "character": character},
+        })
+
+    def references(self, file_path: str, line: int, character: int) -> Any:
+        uri = path_to_uri(file_path)
+        text = self._read_file(file_path)
+        self._ensure_open(uri, text)
+        return self._send("textDocument/references", {
+            "textDocument": {"uri": uri},
+            "position": {"line": line, "character": character},
+            "context": {"includeDeclaration": True},
+        })
+
+    def completion(self, file_path: str, line: int, character: int) -> Any:
+        uri = path_to_uri(file_path)
+        text = self._read_file(file_path)
+        self._ensure_open(uri, text)
+        return self._send("textDocument/completion", {
+            "textDocument": {"uri": uri},
+            "position": {"line": line, "character": character},
+        })
+
+    def has_diagnostics(self, file_path: str) -> bool:
+        uri = path_to_uri(file_path)
+        with self._diag_lock:
+            return bool(self._diagnostics.get(uri))
+
+    def shutdown(self) -> None:
+        if self.proc is None:
+            return
+        try:
+            self._send("shutdown", {})
+        except Exception:
+            pass
+        try:
+            self._send("exit", {}, is_notification=True)
+        except Exception:
+            pass
+        try:
+            self.proc.stdin.close()
+        except Exception:
+            pass
+        try:
+            self.proc.wait(timeout=3)
+        except Exception:
+            try:
+                self.proc.kill()
+            except Exception:
+                pass
+        self._started = False
+
+
+class LspManager:
+    """管理多个语言服务器，按文件扩展名路由诊断/智能感知请求"""
+
+    def __init__(self):
+        self.clients: Dict[str, LspClient] = {}      # language_id -> client
+        self._ext_index: Dict[str, str] = {}         # ext -> language_id
+        self._lock = threading.RLock()
+        self._cleaned = False
+        self._workspace_diag: Dict[str, List[Dict]] = {}  # file -> diagnostics（缓存）
+
+    def connect_all(self, servers: Dict[str, Dict[str, Any]],
+                    cwd: Optional[str] = None,
+                    root_uri: Optional[str] = None) -> None:
+        """连接所有已声明且未禁用的语言服务器。
+
+        servers: {language_id 或扩展名: {"command", "args", "env", "cwd", "disabled", "languageId"}}
+        键若是扩展名（以 '.' 开头）则直接登记该扩展名；否则视为 languageId，
+        并把其常见扩展名都映射到它。
+        """
+        for key, spec in servers.items():
+            if not isinstance(spec, dict):
+                continue
+            if spec.get("disabled"):
+                logger.info(f"LSP server '{key}' 已禁用，跳过")
+                continue
+            command = spec.get("command")
+            if not command:
+                logger.warning(f"LSP server '{key}' 缺少 command，跳过")
+                continue
+            args = spec.get("args") or []
+            env = spec.get("env") or {}
+            server_cwd = spec.get("cwd") or cwd
+            explicit_lang = spec.get("languageId") or ""
+            if key.startswith("."):
+                lang_id = explicit_lang or (ext_to_language(key) or key[1:])
+                self._ext_index[key.lower()] = lang_id
+            else:
+                lang_id = key
+                for ext, lid in _EXT_TO_LANG.items():
+                    if lid == lang_id:
+                        self._ext_index[ext] = lang_id
+            if lang_id in self.clients:
+                continue  # 同一语言只起一个 server
+            client = LspClient(lang_id, command, args, env, server_cwd,
+                               root_uri=root_uri, language_id=lang_id)
+            try:
+                client.start()
+                client.initialize()
+                with self._lock:
+                    self.clients[lang_id] = client
+                logger.info(f"已连接 LSP server '{lang_id}' ({command})")
+            except Exception as e:
+                try:
+                    client.shutdown()
+                except Exception:
+                    pass
+                logger.error(f"连接 LSP server '{lang_id}' 失败: {e}")
+
+    def _resolve_client(self, file_path: str) -> Optional[LspClient]:
+        ext = os.path.splitext(file_path)[1].lower()
+        with self._lock:
+            lang_id = self._ext_index.get(ext)
+            if lang_id:
+                return self.clients.get(lang_id)
+        # 退化：按 languageId 推断
+        lid = ext_to_language(file_path)
+        if lid:
+            with self._lock:
+                return self.clients.get(lid)
+        return None
+
+    def _format_diagnostics(self, file_path: str, diags: List[Dict]) -> str:
+        if not diags:
+            return f"✅ {file_path}：无诊断（未发现问题）"
+        lines = [f"🔍 {file_path}：{len(diags)} 条诊断"]
+        for d in diags:
+            rng = d.get("range", {}).get("start", {})
+            ln = rng.get("line", "?")
+            ch = rng.get("character", "?")
+            sev = d.get("severity", 1)
+            sev_map = {1: "错误", 2: "警告", 3: "信息", 4: "提示"}
+            label = sev_map.get(sev, "问题")
+            msg = d.get("message", "")
+            src = d.get("source", "")
+            src_tag = f" [{src}]" if src else ""
+            lines.append(f"  {label}{src_tag} L{ln + 1}:{ch + 1} — {msg}")
+        return "\n".join(lines)
+
+    def diagnostics(self, file_path: str, wait: float = 2.0) -> str:
+        client = self._resolve_client(file_path)
+        if client is None:
+            return (f"⚠️ 没有为 {file_path} 配置 LSP server"
+                    f"（在 config.json 的 lspServers 中添加对应语言）")
+        diags = client.diagnostics(file_path, wait)
+        text = self._format_diagnostics(file_path, diags)
+        # 缓存，供系统提示自动注入
+        with self._lock:
+            self._workspace_diag[file_path] = diags
+        return text
+
+    def hover(self, file_path: str, line: int, character: int) -> str:
+        client = self._resolve_client(file_path)
+        if client is None:
+            return f"⚠️ 没有为 {file_path} 配置 LSP server"
+        try:
+            res = client.hover(file_path, line, character)
+        except Exception as e:
+            return f"❌ hover 失败: {e}"
+        return self._format_hover(res)
+
+    def definition(self, file_path: str, line: int, character: int) -> str:
+        client = self._resolve_client(file_path)
+        if client is None:
+            return f"⚠️ 没有为 {file_path} 配置 LSP server"
+        try:
+            res = client.definition(file_path, line, character)
+        except Exception as e:
+            return f"❌ definition 失败: {e}"
+        return self._format_locations(res, "定义")
+
+    def references(self, file_path: str, line: int, character: int) -> str:
+        client = self._resolve_client(file_path)
+        if client is None:
+            return f"⚠️ 没有为 {file_path} 配置 LSP server"
+        try:
+            res = client.references(file_path, line, character)
+        except Exception as e:
+            return f"❌ references 失败: {e}"
+        return self._format_locations(res, "引用")
+
+    def completion(self, file_path: str, line: int, character: int) -> str:
+        client = self._resolve_client(file_path)
+        if client is None:
+            return f"⚠️ 没有为 {file_path} 配置 LSP server"
+        try:
+            res = client.completion(file_path, line, character)
+        except Exception as e:
+            return f"❌ completion 失败: {e}"
+        return self._format_completion(res)
+
+    @staticmethod
+    def _format_hover(res: Any) -> str:
+        if not res:
+            return "（无 hover 信息）"
+        contents = res.get("contents") if isinstance(res, dict) else None
+        if isinstance(contents, str):
+            return contents
+        if isinstance(contents, dict):
+            return contents.get("value", "") or json.dumps(contents, ensure_ascii=False)
+        if isinstance(contents, list):
+            parts = []
+            for c in contents:
+                if isinstance(c, str):
+                    parts.append(c)
+                elif isinstance(c, dict):
+                    parts.append(c.get("value", ""))
+            return "\n".join(p for p in parts if p)
+        return json.dumps(res, ensure_ascii=False)
+
+    @staticmethod
+    def _format_locations(res: Any, label: str) -> str:
+        if not res:
+            return f"（无{label}）"
+        items = res if isinstance(res, list) else [res]
+        out = [f"{label}位置（{len(items)}）："]
+        for it in items:
+            loc = it.get("location", it) if isinstance(it, dict) else {}
+            uri = loc.get("uri", "")
+            rng = loc.get("range", {}).get("start", {})
+            ln = rng.get("line", "?")
+            ch = rng.get("character", "?")
+            fp = uri.replace("file://", "") if uri.startswith("file://") else uri
+            out.append(f"  {fp} L{ln + 1}:{ch + 1}")
+        return "\n".join(out)
+
+    @staticmethod
+    def _format_completion(res: Any) -> str:
+        if not res:
+            return "（无补全项）"
+        if isinstance(res, dict):
+            items = res.get("items", [])
+        elif isinstance(res, list):
+            items = res
+        else:
+            items = []
+        if not items:
+            return "（无补全项）"
+        out = [f"补全项（{len(items)}）："]
+        for it in items[:30]:
+            label = it.get("label", "")
+            kind = it.get("kind", "")
+            detail = it.get("detail", "")
+            txt = f"  {label}"
+            if detail:
+                txt += f" — {detail}"
+            elif kind:
+                txt += f" (kind={kind})"
+            out.append(txt)
+        return "\n".join(out)
+
+    def get_context(self) -> str:
+        """返回所有已分析文件的诊断汇总，用于注入系统提示"""
+        with self._lock:
+            entries = list(self._workspace_diag.items())
+        if not entries:
+            return ""
+        blocks = []
+        for fp, diags in entries:
+            if diags:
+                blocks.append(self._format_diagnostics(fp, diags))
+        return "\n\n".join(blocks)
+
+    def get_servers(self) -> List[str]:
+        with self._lock:
+            return list(self.clients.keys())
+
+    def cleanup(self) -> None:
+        with self._lock:
+            if self._cleaned:
+                return
+            self._cleaned = True
+            clients = list(self.clients.values())
+            self.clients.clear()
+            self._ext_index.clear()
+            self._workspace_diag.clear()
+        for client in clients:
+            try:
+                client.shutdown()
+            except Exception:
+                pass

mcp_client.py

@@ -0,0 +1,306 @@
+"""
+小狸 MCP 客户端
+===============
+连接外部 MCP (Model Context Protocol) server，将其工具桥接进小狸的统一工具体系。
+
+- transport = stdio（subprocess + 逐行 JSON-RPC 2.0），与项目根 mcp_server.py（暴露侧）对称。
+- 同步 + 线程安全：契合小狸「同步主循环 + ThreadPoolExecutor」模型，不引入官方 mcp SDK
+  （其 asyncio 风格与现有同步调用链不搭）。
+- 单个 server 失败不阻断其他 server，也不阻断主程序启动。
+
+后续可在此文件扩展 SSE / HTTP transport（McpClient 子类即可）。
+"""
+
+import os
+import json
+import time
+import threading
+import subprocess
+import logging
+from typing import Dict, List, Any, Optional, Tuple
+
+logger = logging.getLogger(__name__)
+
+PROTOCOL_VERSION = "2024-11-05"
+
+
+class McpClientError(RuntimeError):
+    """MCP 客户端错误"""
+
+
+class McpClient:
+    """单个 MCP server 的 stdio 客户端（同步、线程安全）"""
+
+    def __init__(self, name: str, command: str, args: Optional[List[str]] = None,
+                 env: Optional[Dict[str, str]] = None, cwd: Optional[str] = None,
+                 timeout: float = 30.0):
+        self.name = name
+        self.command = command
+        self.args = list(args or [])
+        self.env = env or {}
+        self.cwd = cwd
+        self.timeout = timeout
+
+        self.proc = None
+        self._lock = threading.RLock()
+        self._next_id = 1
+        self._pending: Dict[int, Dict[str, Any]] = {}
+        self._reader_thread = None
+        self._started = False
+
+    # ── 生命周期 ──
+    def start(self) -> None:
+        """启动子进程并开启 reader 线程"""
+        if self._started:
+            return
+        full_env = dict(os.environ)
+        full_env.update(self.env)
+        cmd = [self.command] + self.args
+        try:
+            self.proc = subprocess.Popen(
+                cmd,
+                stdin=subprocess.PIPE,
+                stdout=subprocess.PIPE,
+                stderr=subprocess.DEVNULL,
+                cwd=self.cwd,
+                env=full_env,
+                text=True,
+                encoding="utf-8",
+                bufsize=1,
+            )
+        except Exception as e:
+            raise McpClientError(f"启动 MCP server '{self.name}' 失败: {e}")
+
+        self._reader_thread = threading.Thread(target=self._reader_loop, daemon=True)
+        self._reader_thread.start()
+        self._started = True
+
+    def _reader_loop(self):
+        """后台读线程：逐行解析 JSON-RPC 响应，按 id 投递到 _pending 表"""
+        try:
+            for raw in self.proc.stdout:
+                line = raw.strip()
+                if not line:
+                    continue
+                try:
+                    msg = json.loads(line)
+                except json.JSONDecodeError:
+                    # 容错：跳过 server 可能混入 stdout 的非 JSON 噪声
+                    continue
+                msg_id = msg.get("id")
+                if msg_id is None:
+                    # 通知（如进度），忽略
+                    continue
+                with self._lock:
+                    pend = self._pending.pop(msg_id, None)
+                if pend is not None:
+                    if "error" in msg:
+                        pend["error"] = msg["error"]
+                    else:
+                        pend["value"] = msg.get("result")
+                    pend["event"].set()
+        except Exception:
+            pass
+        finally:
+            # 连接断开：唤醒所有等待中的请求，避免永久阻塞
+            with self._lock:
+                pendings = list(self._pending.values())
+                self._pending.clear()
+            for pend in pendings:
+                pend["error"] = {"code": -1, "message": "MCP server 连接断开"}
+                pend["event"].set()
+
+    def _send(self, method: str, params: Any = None,
+              is_notification: bool = False) -> Optional[Any]:
+        """发送一个 JSON-RPC 请求；非通知则等待响应（带超时）"""
+        if not self._started or self.proc is None:
+            raise McpClientError(f"MCP client '{self.name}' 未启动")
+
+        with self._lock:
+            if is_notification:
+                req = {"jsonrpc": "2.0", "method": method, "params": params or {}}
+                pend = None
+                req_id = None
+            else:
+                req_id = self._next_id
+                self._next_id += 1
+                pend = {"event": threading.Event(), "value": None, "error": None}
+                self._pending[req_id] = pend
+                req = {"jsonrpc": "2.0", "id": req_id, "method": method, "params": params or {}}
+
+        try:
+            self.proc.stdin.write(json.dumps(req, ensure_ascii=False) + "\n")
+            self.proc.stdin.flush()
+        except (BrokenPipeError, ValueError, OSError) as e:
+            raise McpClientError(f"向 MCP server '{self.name}' 写入失败: {e}")
+
+        if is_notification:
+            return None
+
+        if not pend["event"].wait(self.timeout):
+            with self._lock:
+                self._pending.pop(req_id, None)
+            raise McpClientError(f"MCP server '{self.name}' 响应超时: {method}")
+        if pend["error"]:
+            err = pend["error"]
+            raise McpClientError(
+                f"MCP server '{self.name}' 错误 [{err.get('code')}]: {err.get('message')}")
+        return pend["value"]
+
+    # ── MCP 协议方法 ──
+    def initialize(self) -> Dict[str, Any]:
+        """握手：协商协议版本并通知初始化完成"""
+        result = self._send("initialize", {
+            "protocolVersion": PROTOCOL_VERSION,
+            "capabilities": {},
+            "clientInfo": {"name": "xiaoli-cli", "version": "8.0.3"},
+        })
+        # 通知 server 初始化完成（无响应，无需等待）
+        self._send("notifications/initialized", {}, is_notification=True)
+        return result or {}
+
+    def list_tools(self) -> List[Dict[str, Any]]:
+        """列出 server 提供的工具定义"""
+        result = self._send("tools/list", {})
+        if isinstance(result, dict):
+            return result.get("tools", [])
+        return []
+
+    def call_tool(self, tool_name: str, arguments: Dict[str, Any]) -> str:
+        """调用一个工具，返回文本结果"""
+        result = self._send("tools/call", {"name": tool_name, "arguments": arguments or {}})
+        return self._extract_content(result)
+
+    @staticmethod
+    def _extract_content(result: Any) -> str:
+        """从 tools/call 的 result 提取文本结果"""
+        if result is None:
+            return ""
+        if isinstance(result, str):
+            return result
+        if isinstance(result, dict):
+            content = result.get("content")
+            if isinstance(content, list):
+                parts = []
+                for item in content:
+                    if isinstance(item, dict):
+                        if item.get("type") == "text" or "text" in item:
+                            parts.append(str(item.get("text", "")))
+                return "\n".join(p for p in parts if p)
+            if isinstance(result.get("content"), str):
+                return result["content"]
+            # 部分 server 直接把结果放 result 顶层
+            if "isError" in result and result.get("isError"):
+                return f"[工具错误] {json.dumps(result, ensure_ascii=False)}"
+            return json.dumps(result, ensure_ascii=False)
+        return str(result)
+
+    def ping(self) -> bool:
+        try:
+            self._send("ping", {})
+            return True
+        except Exception:
+            return False
+
+    def stop(self) -> None:
+        """优雅关闭子进程，失败则强杀"""
+        if self.proc is None:
+            return
+        try:
+            self._send("shutdown", {})
+        except Exception:
+            pass
+        try:
+            self._send("exit", {}, is_notification=True)
+        except Exception:
+            pass
+        try:
+            self.proc.stdin.close()
+        except Exception:
+            pass
+        try:
+            self.proc.wait(timeout=3)
+        except Exception:
+            try:
+                self.proc.kill()
+            except Exception:
+                pass
+        self._started = False
+
+
+class McpClientManager:
+    """管理多个外部 MCP server 客户端（按配置并发连接、聚合工具）"""
+
+    def __init__(self):
+        self.clients: Dict[str, McpClient] = {}
+        self._tool_index: List[Tuple[str, Dict[str, Any]]] = []  # (server_name, tool_def)
+        self._lock = threading.RLock()
+        self._cleaned = False
+
+    def connect_all(self, servers: Dict[str, Dict[str, Any]],
+                    cwd: Optional[str] = None) -> None:
+        """连接所有已声明且未禁用的 server。单个失败不阻断其他。
+
+        servers: {name: {"command", "args", "env", "cwd", "disabled"}}
+        """
+        for name, spec in servers.items():
+            if not isinstance(spec, dict):
+                continue
+            if spec.get("disabled"):
+                logger.info(f"MCP server '{name}' 已禁用，跳过")
+                continue
+            command = spec.get("command")
+            if not command:
+                logger.warning(f"MCP server '{name}' 缺少 command，跳过")
+                continue
+            args = spec.get("args") or []
+            env = spec.get("env") or {}
+            server_cwd = spec.get("cwd") or cwd
+            client = McpClient(name, command, args, env, server_cwd)
+            try:
+                client.start()
+                client.initialize()
+                tools = client.list_tools()
+                with self._lock:
+                    self.clients[name] = client
+                    for tool in tools:
+                        self._tool_index.append((name, tool))
+                logger.info(f"已连接 MCP server '{name}': {len(tools)} 个工具")
+            except Exception as e:
+                try:
+                    client.stop()
+                except Exception:
+                    pass
+                logger.error(f"连接 MCP server '{name}' 失败: {e}")
+                # 不向上抛，继续下一个 server
+
+    def get_all_tools(self) -> List[Tuple[str, Dict[str, Any]]]:
+        """返回 [(server_name, tool_def), ...]"""
+        with self._lock:
+            return list(self._tool_index)
+
+    def get_servers(self) -> List[str]:
+        with self._lock:
+            return list(self.clients.keys())
+
+    def call_tool(self, server_name: str, tool_name: str, arguments: Dict[str, Any]) -> str:
+        with self._lock:
+            client = self.clients.get(server_name)
+        if client is None:
+            raise McpClientError(f"未连接的 MCP server: {server_name}")
+        return client.call_tool(tool_name, arguments)
+
+    def cleanup(self) -> None:
+        """幂等：终止所有子进程。每个桥接工具的 shutdown 都会调一次，故必须可重入。"""
+        with self._lock:
+            if self._cleaned:
+                return
+            self._cleaned = True
+            clients = list(self.clients.values())
+            self.clients.clear()
+            self._tool_index.clear()
+        for client in clients:
+            try:
+                client.stop()
+            except Exception:
+                pass

plugins/code_editor.py

@@ -599,6 +599,88 @@ else:
         else:
             return f"\n{result}"
 
+    def _auto_lsp_check(self, filepath: str) -> str:
+        """写/改文件后自动让已连接的语言服务器(pylsp 等)重新诊断该文件。
+
+        诊断结果会被缓存进 LSP 管理器的 _workspace_diag，由 get_lsp_context()
+        在下一轮静默注入模型上下文；同时仅在确有问题时给一行可见提示。
+        无对应 server / 文件干净时静默跳过，绝不影响主流程。
+        """
+        manager = None
+        try:
+            cli = self.cli
+            utm = getattr(cli, 'liugin_manager', None)
+            if utm is not None:
+                manager = getattr(utm, '_lsp_manager', None)
+        except Exception:
+            manager = None
+        if manager is None:
+            return ""
+        try:
+            diag_text = manager.diagnostics(filepath, wait=1.5)
+        except Exception:
+            return ""
+        if not diag_text or "没有为" in diag_text:
+            return ""  # 该扩展名未配置 LSP server，跳过
+        if "无诊断" in diag_text:
+            return ""  # 干净，无需呈现（缓存也为空）
+        first = diag_text.splitlines()[0] if diag_text else ""
+        return f"\n  LSP 诊断: {first}"
+
+    def _load_py_detect(self):
+        """延迟加载 py_detect 核心（plugins/py_detect.py），失败返回 None 并缓存"""
+        cached = getattr(self, "_py_detect_mod", "unset")
+        if cached != "unset":
+            return cached
+        try:
+            import importlib.util
+            plugin_path = os.path.join(
+                os.path.dirname(os.path.abspath(__file__)), "py_detect.py")
+            if not os.path.exists(plugin_path):
+                self._py_detect_mod = None
+                return None
+            spec = importlib.util.spec_from_file_location("py_detect_core", plugin_path)
+            mod = importlib.util.module_from_spec(spec)
+            spec.loader.exec_module(mod)
+            self._py_detect_mod = mod
+            return mod
+        except Exception:
+            self._py_detect_mod = None
+            return None
+
+    def _auto_py_detect(self, filepath: str) -> str:
+        """写/改 .py 文件后自动智能检测（py_detect，纯本地零依赖）。
+
+        仅在 .py 文件且检测到问题时输出（截断为前 6 条，避免刷屏），
+        干净或非 py 文件静默返回空串——与 _auto_lsp_check 同风格。
+        """
+        if not filepath.lower().endswith(".py"):
+            return ""
+        mod = self._load_py_detect()
+        if mod is None:
+            return ""
+        try:
+            with open(filepath, "r", encoding="utf-8", errors="replace") as f:
+                source = f.read()
+            issues = mod.analyze(source, filepath)
+            if not issues:
+                return ""
+            order = {"错误": 0, "警告": 1, "提示": 2}
+            issues = sorted(issues, key=lambda d: (order.get(d["severity"], 9), d["line"], d["col"]))
+            n = len(issues)
+            n_e = sum(1 for d in issues if d["severity"] == "错误")
+            n_w = sum(1 for d in issues if d["severity"] == "警告")
+            n_i = sum(1 for d in issues if d["severity"] == "提示")
+            lines = [f"\n  [py_detect] 检测到 {n} 个问题 (错误 {n_e} · 警告 {n_w} · 提示 {n_i}):"]
+            for d in issues[:6]:
+                lines.append(f"    [{d['code']}] {d['severity']} @{d['line']}:{d['col']} {d['title']}"
+                             + (f" ({d['extra']})" if d.get("extra") else ""))
+            if n > 6:
+                lines.append(f"    … 还有 {n - 6} 条，py_detect check {filepath} 查看全部")
+            return "\n".join(lines)
+        except Exception:
+            return ""
+
     # ══════════════════════════════════════
     #  Diff 弹窗系统
     # ══════════════════════════════════════
@@ -975,7 +1057,7 @@ else:
 
         self._spawn_diff_popup(fp, before_content, new_content)
         result = f" 已编辑: {fp}\n\n{self._make_diff(old_text, new_text)}"
-        result += self._auto_syntax_check(fp)
+        result += self._auto_syntax_check(fp) + self._auto_lsp_check(fp) + self._auto_py_detect(fp)
         return result
 
     def _op_multi_edit(self, path: str, rest: str) -> str:
@@ -1013,7 +1095,7 @@ else:
             f.write(content)
         self._spawn_diff_popup(fp, before_content, content)
         result = f" 批量编辑: {fp}\n" + "\n".join(results)
-        result += self._auto_syntax_check(fp)
+        result += self._auto_syntax_check(fp) + self._auto_lsp_check(fp) + self._auto_py_detect(fp)
         return result
 
     def _op_insert(self, path: str, rest: str) -> str:
@@ -1046,7 +1128,7 @@ else:
         after_content = ''.join(lines)
         self._spawn_diff_popup(fp, before_content, after_content)
         result = f" 已在第 {line_num} 行插入 {len(insert_lines)} 行"
-        result += self._auto_syntax_check(fp)
+        result += self._auto_syntax_check(fp) + self._auto_lsp_check(fp) + self._auto_py_detect(fp)
         return result
 
     def _op_delete_lines(self, path: str, rest: str) -> str:
@@ -1079,7 +1161,7 @@ else:
         after_content = ''.join(lines)
         self._spawn_diff_popup(fp, before_content, after_content)
         result = f" 已删除第 {start}-{end} 行 ({len(deleted)} 行)"
-        result += self._auto_syntax_check(fp)
+        result += self._auto_syntax_check(fp) + self._auto_lsp_check(fp) + self._auto_py_detect(fp)
         return result
 
     def _op_create(self, path: str, rest: str) -> str:
@@ -1095,7 +1177,7 @@ else:
             f.write(rest or "")
         self._spawn_diff_popup(fp, "", rest or "")
         result = f" 已创建: {fp} ({len(rest or '')} 字符)"
-        result += self._auto_syntax_check(fp)
+        result += self._auto_syntax_check(fp) + self._auto_lsp_check(fp) + self._auto_py_detect(fp)
         return result
 
     def _op_write(self, path: str, rest: str) -> str:
@@ -1110,7 +1192,7 @@ else:
             f.write(rest)
         self._spawn_diff_popup(fp, before_content, rest)
         result = f" 已写入: {fp} ({len(rest)} 字符)"
-        result += self._auto_syntax_check(fp)
+        result += self._auto_syntax_check(fp) + self._auto_lsp_check(fp) + self._auto_py_detect(fp)
         return result
 
     def _op_append(self, path: str, rest: str) -> str:
@@ -1125,7 +1207,7 @@ else:
         after_content = self._read_file_safe(fp)
         self._spawn_diff_popup(fp, before_content, after_content)
         result = f" 已追加: {fp} ({len(rest)} 字符)"
-        result += self._auto_syntax_check(fp)
+        result += self._auto_syntax_check(fp) + self._auto_lsp_check(fp) + self._auto_py_detect(fp)
         return result
 
     # ══════════════════════════════════════

plugins/file_manager.py

@@ -3,6 +3,7 @@
 聚焦文件系统操作，代码编辑交给 code_editor
 """
 import os
+import re
 import json
 import shutil
 import sys
@@ -10,6 +11,30 @@ from datetime import datetime
 from pathlib import Path
 from typing import List
 
+
+# ── 跨平台路径归一化 ──────────────────────────────────────────────
+# 模型常给出 Linux/Mac 风格绝对路径（如 /home/<user>/Desktop、
+# /Users/<user>/Desktop、/root/Desktop）。在 Windows 下 os.path.abspath 会
+# 错误地拼成 C:\home\user\Desktop。这里把这类路径映射到当前用户家目录
+# 对应位置；Linux/Mac 下这些路径本身合法，原样保留。
+_HOME_OR_USER_RE = re.compile(r'^/(?:home|Users)/[^/]+(/.*)?$')
+_ROOT_RE = re.compile(r'^/root(/.*)?$')
+
+
+def _resolve_cross_platform_path(path: str) -> str:
+    """把模型给出的路径归一化到当前平台后再转绝对路径。"""
+    if os.name == 'nt' and path:
+        p = path.replace('\\', '/')
+        m = _HOME_OR_USER_RE.match(p)
+        if m:
+            rest = (m.group(1) or '').lstrip('/')
+            return os.path.join(os.path.expanduser('~'), rest)
+        m = _ROOT_RE.match(p)
+        if m:
+            rest = (m.group(1) or '').lstrip('/')
+            return os.path.join(os.path.expanduser('~'), rest)
+    return os.path.abspath(path)
+
 # 添加项目根目录到 path
 sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 from xcli_core.tool_result import ToolResult, ErrorCode
@@ -38,6 +63,34 @@ class Liugin:
     def set_cli(self, cli):
         self.cli = cli
 
+    def _auto_lsp_check(self, filepath: str) -> str:
+        """写/改文件后自动让已连接的语言服务器(pylsp 等)重新诊断该文件。
+
+        诊断结果缓存进 LSP 管理器的 _workspace_diag，由 get_lsp_context()
+        在下一轮静默注入模型上下文；仅在有问题时返回一行提示。
+        无对应 server / 文件干净时静默跳过。
+        """
+        manager = None
+        try:
+            cli = self.cli
+            utm = getattr(cli, 'liugin_manager', None)
+            if utm is not None:
+                manager = getattr(utm, '_lsp_manager', None)
+        except Exception:
+            manager = None
+        if manager is None:
+            return ""
+        try:
+            diag_text = manager.diagnostics(filepath, wait=1.5)
+        except Exception:
+            return ""
+        if not diag_text or "没有为" in diag_text:
+            return ""
+        if "无诊断" in diag_text:
+            return ""
+        first = diag_text.splitlines()[0] if diag_text else ""
+        return f"\n  LSP 诊断: {first}"
+
     def get_tool_info(self):
         return {
             "name": "file_manager",
@@ -130,7 +183,7 @@ class Liugin:
                 path = args[i]
             i += 1
 
-        path = os.path.abspath(path)
+        path = _resolve_cross_platform_path(path)
         if not os.path.isdir(path):
             return ToolResult.not_found(f"目录: {path}")
 
@@ -165,7 +218,7 @@ class Liugin:
     def _op_read(self, args: List[str]):
         if not args:
             return ToolResult.missing("文件路径")
-        file_path = os.path.abspath(args[0])
+        file_path = _resolve_cross_platform_path(args[0])
         max_lines = None
         i = 1
         while i < len(args):
@@ -207,7 +260,7 @@ class Liugin:
     def _op_write(self, args: List[str]):
         if len(args) < 2:
             return ToolResult.fail("格式: write <文件> -c <内容>", ErrorCode.MISSING_ARGS)
-        fp = os.path.abspath(args[0])
+        fp = _resolve_cross_platform_path(args[0])
         content = None
         i = 1
         while i < len(args):
@@ -219,12 +272,16 @@ class Liugin:
         os.makedirs(os.path.dirname(fp) or ".", exist_ok=True)
         with open(fp, 'w', encoding='utf-8') as f:
             f.write(content)
-        return ToolResult.ok(f"已写入: {fp} ({len(content)} 字符)", tool_name="file_manager")
+        msg = f"已写入: {fp} ({len(content)} 字符)"
+        lsp_hint = self._auto_lsp_check(fp)
+        if lsp_hint:
+            msg += lsp_hint
+        return ToolResult.ok(msg, tool_name="file_manager")
 
     def _op_append(self, args: List[str]):
         if len(args) < 2:
             return ToolResult.fail("格式: append <文件> -c <内容>", ErrorCode.MISSING_ARGS)
-        fp = os.path.abspath(args[0])
+        fp = _resolve_cross_platform_path(args[0])
         content = None
         i = 1
         while i < len(args):
@@ -237,12 +294,16 @@ class Liugin:
             return ToolResult.not_found(f"文件: {fp}")
         with open(fp, 'a', encoding='utf-8') as f:
             f.write(content)
-        return ToolResult.ok(f"已追加: {fp}", tool_name="file_manager")
+        msg = f"已追加: {fp}"
+        lsp_hint = self._auto_lsp_check(fp)
+        if lsp_hint:
+            msg += lsp_hint
+        return ToolResult.ok(msg, tool_name="file_manager")
 
     def _op_copy(self, args: List[str]):
         if len(args) < 2:
             return ToolResult.missing("源和目标路径")
-        src, dst = os.path.abspath(args[0]), os.path.abspath(args[1])
+        src, dst = _resolve_cross_platform_path(args[0]), _resolve_cross_platform_path(args[1])
         if not os.path.exists(src):
             return ToolResult.not_found(f"源: {src}")
         os.makedirs(os.path.dirname(dst) or ".", exist_ok=True)
@@ -255,7 +316,7 @@ class Liugin:
     def _op_move(self, args: List[str]):
         if len(args) < 2:
             return ToolResult.missing("源和目标路径")
-        src, dst = os.path.abspath(args[0]), os.path.abspath(args[1])
+        src, dst = _resolve_cross_platform_path(args[0]), _resolve_cross_platform_path(args[1])
         if not os.path.exists(src):
             return ToolResult.not_found(f"源: {src}")
         os.makedirs(os.path.dirname(dst) or ".", exist_ok=True)
@@ -265,7 +326,7 @@ class Liugin:
     def _op_delete(self, args: List[str]):
         if not args:
             return ToolResult.missing("路径")
-        path = os.path.abspath(args[0])
+        path = _resolve_cross_platform_path(args[0])
         recursive = "-r" in args
         if not os.path.exists(path):
             return ToolResult.not_found(f"路径: {path}")
@@ -285,7 +346,7 @@ class Liugin:
     def _op_search(self, args: List[str]):
         if not args:
             return ToolResult.missing("搜索目录")
-        directory = os.path.abspath(args[0])
+        directory = _resolve_cross_platform_path(args[0])
         pattern, content_search = None, None
         i = 1
         while i < len(args):
@@ -327,7 +388,7 @@ class Liugin:
     def _op_info(self, args: List[str]):
         if not args:
             return ToolResult.missing("路径")
-        path = os.path.abspath(args[0])
+        path = _resolve_cross_platform_path(args[0])
         if not os.path.exists(path):
             return ToolResult.not_found(f"路径: {path}")
 
@@ -347,7 +408,7 @@ class Liugin:
     def _op_mkdir(self, args: List[str]):
         if not args:
             return ToolResult.missing("目录路径")
-        path = os.path.abspath(args[0])
+        path = _resolve_cross_platform_path(args[0])
         if os.path.exists(path):
             return ToolResult.fail(f"已存在: {path}", ErrorCode.INVALID_ARGS)
         os.makedirs(path, exist_ok=True)

plugins/operit_scripts/beeimg_image_uploader.py

@@ -0,0 +1,176 @@
+# METADATA
+# {
+#     "name": "beeimg_image_uploader_v2",
+#
+#     "display_name": {
+#         "zh": "Beeimg 图片上传器 V2",
+#         "en": "Beeimg Image Uploader V2"
+#     },
+#     "description": {
+#         "zh": "BeeIMG（https://beeimg.com/）工具将本地图片上传到图床并返回图片url，配合图片生成工具实现图生图（图生图务必开启）。",
+#         "en": "Upload a local image to BeeIMG (https://beeimg.com/) and return a public image URL. Useful for image-to-image workflows (make sure img2img is enabled)."
+#     },
+#     "env": [
+#         "BEEIMG_API_KEY"
+#     ],
+#     "category": "Media",
+#     "tools": [
+#         {
+#             "name": "upload_image",
+#             "description": {
+#                 "zh": "使用 multipart 上传将本地图片上传到 BeeIMG 图床并返回图片 url。",
+#                 "en": "Upload a local image to BeeIMG via multipart upload and return the image URL."
+#             },
+#             "parameters": [
+#                 { "name": "file_path", "description": { "zh": "要上传的图片文件绝对路径 (建议使用 /sdcard/ 开头的完整路径)。", "en": "Absolute path of the image file to upload (recommended: full path starting with /sdcard/)." }, "type": "string", "required": true },
+#                 { "name": "album_id", "description": { "zh": "相册ID (可选)。", "en": "Album ID (optional)." }, "type": "string", "required": false },
+#                 { "name": "privacy", "description": { "zh": "隐私设置，'public' 或 'private' (可选)。", "en": "Privacy setting: 'public' or 'private' (optional)." }, "type": "string", "required": false }
+#             ]
+#         }
+#     ]
+# }
+
+import json
+import traceback
+
+API_ENDPOINT = "https://beeimg.com/api/upload/file/json/"
+
+
+def isRecord(value):
+    return isinstance(value, dict)
+
+
+def getErrorMessage(error):
+    if isinstance(error, Exception):
+        return str(error)
+    return str(error)
+
+
+def getErrorStack(error):
+    if isinstance(error, Exception):
+        return traceback.format_exc()
+    return None
+
+
+def getApiKey():
+    return getEnv("BEEIMG_API_KEY") or ""
+
+
+def guessMimeTypeFromPath(filePath):
+    lower = filePath.lower()
+    if lower.endswith(".png"):
+        return "image/png"
+    if lower.endswith(".jpg") or lower.endswith(".jpeg"):
+        return "image/jpeg"
+    if lower.endswith(".webp"):
+        return "image/webp"
+    if lower.endswith(".gif"):
+        return "image/gif"
+    return "application/octet-stream"
+
+
+def safeJsonParseLoose(text):
+    trimmed = (text or "").strip()
+    if not trimmed:
+        return None
+    try:
+        return json.loads(trimmed)
+    except Exception as e:
+        start = trimmed.find("{")
+        end = trimmed.rfind("}")
+        if start != -1 and end != -1 and end > start:
+            return json.loads(trimmed[start:end + 1])
+        raise e
+
+
+async def upload_image(params):
+    file_path = params.get("file_path") if params else None
+    album_id = params.get("album_id") if params else None
+    privacy = params.get("privacy") if params else None
+
+    if not file_path or len(str(file_path).strip()) == 0:
+        raise Exception("参数 file_path 不能为空。")
+
+    # 1. 检查文件是否存在
+    fileExists = await Tools.Files.exists(file_path)
+    if not fileExists["exists"]:
+        raise Exception(f"文件未找到: {file_path} (请尝试使用 /sdcard/ 开头的绝对路径)")
+
+    # 2. 准备参数
+    apiKey = getApiKey()
+    if not apiKey:
+        raise Exception("BEEIMG_API_KEY 未配置，请在环境变量中设置 BeeIMG 的 API Key。")
+
+    form_data = {
+        "apikey": apiKey
+    }
+    if album_id:
+        form_data["albumid"] = str(album_id)
+    if privacy:
+        form_data["privacy"] = str(privacy)
+
+    # 3. 发起 multipart 上传
+    console.log("正在执行上传...")
+
+    resp = await Tools.Net.uploadFile({
+        "url": API_ENDPOINT,
+        "method": "POST",
+        "headers": {
+            # 尽量模拟浏览器 UA，降低被风控概率
+            "User-Agent": "Mozilla/5.0 (Linux; Android 10; K) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/114.0.0.0 Mobile Safari/537.36"
+        },
+        "form_data": form_data,
+        "files": [
+            {
+                "field_name": "file",
+                "file_path": file_path,
+                "content_type": guessMimeTypeFromPath(file_path)
+            }
+        ]
+    })
+
+    if resp["statusCode"] < 200 or resp["statusCode"] >= 300:
+        raise Exception(f"上传请求失败 (HTTP {resp['statusCode']})。可能是网络问题或 IP 被屏蔽。\n响应: {resp['content']}")
+
+    responseText = (resp.get("content") or "").strip()
+    if not responseText:
+        raise Exception("上传失败：服务器返回了空内容。请检查网络连接或尝试关闭 VPN。")
+
+    try:
+        jsonResponse = safeJsonParseLoose(responseText)
+    except Exception as e:
+        raise Exception(f"API 响应解析失败。服务器返回内容不是 JSON:\n{responseText[:200]}...")
+
+    files = jsonResponse["files"] if isRecord(jsonResponse) and isRecord(jsonResponse.get("files")) else None
+    ok = bool(files) and (files.get("status") == "Success" or files.get("code") == "200" or files.get("code") == 200)
+    url = files.get("url") if files else None
+
+    if ok and (isinstance(url, str) or isinstance(url, (int, float))) and len(str(url).strip()) > 0:
+        return {
+            "url": str(url),
+            "thumbnail_url": str(files["thumbnail_url"]) if files and files.get("thumbnail_url") else None,
+            "page_url": str(files["view_url"]) if files and files.get("view_url") else None,
+            "details": f"上传成功! URL: {str(url)}"
+        }
+    else:
+        if files:
+            errMsg = files.get("status") or files.get("message") or json.dumps(files, ensure_ascii=False)
+        else:
+            errMsg = "未知错误"
+        raise Exception(f"BeeIMG API 报错: {errMsg}")
+
+
+async def wrap(func, params):
+    try:
+        result = await func(params)
+        complete({"success": True, "message": "图片上传成功", "data": result})
+    except Exception as error:
+        console.error(f"Error: {getErrorMessage(error)}")
+        complete({"success": False, "message": getErrorMessage(error), "error_stack": getErrorStack(error)})
+
+
+async def upload_image_wrapper(params):
+    await wrap(upload_image, params)
+
+
+exports.upload_image = upload_image_wrapper

plugins/operit_scripts/file_manager.py

@@ -0,0 +1,114 @@
+# METADATA
+# {
+#     "name": "file_manager",
+#     "display_name": {
+#         "zh": "Android 文件管理器",
+#         "en": "Android File Manager"
+#     },
+#     "description": {
+#         "zh": "通过 Environment 获取外部存储目录，支持列目录、读文本、写文本（真实文件系统操作）。",
+#         "en": "Use Environment to get external storage dir; list dir, read/write text files on real filesystem."
+#     },
+#     "category": "Files",
+#     "tools": [
+#         {
+#             "name": "list_dir",
+#             "description": {
+#                 "zh": "列出目录内容",
+#                 "en": "List directory contents"
+#             },
+#             "parameters": [
+#                 { "name": "path", "description": { "zh": "目录路径（默认外部存储根）", "en": "Dir path (default external storage root)" }, "type": "string", "required": false }
+#             ]
+#         },
+#         {
+#             "name": "write_text",
+#             "description": {
+#                 "zh": "写入文本文件",
+#                 "en": "Write text file"
+#             },
+#             "parameters": [
+#                 { "name": "file_path", "description": { "zh": "文件绝对路径", "en": "Absolute file path" }, "type": "string", "required": true },
+#                 { "name": "content", "description": { "zh": "文本内容", "en": "Text content" }, "type": "string", "required": true }
+#             ]
+#         },
+#         {
+#             "name": "read_text",
+#             "description": {
+#                 "zh": "读取文本文件",
+#                 "en": "Read text file"
+#             },
+#             "parameters": [
+#                 { "name": "file_path", "description": { "zh": "文件绝对路径", "en": "Absolute file path" }, "type": "string", "required": true }
+#             ]
+#         }
+#     ]
+# }
+
+TAG = "FileManager"
+
+
+def _resolve(path):
+    """解析路径：默认落到外部存储根"""
+    if path and path.strip():
+        return os.path.abspath(path.strip())
+    return Environment.getExternalStorageDirectory()
+
+
+async def list_dir(params):
+    path = _resolve((params or {}).get("path", ""))
+    if not os.path.isdir(path):
+        raise Exception(f"目录不存在: {path}")
+    entries = []
+    for name in sorted(os.listdir(path)):
+        full = os.path.join(path, name)
+        entries.append({
+            "name": name,
+            "type": "dir" if os.path.isdir(full) else "file",
+            "size": os.path.getsize(full) if os.path.isfile(full) else None,
+        })
+    Log.i(TAG, f"列出 {path}: {len(entries)} 项")
+    result = {"path": path, "count": len(entries), "entries": entries[:100]}
+    complete({"success": True, "message": f"已列出 {len(entries)} 项", "data": result})
+    return result
+
+
+async def write_text(params):
+    file_path = os.path.abspath((params or {}).get("file_path", ""))
+    content = (params or {}).get("content", "")
+    if not file_path:
+        raise Exception("file_path 不能为空")
+    os.makedirs(os.path.dirname(file_path) or ".", exist_ok=True)
+    with open(file_path, "w", encoding="utf-8") as fh:
+        fh.write(content)
+    size = os.path.getsize(file_path)
+    Log.w(TAG, f"已写入 {file_path} ({size} 字节)")
+    result = {"file_path": file_path, "bytes": size}
+    complete({"success": True, "message": f"写入成功 ({size} 字节)", "data": result})
+    return result
+
+
+async def read_text(params):
+    file_path = os.path.abspath((params or {}).get("file_path", ""))
+    if not file_path:
+        raise Exception("file_path 不能为空")
+    if not os.path.isfile(file_path):
+        raise Exception(f"文件不存在: {file_path}")
+    with open(file_path, "r", encoding="utf-8", errors="replace") as fh:
+        content = fh.read()
+    result = {"file_path": file_path, "bytes": len(content.encode("utf-8")), "content": content[:20000]}
+    complete({"success": True, "message": f"读取成功 ({len(content)} 字符)", "data": result})
+    return result
+
+
+async def _dispatch(params):
+    action = (params or {}).get("action", "list_dir")
+    func = {"list_dir": list_dir, "write_text": write_text, "read_text": read_text}.get(action)
+    if not func:
+        raise Exception(f"未知操作: {action}（支持 list_dir/write_text/read_text）")
+    return await func(params)
+
+
+exports.list_dir = list_dir
+exports.write_text = write_text
+exports.read_text = read_text

plugins/operit_scripts/intent_builder.py

@@ -0,0 +1,76 @@
+# METADATA
+# {
+#     "name": "intent_builder",
+#     "display_name": {
+#         "zh": "Intent 意图构建器",
+#         "en": "Intent Builder"
+#     },
+#     "description": {
+#         "zh": "用 android.content.Intent 构建意图：ACTION_VIEW 打开 URL、ACTION_SEND 分享文本，解析 Uri，并用 Log 输出。",
+#         "en": "Build intents with android.content.Intent: ACTION_VIEW for URLs, ACTION_SEND for sharing text, parse Uri, log via Log."
+#     },
+#     "category": "System",
+#     "tools": [
+#         {
+#             "name": "build_view_intent",
+#             "description": {
+#                 "zh": "构建 ACTION_VIEW 意图（打开 URL）",
+#                 "en": "Build ACTION_VIEW intent (open URL)"
+#             },
+#             "parameters": [
+#                 { "name": "url", "description": { "zh": "要打开的网址", "en": "URL to open" }, "type": "string", "required": true }
+#             ]
+#         },
+#         {
+#             "name": "build_share_intent",
+#             "description": {
+#                 "zh": "构建 ACTION_SEND 分享意图",
+#                 "en": "Build ACTION_SEND share intent"
+#             },
+#             "parameters": [
+#                 { "name": "text", "description": { "zh": "分享的文本", "en": "Text to share" }, "type": "string", "required": true }
+#             ]
+#         }
+#     ]
+# }
+
+TAG = "IntentBuilder"
+
+
+async def build_view_intent(params):
+    url = (params or {}).get("url", "")
+    if not url:
+        raise Exception("url 不能为空")
+    intent = Intent(Intent.ACTION_VIEW)
+    intent.setData(Uri.parse(url))
+    Log.i(TAG, f"ACTION_VIEW -> {url}")
+    result = {
+        "action": intent.getAction(),
+        "data": intent.getData(),
+        "flags": intent.getFlags(),
+        "resolved": intent.toString(),
+    }
+    complete({"success": True, "message": "意图构建成功（桌面环境不实际拉起）", "data": result})
+    return result
+
+
+async def build_share_intent(params):
+    text = (params or {}).get("text", "")
+    if not text:
+        raise Exception("text 不能为空")
+    intent = Intent(Intent.ACTION_SEND)
+    intent.setType("text/plain")
+    intent.putExtra(Intent.EXTRA_TEXT, text)
+    Log.i(TAG, f"ACTION_SEND: {text[:40]}")
+    result = {
+        "action": intent.getAction(),
+        "type": intent.getType(),
+        "extras": intent.getExtras(),
+        "resolved": intent.toString(),
+    }
+    complete({"success": True, "message": "分享意图构建成功", "data": result})
+    return result
+
+
+exports.build_view_intent = build_view_intent
+exports.build_share_intent = build_share_intent

plugins/operit_scripts/market_crossref.py

@@ -0,0 +1,205 @@
+# METADATA
+# {
+#   "name": "crossref",
+#   "display_name": { "zh": "Crossref 学术文献查询", "en": "Crossref Academic Literature Search" },
+#   "description": { "zh": "Crossref 学术文献查询（DOI/关键词/作者/标题/ISSN）", "en": "Crossref scholarly literature search." },
+#   "category": "Search",
+#   "tools": [
+#     { "name": "search_by_doi", "description": { "zh": "通过 DOI 查询文章", "en": "Query by DOI" } },
+#     { "name": "search_by_keyword", "description": { "zh": "通过关键词搜索", "en": "Search by keyword" } },
+#     { "name": "search_by_author", "description": { "zh": "通过作者搜索", "en": "Search by author" } },
+#     { "name": "search_by_title", "description": { "zh": "通过标题搜索", "en": "Search by title" } },
+#     { "name": "search_by_issn", "description": { "zh": "通过 ISSN 查询", "en": "Search by ISSN" } }
+#   ]
+# }
+import json as _json
+from urllib.parse import quote_plus
+
+HEADERS = {'User-Agent': 'Operit/1.0 (mailto:support@example.com)'}
+BASE_URL = "https://api.crossref.org"
+DEFAULT_ROWS = 10
+MAX_ROWS = 100
+
+
+def getErrorMessage(error):
+    return str(error) if error else "unknown"
+
+
+def getErrorStack(error):
+    import traceback
+    return traceback.format_exc()
+
+
+def format_authors(authors):
+    if not authors:
+        return "N/A"
+    return ", ".join(
+        f"{a.get('given', '')} {a.get('family', '')}".strip()
+        for a in authors[:5] if a.get('given') or a.get('family')
+    ) or "N/A"
+
+
+def format_date(date_parts):
+    if not date_parts or not date_parts[0]:
+        return "N/A"
+    p = date_parts[0]
+    if len(p) == 1:
+        return str(p[0])
+    if len(p) == 2:
+        return f"{p[0]}-{str(p[1]).zfill(2)}"
+    if len(p) == 3:
+        return f"{p[0]}-{str(p[1]).zfill(2)}-{str(p[2]).zfill(2)}"
+    return "N/A"
+
+
+def format_article(item, index=None):
+    lines = []
+    if index is not None:
+        lines.append(f"\n=== Article {index + 1} ===")
+    title = item.get("title", ["No Title"])[0] if item.get("title") else "No Title"
+    lines.append(f"Title: {title}")
+    if item.get("DOI"):
+        lines.append(f"DOI: {item['DOI']}")
+        lines.append(f"URL: https://doi.org/{item['DOI']}")
+    lines.append(f"Authors: {format_authors(item.get('author'))}")
+    dp = (item.get("published") or item.get("published-print") or item.get("published-online") or {}).get("date-parts")
+    lines.append(f"Published: {format_date(dp)}")
+    if item.get("container-title"):
+        lines.append(f"Journal/Conference: {item['container-title'][0]}")
+    if item.get("ISSN"):
+        lines.append(f"ISSN: {', '.join(item['ISSN'])}")
+    if item.get("publisher"):
+        lines.append(f"Publisher: {item['publisher']}")
+    if item.get("type"):
+        lines.append(f"Type: {item['type']}")
+    if item.get("is-referenced-by-count") is not None:
+        lines.append(f"Citations: {item['is-referenced-by-count']}")
+    if item.get("abstract"):
+        ab = __import__("re").sub(r"<[^>]*>", "", item["abstract"])
+        lines.append(f"Abstract: {ab[:500]}{'...' if len(ab) > 500 else ''}")
+    return "\n".join(lines)
+
+
+async def _get_json(url):
+    client = OkHttp.newClient()
+    resp = await client.newRequest().url(url).method("GET").headers(HEADERS).build().execute()
+    if not resp.isSuccessful():
+        raise RuntimeError(f"HTTP {resp.statusCode} - {resp.statusMessage}")
+    return resp.json()
+
+
+async def search_by_doi(params):
+    doi = (params.get("doi") or "").strip()
+    if not doi:
+        return {"success": False, "message": "请提供有效的 DOI"}
+    try:
+        data = await _get_json(f"{BASE_URL}/works/{quote_plus(doi)}")
+        if data.get("status") == "ok" and data.get("message"):
+            return {"success": True, "message": "查询成功", "data": format_article(data["message"])}
+        return {"success": False, "message": "未找到该 DOI 对应的文章"}
+    except Exception as e:
+        return {"success": False, "message": f"查询失败: {getErrorMessage(e)}"}
+
+
+async def search_by_keyword(params):
+    query = (params.get("query") or "").strip()
+    if not query:
+        return {"success": False, "message": "请提供有效的搜索关键词"}
+    rows = max(1, min(int(params.get("rows", DEFAULT_ROWS) or DEFAULT_ROWS), MAX_ROWS))
+    sort = params.get("sort", "relevance")
+    order = params.get("order", "desc")
+    qs = "&".join(f"{k}={quote_plus(v)}" for k, v in
+                  {"query": query, "rows": str(rows), "sort": sort, "order": order}.items())
+    try:
+        data = await _get_json(f"{BASE_URL}/works?{qs}")
+        return _summarize(data, query, f'关键词 "{query}"')
+    except Exception as e:
+        return {"success": False, "message": f"搜索失败: {getErrorMessage(e)}"}
+
+
+async def search_by_author(params):
+    author = (params.get("author") or "").strip()
+    if not author:
+        return {"success": False, "message": "请提供有效的作者名字"}
+    rows = max(1, min(int(params.get("rows", DEFAULT_ROWS) or DEFAULT_ROWS), MAX_ROWS))
+    qs = "&".join(f"{k}={quote_plus(v)}" for k, v in
+                  {"query.author": author, "rows": str(rows)}.items())
+    try:
+        data = await _get_json(f"{BASE_URL}/works?{qs}")
+        return _summarize(data, author, f'作者 "{author}"')
+    except Exception as e:
+        return {"success": False, "message": f"搜索失败: {getErrorMessage(e)}"}
+
+
+async def search_by_title(params):
+    title = (params.get("title") or "").strip()
+    if not title:
+        return {"success": False, "message": "请提供有效的文章标题"}
+    rows = max(1, min(int(params.get("rows", DEFAULT_ROWS) or DEFAULT_ROWS), MAX_ROWS))
+    qs = "&".join(f"{k}={quote_plus(v)}" for k, v in
+                  {"query.title": title, "rows": str(rows)}.items())
+    try:
+        data = await _get_json(f"{BASE_URL}/works?{qs}")
+        return _summarize(data, title, f'标题 "{title}"')
+    except Exception as e:
+        return {"success": False, "message": f"搜索失败: {getErrorMessage(e)}"}
+
+
+async def search_by_issn(params):
+    issn = (params.get("issn") or "").strip()
+    if not issn:
+        return {"success": False, "message": "请提供有效的 ISSN"}
+    rows = max(1, min(int(params.get("rows", DEFAULT_ROWS) or DEFAULT_ROWS), MAX_ROWS))
+    try:
+        data = await _get_json(f"{BASE_URL}/journals/{quote_plus(issn)}/works?rows={rows}")
+        return _summarize(data, issn, f'ISSN {issn}')
+    except Exception as e:
+        return {"success": False, "message": f"查询失败: {getErrorMessage(e)}"}
+
+
+def _summarize(data, key, label):
+    if data.get("status") == "ok" and data.get("message") and data.get("message", {}).get("items"):
+        items = data["message"]["items"]
+        total = data["message"].get("total-results")
+        results = [format_article(it, i) for i, it in enumerate(items)]
+        return {"success": True, "message": "搜索成功",
+                "data": f"Found {total} results for {label} (showing {len(items)}):\n\n" + "\n\n".join(results),
+                "total": total, "count": len(items)}
+    return {"success": False, "message": f"未找到匹配 {label} 的文章"}
+
+
+async def wrap(func, params, ok_msg="操作成功", err_msg="操作失败"):
+    try:
+        result = await func(params)
+        complete({"success": True, "message": result.get("message", ok_msg), "data": result.get("data"),
+                   **{k: v for k, v in result.items() if k not in ("success", "message", "data")}})
+    except Exception as error:
+        console.error(f"Error: {getErrorMessage(error)}")
+        complete({"success": False, "message": getErrorMessage(error), "error_stack": getErrorStack(error)})
+
+
+async def search_by_doi_wrapper(params):
+    await wrap(search_by_doi, params, "查询成功", "查询失败")
+
+
+async def search_by_keyword_wrapper(params):
+    await wrap(search_by_keyword, params, "搜索成功", "搜索失败")
+
+
+async def search_by_author_wrapper(params):
+    await wrap(search_by_author, params, "搜索成功", "搜索失败")
+
+
+async def search_by_title_wrapper(params):
+    await wrap(search_by_title, params, "搜索成功", "搜索失败")
+
+
+async def search_by_issn_wrapper(params):
+    await wrap(search_by_issn, params, "查询成功", "查询失败")
+
+
+exports.search_by_doi = search_by_doi_wrapper
+exports.search_by_keyword = search_by_keyword_wrapper
+exports.search_by_author = search_by_author_wrapper
+exports.search_by_title = search_by_title_wrapper
+exports.search_by_issn = search_by_issn_wrapper

plugins/operit_scripts/market_duckduckgo.py

@@ -0,0 +1,195 @@
+# METADATA
+# {
+#     "name": "market_duckduckgo",
+#     "display_name": {
+#         "zh": "DuckDuckGo 搜索(市场插件翻译)",
+#         "en": "DuckDuckGo Search (translated from operit market)"
+#     },
+#     "description": {
+#         "zh": "从 operit 市场 duckduckgo.ts 忠实翻译：搜索与网页内容抓取。",
+#         "en": "Faithful translation of operit market duckduckgo.ts: search & fetch."
+#     },
+#     "category": "Search",
+#     "tools": [
+#         {
+#             "name": "search",
+#             "description": {
+#                 "zh": "搜索 DuckDuckGo 并返回格式化结果",
+#                 "en": "Search DuckDuckGo and return formatted results"
+#             },
+#             "parameters": [
+#                 {"name": "query", "type": "string", "required": true},
+#                 {"name": "max_results", "type": "string", "required": false}
+#             ]
+#         },
+#         {
+#             "name": "fetch_content",
+#             "description": {
+#                 "zh": "抓取并解析网页内容为纯文本",
+#                 "en": "Fetch and parse webpage content to plain text"
+#             },
+#             "parameters": [
+#                 {"name": "url", "type": "string", "required": true}
+#             ]
+#         }
+#     ]
+# }
+
+import re
+import html
+import asyncio
+
+
+BASE_URL = "https://html.duckduckgo.com/html/"
+
+_UA = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36"
+
+
+class RateLimiter:
+    """轻量异步令牌桶，忠实于 TS 版 RateLimiter.acquire()"""
+    def __init__(self, max_concurrent: int):
+        self._sem = asyncio.Semaphore(max_concurrent)
+
+    async def acquire(self):
+        await self._sem.acquire()
+
+    def release(self):
+        self._sem.release()
+
+
+_search_limiter = RateLimiter(20)
+_fetch_limiter = RateLimiter(20)
+
+
+def getErrorMessage(error):
+    return str(error) if error else "unknown"
+
+
+def getErrorStack(error):
+    import traceback
+    return traceback.format_exc()
+
+
+def _decode_entities(text: str) -> str:
+    return html.unescape(text)
+
+
+def _format_results_for_llm(results: list) -> str:
+    lines = []
+    for r in results:
+        lines.append(f"{r['position']}. {r['title']}")
+        lines.append(f"   URL: {r['link']}")
+        lines.append(f"   {r['snippet']}")
+        lines.append("")
+    return "\n".join(lines).strip()
+
+
+async def wrap(func, params, ok_msg="操作成功", err_msg="操作失败"):
+    try:
+        result = await func(params)
+        complete({"success": True, "message": ok_msg, "data": result})
+    except Exception as error:
+        console.error(f"Error: {getErrorMessage(error)}")
+        complete({"success": False, "message": getErrorMessage(error), "error_stack": getErrorStack(error)})
+
+
+async def search(params):
+    """忠实翻译 duckduckgo.ts search()"""
+    query = params.get("query", "")
+    max_results = 10
+    if params.get("max_results"):
+        try:
+            parsed = int(params["max_results"])
+            if not str(parsed).startswith("NaN"):
+                max_results = parsed
+        except (ValueError, TypeError):
+            pass
+
+    if not query:
+        raise ValueError("查询不能为空")
+
+    await _search_limiter.acquire()
+    try:
+        console.log(f"正在从DuckDuckGo搜索: {query}")
+        client = OkHttp.newClient()
+        request = (client.newRequest()
+                   .url(BASE_URL)
+                   .method("POST")
+                   .body({"q": query, "b": "", "kl": ""}, "form")
+                   .headers({"User-Agent": _UA}))
+
+        response = await request.build().execute()
+        if not response.isSuccessful():
+            raise RuntimeError(f"HTTP 错误! 状态码: {response.statusCode}")
+
+        html_text = response.content
+        results = []
+        result_regex = re.compile(
+            r'<h2 class="result__title">[\s\S]*?<a.*?href="([^"]+)".*?>([\s\S]+?)</a>'
+            r'[\s\S]*?</h2>[\s\S]*?<a class="result__snippet".*?>([\s\S]+?)</a>',
+            re.IGNORECASE,
+        )
+        for m in result_regex.finditer(html_text):
+            if len(results) >= max_results:
+                break
+            link = m.group(1)
+            if "y.js" in link:
+                continue  # 跳过广告
+            if link.startswith("//duckduckgo.com/l/?uddg="):
+                try:
+                    link = _decode_entities(link.split("uddg=")[1].split("&")[0])
+                except Exception:
+                    console.error(f"URL 解码失败: {link}")
+            title = _decode_entities(re.sub(r"<[^>]*>", "", m.group(2)).strip())
+            snippet = _decode_entities(re.sub(r"<[^>]*>", "", m.group(3)).strip())
+            results.append({"title": title, "link": link, "snippet": snippet,
+                            "position": len(results) + 1})
+
+        console.log(f"成功找到 {len(results)} 个结果")
+        return _format_results_for_llm(results)
+    finally:
+        _search_limiter.release()
+
+
+async def fetch_content(params):
+    """忠实翻译 duckduckgo.ts fetch_content()"""
+    url = params.get("url", "")
+    if not url:
+        raise ValueError("URL不能为空")
+
+    await _fetch_limiter.acquire()
+    try:
+        console.log(f"正在抓取内容: {url}")
+        client = OkHttp.newClient()
+        request = (client.newRequest()
+                   .url(url)
+                   .method("GET")
+                   .headers({"User-Agent": _UA}))
+
+        response = await request.build().execute()
+        if not response.isSuccessful():
+            raise RuntimeError(f"无法访问网页 ({response.statusCode})")
+
+        text = response.content
+        for tag in ("style", "script", "nav", "header", "footer"):
+            text = re.sub(rf"<{tag}[^>]*>[\s\S]*?</{tag}>", "", text, flags=re.IGNORECASE)
+        text = re.sub(r"<[^>]+>", " ", text)
+        text = re.sub(r"\s+", " ", text).strip()
+        if len(text) > 8000:
+            text = text[:8000] + "... [内容被截断]"
+        console.log(f"成功抓取并解析内容 ({len(text)} 字符)")
+        return text
+    finally:
+        _fetch_limiter.release()
+
+
+async def search_wrapper(params):
+    await wrap(search, params, "搜索完成", "搜索失败")
+
+
+async def fetch_content_wrapper(params):
+    await wrap(fetch_content, params, "抓取完成", "抓取失败")
+
+
+exports.search = search_wrapper
+exports.fetch_content = fetch_content_wrapper

plugins/operit_scripts/market_network_test.py

@@ -0,0 +1,90 @@
+# METADATA
+# {
+#   "name": "network_test",
+#   "display_name": { "zh": "网络测试", "en": "Network Test" },
+#   "description": { "zh": "基于 OkHttp 的网络请求测试（GET/POST）", "en": "Network request testing via OkHttp." },
+#   "category": "Network",
+#   "tools": [
+#     { "name": "http_get", "description": { "zh": "发送 HTTP GET 请求", "en": "Send HTTP GET" } },
+#     { "name": "http_post", "description": { "zh": "发送 HTTP POST 请求", "en": "Send HTTP POST" } }
+#   ]
+# }
+import json as _json
+
+
+def getErrorMessage(error):
+    return str(error) if error else "unknown"
+
+
+def getErrorStack(error):
+    import traceback
+    return traceback.format_exc()
+
+
+def format_response(resp):
+    content = resp.content or ""
+    result = {
+        "status_code": resp.statusCode,
+        "status_message": resp.statusMessage,
+        "content_length": len(content),
+        "headers": resp.headers(),
+        "content": content[:12000],
+    }
+    try:
+        result["json"] = _json.loads(content)
+    except Exception:
+        result["json"] = None
+    return result
+
+
+async def http_get(params):
+    if not params.get("url"):
+        return {"success": False, "message": "URL不能为空"}
+    try:
+        console.log(f"发送GET请求: {params['url']}")
+        req = OkHttp.newClient().newRequest().url(params["url"]).method("GET")
+        if params.get("headers"):
+            req = req.headers(params["headers"])
+        resp = await req.build().execute()
+        return {"success": True, "message": "GET请求完成", "data": format_response(resp)}
+    except Exception as e:
+        return {"success": False, "message": f"GET请求失败: {getErrorMessage(e)}"}
+
+
+async def http_post(params):
+    if not params.get("url"):
+        return {"success": False, "message": "URL不能为空"}
+    if params.get("body") is None:
+        return {"success": False, "message": "请求体不能为空"}
+    try:
+        req = OkHttp.newClient().newRequest().url(params["url"]).method("POST")
+        if params.get("headers"):
+            req = req.headers(params["headers"])
+        body_type = params.get("body_type", "json")
+        req = req.body(params["body"], body_type)
+        resp = await req.build().execute()
+        return {"success": True, "message": "POST请求完成", "data": format_response(resp)}
+    except Exception as e:
+        return {"success": False, "message": f"POST请求失败: {getErrorMessage(e)}"}
+
+
+async def wrap(func, params, ok_msg="操作成功", err_msg="操作失败"):
+    try:
+        result = await func(params)
+        complete({"success": True, "message": result.get("message", ok_msg),
+                   "data": result.get("data"), **{k: v for k, v in result.items() if k not in ("success", "message", "data")}})
+    except Exception as error:
+        console.error(f"Error: {getErrorMessage(error)}")
+        complete({"success": False, "message": getErrorMessage(error), "error_stack": getErrorStack(error)})
+
+
+async def http_get_wrapper(params):
+    await wrap(http_get, params, "GET请求完成", "GET请求失败")
+
+
+async def http_post_wrapper(params):
+    await wrap(http_post, params, "POST请求完成", "POST请求失败")
+
+
+exports.http_get = http_get_wrapper
+exports.http_post = http_post_wrapper

plugins/operit_scripts/market_reader.py

@@ -0,0 +1,85 @@
+# METADATA
+# {
+#   "name": "reader",
+#   "display_name": { "zh": "代码读取器", "en": "Code Reader" },
+#   "description": { "zh": "在文件夹中搜索匹配的代码（本地文件，无需 Android）", "en": "Search matching code in a folder." },
+#   "category": "File",
+#   "tools": [
+#     { "name": "search_code_in_folder", "description": { "zh": "在文件夹中搜索匹配的代码", "en": "Search matching code in a folder" } }
+#   ]
+# }
+import os
+import re
+
+
+def getErrorMessage(error):
+    return str(error) if error else "unknown"
+
+
+def getErrorStack(error):
+    import traceback
+    return traceback.format_exc()
+
+
+async def search_code_in_folder(params):
+    folder_path = params.get("folder_path")
+    pattern = params.get("pattern")
+    file_extensions = params.get("file_extensions", ".js,.ts,.jsx,.tsx,.java,.cs,.py")
+    recursive = str(params.get("recursive", "true")).lower() in ("true", "1", "yes")
+    if not folder_path:
+        return {"success": False, "message": "文件夹路径不能为空"}
+    if not pattern:
+        return {"success": False, "message": "搜索模式不能为空"}
+    try:
+        exts = [e.strip().lower() for e in file_extensions.split(",") if e.strip()]
+        rx = re.compile(pattern, re.MULTILINE)
+        files = []
+        if recursive:
+            for root, _, fnames in os.walk(folder_path):
+                for fn in fnames:
+                    if any(fn.lower().endswith(e) for e in exts):
+                        files.append(os.path.join(root, fn))
+        else:
+            for fn in os.listdir(folder_path):
+                fp = os.path.join(folder_path, fn)
+                if os.path.isfile(fp) and any(fn.lower().endswith(e) for e in exts):
+                    files.append(fp)
+
+        results = []
+        for fp in files:
+            try:
+                with open(fp, "r", encoding="utf-8", errors="replace") as fh:
+                    content = fh.read()
+            except Exception as e:
+                console.error(f"处理文件 {fp} 时出错: {getErrorMessage(e)}")
+                continue
+            matches = []
+            for m in rx.finditer(content):
+                ln = content.count("\n", 0, m.start()) + 1
+                matches.append({"match": m.group(0), "line": ln,
+                                 "content": content.split("\n")[ln - 1] if ln - 1 < len(content.split("\n")) else ""})
+            if matches:
+                results.append({"file": fp, "matches": matches})
+        total = sum(len(r["matches"]) for r in results)
+        return {"success": True, "message": f"在 {len(files)} 个文件中找到 {total} 处匹配",
+                "data": {"files_searched": len(files), "files_with_matches": len(results),
+                         "total_matches": total, "results": results[:50]}}
+    except Exception as e:
+        return {"success": False, "message": f"搜索失败: {getErrorMessage(e)}"}
+
+
+async def wrap(func, params, ok_msg="操作成功", err_msg="操作失败"):
+    try:
+        result = await func(params)
+        complete({"success": True, "message": result.get("message", ok_msg),
+                   "data": result.get("data"), **{k: v for k, v in result.items() if k not in ("success", "message", "data")}})
+    except Exception as error:
+        console.error(f"Error: {getErrorMessage(error)}")
+        complete({"success": False, "message": getErrorMessage(error), "error_stack": getErrorStack(error)})
+
+
+async def search_code_in_folder_wrapper(params):
+    await wrap(search_code_in_folder, params, "代码搜索完成", "代码搜索失败")
+
+
+exports.search_code_in_folder = search_code_in_folder_wrapper

plugins/operit_scripts/market_time.py

@@ -0,0 +1,103 @@
+# METADATA
+# {
+#     "name": "market_time",
+#     "display_name": {
+#         "zh": "时间(市场插件翻译)",
+#         "en": "Time (translated from operit market)"
+#     },
+#     "description": {
+#         "zh": "从 operit 市场 time.ts 忠实翻译：获取/格式化当前时间。",
+#         "en": "Faithful translation of operit market time.ts: get/format current time."
+#     },
+#     "category": "Utility",
+#     "tools": [
+#         {
+#             "name": "get_time",
+#             "description": {
+#                 "zh": "获取当前时间各字段",
+#                 "en": "Get current time fields"
+#             },
+#             "parameters": []
+#         },
+#         {
+#             "name": "format_time",
+#             "description": {
+#                 "zh": "格式化当前时间为 24h/12h 字符串",
+#                 "en": "Format current time as 24h/12h strings"
+#             },
+#             "parameters": []
+#         }
+#     ]
+# }
+
+import datetime
+
+
+def getErrorMessage(error):
+    return str(error) if error else "unknown"
+
+
+def getErrorStack(error):
+    import traceback
+    return traceback.format_exc()
+
+
+async def wrap(func, params, ok_msg="操作成功", err_msg="操作失败"):
+    try:
+        result = await func(params)
+        complete({"success": True, "message": ok_msg, "data": result})
+    except Exception as error:
+        console.error(f"Error: {getErrorMessage(error)}")
+        complete({"success": False, "message": getErrorMessage(error), "error_stack": getErrorStack(error)})
+
+
+async def get_time(params):
+    """获取当前时间各字段（忠实于 time.ts get_time）"""
+    now = datetime.datetime.now()
+    return {
+        "timestamp": int(now.timestamp() * 1000),
+        "iso": now.isoformat(),
+        "local": now.strftime("%Y-%m-%d %H:%M:%S"),
+        "date": {
+            "year": now.year,
+            "month": now.month,
+            "day": now.day,
+            "weekday": now.strftime("%A"),
+        },
+        "time": {
+            "hours": now.hour,
+            "minutes": now.minute,
+            "seconds": now.second,
+        },
+    }
+
+
+async def format_time(params):
+    """格式化当前时间为 24h/12h 字符串（忠实于 time.ts format_time）"""
+    now = datetime.datetime.now()
+    pad = lambda n: str(n).zfill(2)
+    hours = now.hour
+    minutes = now.minute
+    seconds = now.second
+    time24h = f"{pad(hours)}:{pad(minutes)}:{pad(seconds)}"
+    suffix = "PM" if hours >= 12 else "AM"
+    hour12 = 12 if hours % 12 == 0 else hours % 12
+    time12h = f"{pad(hour12)}:{pad(minutes)}:{pad(seconds)} {suffix}"
+    return {
+        "iso": now.isoformat(),
+        "date": f"{now.year}-{pad(now.month)}-{pad(now.day)}",
+        "time24h": time24h,
+        "time12h": time12h,
+    }
+
+
+async def get_time_wrapper(params):
+    await wrap(get_time, params, "时间获取完成", "时间获取失败")
+
+
+async def format_time_wrapper(params):
+    await wrap(format_time, params, "时间格式化完成", "时间格式化失败")
+
+
+exports.get_time = get_time_wrapper
+exports.format_time = format_time_wrapper

plugins/operit_scripts/system_info.py

@@ -0,0 +1,78 @@
+# METADATA
+# {
+#     "name": "system_info",
+#     "display_name": {
+#         "zh": "系统信息查看器",
+#         "en": "System Info Viewer"
+#     },
+#     "description": {
+#         "zh": "通过 Android API 转译层获取系统信息：存储目录、构建信息、平台环境，并用 Log 输出日志。",
+#         "en": "Get system info via Android API translation layer: storage dirs, build info, platform. Log via android.util.Log."
+#     },
+#     "category": "System",
+#     "tools": [
+#         {
+#             "name": "get_info",
+#             "description": {
+#                 "zh": "获取系统信息",
+#                 "en": "Get system info"
+#             },
+#             "parameters": []
+#         }
+#     ]
+# }
+
+import platform
+import sys
+
+TAG = "SystemInfo"
+
+
+def getErrorMessage(error):
+    return str(error) if error else "unknown"
+
+
+def getErrorStack(error):
+    return traceback.format_exc()
+
+
+async def wrap(func, params):
+    try:
+        result = await func(params)
+        complete({"success": True, "message": "操作成功", "data": result})
+    except Exception as error:
+        console.error(f"Error: {getErrorMessage(error)}")
+        complete({"success": False, "message": getErrorMessage(error), "error_stack": getErrorStack(error)})
+
+
+async def get_info(params):
+    """通过 Android API 转译层采集系统信息"""
+    try:
+        info = {
+            "storage_external": Environment.getExternalStorageDirectory(),
+            "storage_data": Environment.getDataDirectory(),
+            "storage_root": Environment.getRootDirectory(),
+            "storage_public_pictures": Environment.getExternalStoragePublicDirectory(Environment.DIRECTORY_PICTURES),
+            "build_brand": getattr(Build, "BRAND", "unknown"),
+            "build_model": getattr(Build, "MODEL", "unknown"),
+            "build_version": getattr(Build, "VERSION", {}).get("RELEASE", "unknown") if isinstance(getattr(Build, "VERSION", {}), dict) else "unknown",
+            "os_name": os.name,
+            "sys_platform": sys.platform,
+            "python_version": platform.python_version(),
+            "python_platform": platform.platform(),
+            "sdcard_mounted": Environment.getExternalStorageState() == Environment.MEDIA_MOUNTED,
+        }
+        Log.i(TAG, f"采集完成: external={info['storage_external']}")
+        complete({"success": True, "message": "系统信息采集完成", "data": info})
+        return info
+    except Exception as error:
+        console.error(f"{TAG} 错误: {getErrorMessage(error)}")
+        complete({"success": False, "message": f"采集失败: {getErrorMessage(error)}", "error_stack": getErrorStack(error)})
+        raise
+
+
+async def get_info_wrapper(params):
+    await wrap(get_info, params)
+
+
+exports.get_info = get_info_wrapper

plugins/operit_translate.py

@@ -0,0 +1,175 @@
+"""operit 市场插件 (.ts) → xiaoli-cli 可运行 .py 的脚手架生成器。
+
+设计原则（不需要 android_pyapi）：
+  operit 插件使用的是「通用平台 API」(OkHttp / Tools / toolCall /
+  NativeInterface / complete / console / exports)，不是 Android API。
+  xiaoli-cli 的 operit_loader 已经 1:1 镜像了这套 API（全部用 requests/
+  httpx 在 Python 端原生实现），所以「翻译」只是把 TS 语法换成 Python 语法，
+  全局名原样保留 —— 不存在 Android 依赖。
+
+本工具只做「可靠」的样板自动化：
+  1. 从 /* METADATA {json} */ 提取插件元数据，生成为 Python 友好的 # METADATA 块
+  2. 生成 run_tool 契约所需的胶水：getErrorMessage / getErrorStack / wrap /
+     exports.x = wrapper
+  3. 为每个工具生成 async def 骨架，并把原 .ts 函数体作为 `/// TS 参考` 注释嵌入
+  4. 复杂逻辑用 TODO 标注，绝不像「全量 TS→PY 编译器」那样产生半坏代码
+
+用法：
+  python operit_translate.py <市场插件.ts> [-o <输出.py>] [-n <插件名>]
+"""
+import sys
+import json
+import re
+import os
+
+
+def extract_metadata(ts: str):
+    """从 /* METADATA ... */ 提取 JSON 元数据，返回 (meta_dict, cleaned_ts)。"""
+    m = re.search(r"/\*\s*METADATA\s*(.*?)\*/", ts, re.DOTALL)
+    if not m:
+        return None, ts
+    raw = m.group(1)
+    try:
+        meta = json.loads(raw)
+    except json.JSONDecodeError:
+        # 宽松：去注释后重试
+        try:
+            meta = json.loads(re.sub(r"//.*", "", raw))
+        except json.JSONDecodeError:
+            meta = None
+    cleaned = ts[:m.start()] + ts[m.end():]
+    return meta, cleaned
+
+
+def py_meta_block(meta: dict) -> str:
+    """把元数据渲染成 # METADATA 块（operit_loader 逐行解析的格式）。"""
+    js = json.dumps(meta, ensure_ascii=False, indent=2)
+    lines = ["# METADATA", "# {"] + ["#     " + ln for ln in js[1:-1].split("\n")] + ["# }", ""]
+    return "\n".join(lines)
+
+
+def strip_ts_types_and_comments(ts: str) -> str:
+    """去掉类型注解与注释，仅用于把 .ts 体作为参考注释嵌入（不参与执行）。"""
+    # 去块注释 / 行注释
+    ts = re.sub(r"/\*.*?\*/", "", ts, flags=re.DOTALL)
+    ts = re.sub(r"//[^\n]*", "", ts)
+    # as 断言
+    ts = re.sub(r"\bas\s+[\w\[\]]+", "", ts)
+    # 泛型
+    ts = re.sub(r"\b\w+<[^>]+>", lambda m: m.group(0).split("<")[0], ts)
+    # 修饰符
+    ts = re.sub(r"\b(private|public|protected|readonly|static|export|async|const|let|var)\b", "", ts)
+    # 参数/返回类型
+    ts = re.sub(r"\)\s*:\s*[^{]+?(\{)", r")\1", ts)
+    ts = re.sub(r"\(([^()]*)\)\s*:\s*[^{]+", r"(\1)", ts)
+    return ts
+
+
+def indented(block: str, prefix: str = "    ") -> str:
+    return "\n".join((prefix + ln if ln.strip() else ln) for ln in block.split("\n"))
+
+
+def generate(py_name: str, ts_path: str) -> str:
+    with open(ts_path, "r", encoding="utf-8") as f:
+        ts = f.read()
+
+    meta, cleaned = extract_metadata(ts)
+    if not meta:
+        meta = {"name": py_name, "display_name": {"zh": py_name, "en": py_name},
+                "description": {"zh": "从 operit 市场翻译", "en": "Translated from operit market"},
+                "category": "Utility", "tools": []}
+
+    tools = meta.get("tools", [])
+    if not tools:
+        # 退路：从 exports.x = 猜测工具名
+        for mm in re.finditer(r"exports\.(\w+)\s*=", cleaned):
+            tools.append({"name": mm.group(1)})
+
+    out = []
+    out.append(py_meta_block(meta))
+    out.append("import re")
+    out.append("import html")
+    out.append("import asyncio")
+    out.append("")
+    out.append("")
+    out.append("def getErrorMessage(error):")
+    out.append("    return str(error) if error else \"unknown\"")
+    out.append("")
+    out.append("")
+    out.append("def getErrorStack(error):")
+    out.append("    import traceback")
+    out.append("    return traceback.format_exc()")
+    out.append("")
+    out.append("")
+    out.append("async def wrap(func, params, ok_msg=\"操作成功\", err_msg=\"操作失败\"):")
+    out.append("    \"\"\"operit 的 complete({success,message,data}) 契约封装\"\"\"")
+    out.append("    try:")
+    out.append("        result = await func(params)")
+    out.append("        complete({\"success\": True, \"message\": ok_msg, \"data\": result})")
+    out.append("    except Exception as error:")
+    out.append("        console.error(f\"Error: {getErrorMessage(error)}\")")
+    out.append("        complete({\"success\": False, \"message\": getErrorMessage(error),")
+    out.append("                 \"error_stack\": getErrorStack(error)})")
+    out.append("")
+    out.append("")
+    # 参考注释：原 .ts 去类型后的体
+    ref = strip_ts_types_and_comments(cleaned).strip()
+    if ref:
+        out.append("# ==================== 原 .ts 参考（去类型/注释，仅供翻译对照） ====================")
+        for ln in ref.split("\n"):
+            out.append("# " + ln)
+        out.append("# ========================================================================================")
+        out.append("")
+
+    # 每个工具：骨架 + wrapper + exports
+    for t in tools:
+        name = t.get("name", "tool")
+        desc = t.get("description", {})
+        desc_zh = desc.get("zh", "") if isinstance(desc, dict) else str(desc)
+        out.append("")
+        out.append(f"async def {name}(params):")
+        out.append(f"    \"\"\"{desc_zh or name}（译自 operit 市场 {meta.get('name','?')}.ts）\"\"\"")
+        out.append("    # TODO: 用下面的 1:1 API 映射把 .ts 逻辑补全（参考上方 /// TS 参考）")
+        out.append("    # OkHttp.newClient().newRequest().url(u).method('GET'/'POST')")
+        out.append("    #        .body(payload, 'json'|'form').headers({...}).build().execute()  # async")
+        out.append("    # response.isSuccessful() / response.statusCode / response.content")
+        out.append("    # Tools.System.sleep(ms) -> {'requestedMs':, 'sleptMs':}")
+        out.append("    # console.log / console.error / complete({success,message,data})")
+        out.append(f'    raise NotImplementedError("{name} 待翻译")')
+        out.append("")
+        out.append("")
+        out.append(f"async def {name}_wrapper(params):")
+        out.append(f"    await wrap({name}, params, \"{name} 完成\", \"{name} 失败\")")
+        out.append("")
+        out.append("")
+        out.append(f"exports.{name} = {name}_wrapper")
+        out.append("")
+
+    return "\n".join(out)
+
+
+def main():
+    if len(sys.argv) < 2:
+        print(__doc__)
+        sys.exit(1)
+    ts_path = sys.argv[1]
+    py_name = None
+    out_path = None
+    if "-n" in sys.argv:
+        py_name = sys.argv[sys.argv.index("-n") + 1]
+    if "-o" in sys.argv:
+        out_path = sys.argv[sys.argv.index("-o") + 1]
+    if not py_name:
+        py_name = os.path.splitext(os.path.basename(ts_path))[0].replace("-", "_")
+    if not out_path:
+        out_path = os.path.join(os.path.dirname(ts_path), py_name + ".py")
+
+    code = generate(py_name, ts_path)
+    with open(out_path, "w", encoding="utf-8") as f:
+        f.write(code)
+    print(f"已生成脚手架: {out_path}  ({len(code)} 字节)")
+    print("下一步: 打开文件，把 TODO 处按 1:1 API 映射补全即可（无需 android_pyapi）。")
+
+
+if __name__ == "__main__":
+    main()

plugins_config.py

@@ -1,20 +1,31 @@
-# 小狸插件配置
+# 小狸插件配置 - 由启动器自动生成
-
+ENABLE_AI_IMAGE = True
 ENABLE_AI_SEARCH = True
 ENABLE_AUDIO_PLAYER = True
+ENABLE_AUTO_ENGINEER = True
+ENABLE_BOOKMARK_MANAGER = True
+ENABLE_BROWSER_AUTO = True
 ENABLE_CMD_EXECUTOR = True
-ENABLE_FILE_MANAGER = True
 ENABLE_CODE_EDITOR = True
-ENABLE_CODE_SEARCH = True
+ENABLE_COLOR_PICKER = True
-ENABLE_GIT_TOOLS = True
+ENABLE_FILE_MANAGER = True
 ENABLE_FRONTEND_TESTER = True
+ENABLE_GIT_TOOLS = True
+ENABLE_GUI_AUTO = True
+ENABLE_ID_GENERATOR = True
+ENABLE_IMAGE_GENERATOR = True
+ENABLE_JSON_TOOLKIT = True
+ENABLE_MARKDOWN_TOOL = True
+ENABLE_MEMORY_PLUGIN = True
 ENABLE_NETWORK_TOOLS = True
+ENABLE_OPERIT_LOADER = True
+ENABLE_PY_DETECT = True
+ENABLE_REGEX_TESTER = True
 ENABLE_SCHEDULER = True
+ENABLE_SEND_IMAGE = True
+ENABLE_SNIPPET_MANAGER = True
 ENABLE_SPEECH_RECOGNITION = True
+ENABLE_SUB_AGENT = True
 ENABLE_TASK_MANAGER = True
 ENABLE_TOOL_SEARCH = True
-ENABLE_AUTO_ENGINEER = True
+ENABLE_WORKFLOW = True
-ENABLE_GIT_WORKFLOW = True
-ENABLE_PROJECT_ANALYZER = True
-
-ENABLE_DGLAB_WS = False

requirements.txt

@@ -1,30 +1,25 @@
-# 小狸 Pro-CLI 依赖库
+# 小狸 Pro-CLI 依赖
-# 修复后的包名（使用 pip 安装名）
+# 生成于 2026-08-05 21:10
 
-# 核心依赖
+PIL
-colorama>=0.4.6
+ascii_magic
-openai>=1.0.0
+colorama
-requests>=2.28.0
+cv2
-ollama>=0.1.0
+numpy
+ollama
+openai
+pygame
+requests
+tkinter
 
-# 图像处理
+# LSP 智能感知（pylsp 作为工具之一：lsp__diagnostics/hover/definition/references/completion）
-Pillow>=9.0.0
+# 写/改 .py 文件后自动触发诊断并注入模型上下文
-opencv-python>=4.5.0
+python-lsp-server
-numpy>=1.21.0
+pyflakes
-ascii-magic>=2.0.0
+pycodestyle
+rope
+mccabe
 
-# 数据处理
+# 自研本地 LSP 服务器 / py_detect L2 语义验证后端
-pandas>=1.3.0
+# 纯 Python 语义引擎，装依赖即自动预装（L2 自动启用；无 jedi 自动降级为纯标准库模式）
-scipy>=1.7.0
+jedi
-
-# 音频
-pygame>=2.0.0
-
-# TUI 模式
-textual>=0.40.0
-
-# WebSocket 远程
-websockets>=11.0
-
-# 类型提示
-typing-extensions>=4.0.0

tests/test_agents.py

@@ -0,0 +1,104 @@
+"""Agents 体系化测试 - AgentManager 加载/解析/委派（不触真实引擎）"""
+import os
+import tempfile
+import textwrap
+
+from xcli_core.agent_manager import AgentManager, AgentDef
+
+
+def _write(path, text):
+    with open(path, "w", encoding="utf-8") as f:
+        f.write(text)
+
+
+def test_parse_frontmatter():
+    text = textwrap.dedent("""\
+    ---
+    name: coder
+    description: 代码实现专家
+    tools: code_editor, git_tools, file_manager
+    model: gpt-4o
+    ---
+    你是一个资深程序员。
+    任务完成后报告结果。
+    """)
+    d = AgentManager._parse(text, name="coder")
+    assert isinstance(d, AgentDef)
+    assert d.name == "coder"
+    assert d.description == "代码实现专家"
+    assert d.tools == ["code_editor", "git_tools", "file_manager"]
+    assert d.model == "gpt-4o"
+    assert "资深程序员" in d.system_prompt
+
+
+def test_parse_no_frontmatter_uses_filename_and_body():
+    text = "纯文本 agent 定义，无 frontmatter。"
+    d = AgentManager._parse(text, name="plain")
+    assert d.name == "plain"
+    assert d.system_prompt == text
+    assert d.tools == []
+
+
+def test_load_dir_and_list():
+    with tempfile.TemporaryDirectory() as td:
+        _write(os.path.join(td, "coder.md"), textwrap.dedent("""\
+        ---
+        name: coder
+        description: 编码
+        tools: code_editor
+        ---
+        写代码。
+        """))
+        _write(os.path.join(td, "reviewer.md"), textwrap.dedent("""\
+        ---
+        name: reviewer
+        description: 审查
+        tools: code_editor, git_tools
+        ---
+        审代码。
+        """))
+        _write(os.path.join(td, "skip.txt"), "不是 md")
+        mgr = AgentManager()
+        n = mgr.load_dir(td)
+        assert n == 2
+        assert set(mgr.names()) == {"coder", "reviewer"}
+        assert mgr.get("coder").description == "编码"
+        assert mgr.get("missing") is None
+
+
+def test_load_dir_missing_dir_is_noop():
+    mgr = AgentManager()
+    assert mgr.load_dir("") == 0
+    assert mgr.load_dir("/no/such/dir/xyz") == 0
+
+
+def test_dispatch_calls_runner_with_fields():
+    mgr = AgentManager()
+    mgr._agents["coder"] = AgentDef(
+        name="coder", description="编码",
+        tools=["code_editor"], model="gpt-4o",
+        system_prompt="你是 coder。",
+    )
+    captured = {}
+    def runner(system_prompt, task, tools, model):
+        captured["system_prompt"] = system_prompt
+        captured["task"] = task
+        captured["tools"] = tools
+        captured["model"] = model
+        return f"done:{task}"
+    out = mgr.dispatch("coder", "写 hello.py", runner)
+    assert out == "done:写 hello.py"
+    assert captured["system_prompt"] == "你是 coder。"
+    assert captured["tools"] == ["code_editor"]
+    assert captured["model"] == "gpt-4o"
+
+
+def test_dispatch_missing_raises():
+    mgr = AgentManager()
+    def runner(*a):
+        return "x"
+    try:
+        mgr.dispatch("nope", "t", runner)
+        assert False, "expected KeyError"
+    except KeyError:
+        pass

tests/test_compress_quality.py

@@ -0,0 +1,103 @@
+"""上下文压缩质量测试：覆盖保真度、可配置 keep_recent、避免二次失真。"""
+import os
+import sys
+import pytest
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from xcli_core.memory import MemoryManager
+
+
+class FakeEngine:
+    """捕获发给引擎的摘要 prompt，并返回确定结构化摘要。"""
+    def __init__(self):
+        self.last_prompt = None
+        self.calls = 0
+
+    def generate_response(self, user_input, system_prompt=None, tools=None):
+        self.last_prompt = user_input
+        self.calls += 1
+        return "[用户]: 需求A\n[助手]: 调用 code_editor 创建 main.py\n[助手]: 调用 git_tools 提交"
+
+
+class _Msg(dict):
+    def __init__(self, role, content):
+        super().__init__(role=role, content=content)
+
+
+def _make_convo(n):
+    msgs = []
+    for i in range(n):
+        msgs.append(_Msg("user" if i % 2 == 0 else "assistant",
+                         f"消息{i} 内容占位 " + "x" * 5))
+    return msgs
+
+
+def test_prepare_keeps_tool_json_full():
+    """工具调用 JSON 不应被 200 字截断，应完整保留。"""
+    mm = MemoryManager(project_dir="C:/tmp/xiaoli_test_mm")
+    tool_json = ('{"action": "use_tool", "tool": "code_editor", '
+                 '"args": "edit C:/repo/main.py old_block <<<>>> new_block"}')
+    msg = _Msg("assistant", "我先调用工具：\n" + tool_json)
+    out = mm._prepare_for_summary(msg)
+    assert tool_json in out, "工具调用 JSON 必须完整保留"
+    assert "[助手] [工具调用]" in out or "[assistant] [工具调用]" in out
+
+
+def test_prepare_keeps_code_fence_when_long():
+    """超长内容但含代码块时，应保留代码块而非简单前 1500 字截断。"""
+    mm = MemoryManager(project_dir="C:/tmp/xiaoli_test_mm")
+    long_prose = "p" * 3000
+    code = "```python\ndef add(a, b):\n    return a + b\n```"
+    msg = _Msg("assistant", long_prose + "\n" + code)
+    out = mm._prepare_for_summary(msg)
+    assert "def add" in out, "长内容中的代码块应被保留"
+    assert "保留代码块" in out
+
+
+def test_compress_uses_current_engine_and_keeps_recent():
+    """压缩调用当前引擎，并保留最近 keep_recent 条。"""
+    mm = MemoryManager(project_dir="C:/tmp/xiaoli_test_mm")
+    eng = FakeEngine()
+    conv = _make_convo(30)  # 30 条 > keep_recent(12)
+    result = mm.compress_context(conv, engine=eng)
+    # 返回 = 1 条摘要 system + 最近 12 条
+    assert len(result) == 1 + mm.keep_recent
+    assert result[0]["role"] == "system"
+    assert "[上下文压缩]" in result[0]["content"]
+    # 最近的消息在末尾
+    assert result[-1] is conv[-1]
+    assert eng.calls == 1
+
+
+def test_no_double_compaction_on_growth():
+    """历史摘要标记不会被二次摘要；真正新增的消息才会产生新摘要。"""
+    mm = MemoryManager(project_dir="C:/tmp/xiaoli_test_mm")
+    eng = FakeEngine()
+    conv = _make_convo(30)
+    r1 = mm.compress_context(conv, engine=eng)
+    assert len(mm.compressed_blobs) == 1
+
+    # 幂等：对返回的同一历史再次压缩，不应二次摘要旧摘要、不调引擎
+    eng2 = FakeEngine()
+    r2 = mm.compress_context(r1, engine=eng2)
+    assert r2 is r1, "对已压缩历史再压缩应为幂等 no-op"
+    assert len(mm.compressed_blobs) == 1, "不应二次摘要旧摘要"
+    assert eng2.calls == 0
+
+    # 真正增长的新消息应被压缩为第二条摘要，且旧摘要标记保留
+    grown2 = list(r1) + [_Msg("user", f"后续{i}") for i in range(15)]
+    eng3 = FakeEngine()
+    r3 = mm.compress_context(grown2, engine=eng3)
+    assert len(mm.compressed_blobs) == 2, "新增的真实消息应被压缩为第二条摘要"
+    # 合并摘要里应同时含第一次与第二次的标记
+    assert "[上下文压缩]" in r3[0]["content"]
+
+
+def test_keep_recent_configurable():
+    """keep_recent 可通过参数覆盖。"""
+    mm = MemoryManager(project_dir="C:/tmp/xiaoli_test_mm")
+    eng = FakeEngine()
+    conv = _make_convo(20)
+    result = mm.compress_context(conv, engine=eng, keep_recent=5)
+    assert len(result) == 1 + 5

tests/test_fc_hybrid.py

@@ -0,0 +1,121 @@
+"""
+FC 混合模式（AI 自选）集成测试
+
+验证「让 AI 自己选纯 JSON 或 FC」的接线：
+1. 支持 FC 的引擎会收到下发的 tools schema（tool_choice=auto，模型自选）
+2. 引擎返回 _fc JSON 时，code_editor 被按名分发执行（走 _handle_fc_response 路径）
+3. 不支持 FC 的引擎（无 _fc_supported）则不收 tools，纯 JSON 照常工作
+"""
+import os
+import sys
+import json
+import tempfile
+
+import pytest
+
+PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+if PROJECT_ROOT not in sys.path:
+    sys.path.insert(0, PROJECT_ROOT)
+
+
+class FakeFcEngine:
+    """模拟一个支持 FC 的引擎：首轮返回 _fc 调用 code_editor，次轮收尾。"""
+    def __init__(self):
+        self.calls = 0
+        self.tools_seen = []
+
+    @property
+    def name(self):
+        return "fakefc"
+
+    def _fc_supported(self):
+        return True
+
+    def generate_response(self, user_input, tool_results=None, system_prompt=None, tools=None):
+        self.calls += 1
+        self.tools_seen.append(tools)
+        if self.calls == 1:
+            return json.dumps({
+                "_fc": True,
+                "tool_calls": [{
+                    "id": "c1",
+                    "name": "code_editor",
+                    "tool": "code_editor",
+                    "args": "create %s print(1)" % self._fp,
+                }],
+            }, ensure_ascii=False)
+        return "已完成。"
+
+
+class FakeNoFcEngine:
+    """模拟一个不支持 FC 的引擎（无 _fc_supported）：应降级为纯 JSON。"""
+    def __init__(self):
+        self.calls = 0
+        self.tools_seen = []
+
+    @property
+    def name(self):
+        return "fakenofc"
+
+    def generate_response(self, user_input, tool_results=None, system_prompt=None, tools=None):
+        self.calls += 1
+        self.tools_seen.append(tools)
+        assert tools is None, f"不支持 FC 的引擎不应收到 tools，实际 {tools!r}"
+        if self.calls == 1:
+            return ('好的，我来创建文件。\n'
+                    '{"action": "use_tool", "tool": "code_editor", '
+                    '"args": "create %s print(1)"}' % self._fp)
+        return "已完成。"
+
+
+def _run(engine):
+    from xcli_core.cli_core import AICLI
+    from xcli_core.safety import get_safety, MODE_UNRESTRICTED
+
+    cli = AICLI()
+    cli.current_engine = engine
+    cli.tui_output_callback = lambda *a, **k: None  # 同步 TUI 分支，免动画线程
+    get_safety().set_mode(MODE_UNRESTRICTED)
+
+    d = tempfile.mkdtemp(prefix="pj_")
+    fp = os.path.join(d, "demo.py")
+    engine._fp = fp
+    cli.process_conversation("帮我创建一个 python 文件")
+    return engine, fp
+
+
+def test_fc_engine_receives_tools_and_dispatches():
+    engine, fp = _run(FakeFcEngine())
+    # 首轮应向支持 FC 的引擎下发 tools schema
+    assert engine.tools_seen[0] is not None, "FC 引擎应收到 tools schema"
+    assert isinstance(engine.tools_seen[0], list) and len(engine.tools_seen[0]) > 0
+    assert any(
+        t.get("function", {}).get("name") == "code_editor" for t in engine.tools_seen[0]
+    ), "code_editor 应在下发的 FC 工具集中"
+    # _fc 调用应被按名分发执行并真实创建文件
+    assert os.path.exists(fp), "FC 路径应执行 code_editor 并创建文件"
+    assert "print(1)" in open(fp, encoding="utf-8").read()
+    assert engine.calls >= 2
+
+
+def test_no_fc_engine_stays_pure_json():
+    engine, fp = _run(FakeNoFcEngine())
+    # 不支持 FC 的引擎不应收到 tools，纯 JSON 照常分发
+    assert all(t is None for t in engine.tools_seen), "不支持 FC 的引擎不应收到 tools"
+    assert os.path.exists(fp), "纯 JSON 路径仍应创建文件"
+
+
+def test_build_fc_tools_returns_none_for_unsupported_engine():
+    from xcli_core.cli_core import AICLI
+    cli = AICLI()
+    cli.current_engine = FakeNoFcEngine()
+    assert cli._build_fc_tools("帮我创建文件") is None
+
+
+def test_build_fc_tools_returns_list_for_supported_engine():
+    from xcli_core.cli_core import AICLI
+    cli = AICLI()
+    cli.current_engine = FakeFcEngine()
+    tools = cli._build_fc_tools("帮我创建 python 文件")
+    assert isinstance(tools, list) and len(tools) > 0
+    assert any(t.get("function", {}).get("name") == "code_editor" for t in tools)

tests/test_lazy_ollama_import.py

@@ -0,0 +1,70 @@
+"""
+回归测试：ollama 必须惰性导入。
+
+原因：ollama SDK 在 import 时会触发平台探测（platform.machine() → WMI），
+在部分沙箱/未安装 ollama 的环境会卡死或拖慢启动。
+修复后：
+- ai_engines/ollama_engine 模块顶层不得 import ollama；
+- OllamaAI.__init__ 不得急切点服务（is_service_running 改为惰性 property），
+  只有真正选用 ollama 引擎时才 import ollama。
+
+本测试只 import 模块、不构造 AICLI（避免触发 LSP 连接等环境依赖），
+因此可在任意沙箱稳定运行，专门守住「ollama 惰性导入」这一改动。
+"""
+import sys
+import types
+import builtins
+
+PROJECT_ROOT = sys.path[0] if sys.path else ""
+# 兜底：确保能找到仓库根
+import os
+ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+if ROOT not in sys.path:
+    sys.path.insert(0, ROOT)
+
+
+def test_ollama_engine_module_import_is_lazy():
+    """import ai_engines.ollama_engine 不应触发 import ollama。"""
+    imported = []
+    real_import = builtins.__import__
+
+    def traced(name, *a, **k):
+        if name == "ollama" or name.startswith("ollama."):
+            imported.append(name)
+        return real_import(name, *a, **k)
+
+    builtins.__import__ = traced
+    try:
+        import ai_engines.ollama_engine as oe
+        # 旧顶层 ollama_client 应已移除
+        assert not hasattr(oe, "ollama_client"), \
+            "旧的顶层 ollama_client 应已移除，改为惰性 _ollama() getter"
+        # 应有惰性 getter
+        assert hasattr(oe, "_ollama"), "应有惰性 _ollama() getter"
+        # 模块被 import 后不应已有 ollama 真身
+        assert oe._ollama_import_tried is False, \
+            "模块加载完成时尚不应尝试 import ollama（惰性）"
+    finally:
+        builtins.__import__ = real_import
+
+    assert imported == [], \
+        f"import ai_engines.ollama_engine 不应触发 import ollama，实际 {imported}"
+
+
+def test_ollama_sdk_only_imported_on_use():
+    """调用 _ollama() 才真正 import ollama；未调用前仍是惰性。"""
+    # 用一个假 ollama 顶替，避免真实 SDK / WMI；只验证「调用时才触发」
+    fake = types.ModuleType("ollama")
+    fake.list = lambda *a, **k: []
+    fake.chat = lambda *a, **k: {}
+    fake.pull = lambda *a, **k: []
+    sys.modules["ollama"] = fake
+
+    import ai_engines.ollama_engine as oe
+    assert oe._ollama_import_tried is False
+    client = oe._ollama()          # 首次调用
+    assert oe._ollama_import_tried is True
+    assert client is fake          # 取回的是我们注入的 ollama
+    # 第二次调用应命中缓存，不重复 import
+    again = oe._ollama()
+    assert again is fake

tests/test_local_lsp_server.py

@@ -0,0 +1,229 @@
+"""自研本地 LSP 服务器测试：单元（分帧/协议/引擎）+ 真实 subprocess 集成"""
+import io
+import json
+import os
+import queue
+import re
+import subprocess
+import sys
+import threading
+
+import pytest
+
+PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+sys.path.insert(0, PROJECT_ROOT)
+
+from xcli_core.local_lsp_server import (
+    LocalLspServer,
+    path_to_uri,
+    read_message,
+    uri_to_path,
+    write_message,
+)
+
+SRC = "import math\ndef add(a, b):\n    return a + b\n\nresult = add(1, 2)\nprint(result)\n"
+URI = "file:///C:/Users/Administrator/Desktop/xiaoli-cli/_test_tmp.py"
+
+
+# ── 分帧传输层 ──
+class TestFraming:
+    def test_roundtrip(self):
+        buf = io.BytesIO()
+        write_message(buf, {"jsonrpc": "2.0", "id": 1, "method": "initialize", "params": {}})
+        buf.seek(0)
+        msg = read_message(buf)
+        assert msg == {"jsonrpc": "2.0", "id": 1, "method": "initialize", "params": {}}
+
+    def test_cjk_bytes(self):
+        # 中文消息体: Content-Length 按字节计
+        obj = {"result": "未定义变量 — 中文诊断"}
+        buf = io.BytesIO()
+        write_message(buf, obj)
+        buf.seek(0)
+        assert read_message(buf) == obj
+
+    def test_empty_stream(self):
+        assert read_message(io.BytesIO(b"")) is None
+
+
+# ── 路径工具 ──
+class TestPath:
+    def test_uri_to_path_windows(self):
+        p = uri_to_path("file:///C:/Users/a/b.py")
+        assert os.path.abspath(p) == os.path.abspath("C:/Users/a/b.py")
+
+    def test_path_to_uri(self):
+        assert path_to_uri(r"C:\Users\a\b.py").startswith("file:///C:/Users/a/b.py")
+
+
+# ── 协议层（进程内直接调 handle） ──
+class TestProtocol:
+    def make_server(self):
+        srv = LocalLspServer()
+        srv.handle({"jsonrpc": "2.0", "method": "initialize", "id": 1,
+                    "params": {"rootUri": "file:///" + PROJECT_ROOT.replace("\\", "/")}})
+        return srv
+
+    def test_initialize_capabilities(self):
+        srv = LocalLspServer()
+        outs = srv.handle({"jsonrpc": "2.0", "method": "initialize", "id": 1, "params": {}})
+        assert outs[0]["id"] == 1
+        caps = outs[0]["result"]["capabilities"]
+        assert caps["definitionProvider"] is True
+        assert "completionProvider" in caps
+        assert caps["textDocumentSync"]["change"] == 1
+
+    def test_unknown_method_error(self):
+        srv = LocalLspServer()
+        outs = srv.handle({"jsonrpc": "2.0", "method": "textDocument/xxx", "id": 9, "params": {}})
+        assert outs[0]["error"]["code"] == -32601
+
+    def test_shutdown_and_unknown_notification_no_reply(self):
+        srv = LocalLspServer()
+        assert srv.handle({"jsonrpc": "2.0", "method": "initialized", "params": {}}) == []
+        assert srv.handle({"jsonrpc": "2.0", "method": "shutdown", "id": 2,
+                           "params": {}})[0]["result"] is None
+
+    def test_did_open_pushes_diagnostics(self):
+        srv = self.make_server()
+        outs = srv.handle({"jsonrpc": "2.0", "method": "textDocument/didOpen", "params": {
+            "textDocument": {"uri": URI, "languageId": "python", "version": 1, "text": SRC}}})
+        push = [o for o in outs if o.get("method") == "textDocument/publishDiagnostics"]
+        assert push, "didOpen 应推送诊断"
+        diags = push[0]["params"]["diagnostics"]
+        assert diags, "应检测到问题 (import math 未使用)"
+        first = diags[0]
+        assert first["source"] == "py_detect"
+        assert "range" in first and "severity" in first and "message" in first
+
+    def test_diagnostics_clean_file_no_push(self):
+        srv = self.make_server()
+        clean = "def add(a, b):\n    return a + b\nprint(add(1, 2))\n"
+        outs = srv.handle({"jsonrpc": "2.0", "method": "textDocument/didOpen", "params": {
+            "textDocument": {"uri": URI, "languageId": "python", "version": 1, "text": clean}}})
+        assert not [o for o in outs if o.get("method") == "textDocument/publishDiagnostics"]
+
+
+# ── jedi 语义（进程内，无需真实文件） ──
+class TestSemantic:
+    def open(self):
+        srv = LocalLspServer()
+        srv.handle({"jsonrpc": "2.0", "method": "textDocument/didOpen", "params": {
+            "textDocument": {"uri": URI, "languageId": "python", "version": 1, "text": SRC}}})
+        return srv
+
+    def req(self, srv, method, line, character):
+        outs = srv.handle({"jsonrpc": "2.0", "method": method, "id": 1, "params": {
+            "textDocument": {"uri": URI}, "position": {"line": line, "character": character}}})
+        return outs[0]["result"]
+
+    def test_completion(self):
+        srv = self.open()
+        res = self.req(srv, "textDocument/completion", 6, 10)  # result.
+        items = res.get("items", []) if isinstance(res, dict) else res
+        labels = [i["label"] for i in items]
+        assert "result" in labels, f"应在 result 后补全: {labels}"
+
+    def test_definition(self):
+        srv = self.open()
+        res = self.req(srv, "textDocument/definition", 5, 10)  # add 调用处
+        assert res, "应找到 add 定义"
+        loc = res[0]
+        assert loc["uri"].startswith("file:///")
+        assert loc["range"]["start"]["line"] == 2  # 定义在第 3 行 (0-based)
+
+    def test_references(self):
+        srv = self.open()
+        res = self.req(srv, "textDocument/references", 5, 10)  # add
+        assert len(res) >= 2, "应至少含定义处+调用处"
+
+    def test_hover(self):
+        srv = self.open()
+        res = self.req(srv, "textDocument/hover", 5, 10)  # add
+        assert res is not None
+        contents = res.get("contents", {})
+        value = contents.get("value", "") if isinstance(contents, dict) else str(contents)
+        assert value, "hover 应有内容"
+
+    def test_hover_builtin(self):
+        srv = self.open()
+        # math 已 import; 悬停 print 是内置, 不崩溃即可
+        res = self.req(srv, "textDocument/hover", 6, 1)
+        assert res is None or isinstance(res, dict)
+
+
+# ── 真实 subprocess 集成（完整握手） ──
+class TestSubprocessIntegration:
+    def test_full_handshake(self):
+        proc = subprocess.Popen(
+            [sys.executable, "-m", "xcli_core.local_lsp_server"],
+            stdin=subprocess.PIPE, stdout=subprocess.PIPE,
+            stderr=subprocess.DEVNULL, cwd=PROJECT_ROOT, bufsize=0)
+        try:
+            q = queue.Queue()
+
+            def _reader():
+                try:
+                    while True:
+                        header = b""
+                        while b"\r\n\r\n" not in header:
+                            chunk = proc.stdout.read(1)
+                            if not chunk:
+                                return
+                            header += chunk
+                        m = re.search(rb"Content-Length:\s*(\d+)", header)
+                        body = proc.stdout.read(int(m.group(1))) if m else b""
+                        if body:
+                            q.put(json.loads(body.decode("utf-8")))
+                except Exception:
+                    pass
+
+            threading.Thread(target=_reader, daemon=True).start()
+
+            def send(obj):
+                data = json.dumps(obj, ensure_ascii=False).encode("utf-8")
+                proc.stdin.write(f"Content-Length: {len(data)}\r\n\r\n".encode() + data)
+                proc.stdin.flush()
+
+            def recv(timeout=25):
+                return q.get(timeout=timeout)
+
+            root_uri = "file:///" + PROJECT_ROOT.replace("\\", "/")
+            send({"jsonrpc": "2.0", "id": 1, "method": "initialize",
+                  "params": {"processId": os.getpid(), "rootUri": root_uri,
+                             "capabilities": {"textDocument": {}}}})
+            r = recv()
+            assert "capabilities" in r.get("result", {}), r
+
+            send({"jsonrpc": "2.0", "method": "initialized", "params": {}})
+            send({"jsonrpc": "2.0", "method": "textDocument/didOpen", "params": {
+                "textDocument": {"uri": URI, "languageId": "python", "version": 1,
+                                 "text": SRC}}})
+            diag = recv()
+            assert diag.get("method") == "textDocument/publishDiagnostics", diag
+            assert diag["params"]["diagnostics"], "应推送诊断"
+
+            send({"jsonrpc": "2.0", "id": 2, "method": "textDocument/completion", "params": {
+                "textDocument": {"uri": URI}, "position": {"line": 6, "character": 10}}})
+            c = recv()["result"]
+            items = c.get("items", []) if isinstance(c, dict) else c
+            assert any(i["label"] == "result" for i in items)
+
+            send({"jsonrpc": "2.0", "id": 3, "method": "textDocument/definition", "params": {
+                "textDocument": {"uri": URI}, "position": {"line": 5, "character": 10}}})
+            d = recv()["result"]
+            assert d and d[0]["range"]["start"]["line"] == 2
+
+            send({"jsonrpc": "2.0", "id": 4, "method": "textDocument/hover", "params": {
+                "textDocument": {"uri": URI}, "position": {"line": 5, "character": 10}}})
+            h = recv()["result"]
+            assert h is not None
+
+            send({"jsonrpc": "2.0", "id": 5, "method": "shutdown", "params": {}})
+            assert recv()["result"] is None
+            send({"jsonrpc": "2.0", "method": "exit", "params": {}})
+            proc.stdin.close()
+            proc.wait(timeout=5)
+        finally:
+            if proc.poll() is None:
+                proc.kill()

tests/test_lsp_client.py

@@ -0,0 +1,375 @@
+"""
+LSP 客户端测试
+==============
+用 inline 的 fake LSP server（Content-Length 分帧 JSON-RPC）验证：
+- LspClient 全链路（initialize / didOpen 触发诊断 / hover / definition / shutdown）
+- LspManager 按扩展名路由、多 server
+- UnifiedToolManager 桥接：5 个 lsp__* 工具注册、get_mcp_tools 可见、execute 返回诊断
+- get_lsp_context 注入
+- fc_tools 保真（lsp 工具进入模型视野）
+全程不触真实引擎 / 网络。
+"""
+
+import os
+import sys
+import json
+import time
+import threading
+
+import pytest
+
+import lsp_client
+from lsp_client import LspClient, LspManager, path_to_uri, ext_to_language
+
+
+# ── Fake LSP server（Content-Length 分帧）──
+FAKE_SERVER_SRC = r'''
+import sys, json, threading, time
+
+def send(obj):
+    data = json.dumps(obj, ensure_ascii=False).encode('utf-8')
+    sys.stdout.buffer.write(b"Content-Length: %d\r\n\r\n" % len(data))
+    sys.stdout.buffer.write(data)
+    sys.stdout.buffer.flush()
+
+def read_msg():
+    header = b""
+    while b"\r\n\r\n" not in header:
+        c = sys.stdin.buffer.read(1)
+        if not c:
+            return None
+        header += c
+    m = __import__("re").search(rb"Content-Length:\s*(\d+)", header)
+    if not m:
+        return None
+    n = int(m.group(1))
+    body = b""
+    while len(body) < n:
+        c = sys.stdin.buffer.read(n - len(body))
+        if not c:
+            return None
+        body += c
+    return json.loads(body.decode("utf-8"))
+
+def main():
+    initialized = threading.Event()
+    while True:
+        msg = read_msg()
+        if msg is None:
+            break
+        mid = msg.get("id")
+        method = msg.get("method")
+        if method == "initialize":
+            send({"jsonrpc": "2.0", "id": mid, "result": {
+                "capabilities": {"hoverProvider": True,
+                                 "definitionProvider": True,
+                                 "referencesProvider": True,
+                                 "completionProvider": {}}}})
+        elif method == "initialized":
+            initialized.set()
+        elif method == "textDocument/didOpen":
+            params = msg.get("params", {})
+            uri = params.get("textDocument", {}).get("uri", "")
+            # 立即推送一条诊断
+            send({"jsonrpc": "2.0", "method": "textDocument/publishDiagnostics",
+                  "params": {"uri": uri, "diagnostics": [
+                      {"range": {"start": {"line": 2, "character": 4},
+                                 "end": {"line": 2, "character": 7}},
+                       "severity": 1, "source": "fake-lsp",
+                       "message": "Undefined variable 'foo'"}]}})
+        elif method == "textDocument/hover":
+            send({"jsonrpc": "2.0", "id": mid, "result": {
+                "contents": {"kind": "markdown", "value": "def foo(): ..."}}})
+        elif method == "textDocument/definition":
+            send({"jsonrpc": "2.0", "id": mid, "result": {
+                "uri": "file:///x.py", "range": {
+                    "start": {"line": 0, "character": 0},
+                    "end": {"line": 0, "character": 3}}}})
+        elif method == "textDocument/references":
+            send({"jsonrpc": "2.0", "id": mid, "result": [
+                {"uri": "file:///x.py", "range": {
+                    "start": {"line": 1, "character": 0},
+                    "end": {"line": 1, "character": 3}}}]})
+        elif method == "textDocument/completion":
+            send({"jsonrpc": "2.0", "id": mid, "result": {
+                "items": [{"label": "bar"}, {"label": "baz"}]}})
+        elif method == "shutdown":
+            send({"jsonrpc": "2.0", "id": mid, "result": None})
+        elif method == "exit":
+            break
+        else:
+            if mid is not None:
+                send({"jsonrpc": "2.0", "id": mid, "result": {}})
+
+if __name__ == "__main__":
+    main()
+'''
+
+
+@pytest.fixture
+def fake_server(tmp_path):
+    p = tmp_path / "fake_lsp_server.py"
+    p.write_text(FAKE_SERVER_SRC)
+    return str(p)
+
+
+def test_path_to_uri_windows():
+    uri = path_to_uri("C:\\Users\\me\\a.py")
+    assert uri.startswith("file:///")
+    assert uri.endswith("/C:/Users/me/a.py")
+
+
+def test_ext_to_language():
+    assert ext_to_language("x.py") == "python"
+    assert ext_to_language("x.rs") == "rust"
+    assert ext_to_language("x.unknown") is None
+
+
+def test_lsp_client_lifecycle(fake_server, tmp_path):
+    code_file = tmp_path / "sample.py"
+    code_file.write_text("print('hi')\nx = 1\nfoo = 2\n")
+
+    client = LspClient("python", sys.executable, [fake_server],
+                       language_id="python", root_uri="file:///tmp")
+    try:
+        client.start()
+        init = client.initialize()
+        assert isinstance(init, dict)
+
+        diags = client.diagnostics(str(code_file), wait=3.0)
+        assert len(diags) == 1
+        assert "Undefined variable" in diags[0]["message"]
+
+        hover = client.hover(str(code_file), 0, 0)
+        assert hover and "foo" in json.dumps(hover)
+
+        definition = client.definition(str(code_file), 0, 0)
+        assert definition
+
+        references = client.references(str(code_file), 0, 0)
+        assert references
+
+        completion = client.completion(str(code_file), 0, 0)
+        assert completion
+    finally:
+        client.shutdown()
+
+
+def test_lsp_manager_routing(fake_server, tmp_path):
+    # 用 .py 扩展名映射到 fake server（languageId=python）
+    servers = {"python": {"command": sys.executable, "args": [fake_server],
+                          "disabled": False}}
+    mgr = LspManager()
+    mgr.connect_all(servers, cwd=str(tmp_path), root_uri="file:///tmp")
+    assert "python" in mgr.get_servers()
+
+    code_file = tmp_path / "sample.py"
+    code_file.write_text("a = 1\nb = 2\nfoo = 3\n")
+    out = mgr.diagnostics(str(code_file))
+    assert "Undefined variable" in out
+    # 诊断被缓存，供上下文注入
+    ctx = mgr.get_context()
+    assert "Undefined variable" in ctx
+
+    hover_out = mgr.hover(str(code_file), 0, 0)
+    assert "foo" in hover_out
+
+
+def test_lsp_manager_unknown_ext_graceful(fake_server, tmp_path):
+    servers = {"python": {"command": sys.executable, "args": [fake_server]}}
+    mgr = LspManager()
+    mgr.connect_all(servers, cwd=str(tmp_path), root_uri="file:///tmp")
+    # .rs 没有配置 server，应优雅提示而非崩溃
+    out = mgr.diagnostics(str(tmp_path / "x.rs"))
+    assert "没有为" in out or "LSP server" in out
+
+
+def test_lsp_manager_connect_failure_graceful(tmp_path):
+    # 命令不存在 → 不应抛异常，仅记录
+    servers = {"python": {"command": "this_command_does_not_exist_xyz"}}
+    mgr = LspManager()
+    mgr.connect_all(servers, cwd=str(tmp_path), root_uri="file:///tmp")
+    assert mgr.get_servers() == []
+
+
+def test_unified_manager_bridges_lsp(fake_server, tmp_path, monkeypatch):
+    code_file = tmp_path / "sample.py"
+    code_file.write_text("a = 1\nb = 2\nfoo = 3\n")
+    cfg = {
+        "mcpServers": {},
+        "lspServers": {"python": {"command": sys.executable, "args": [fake_server],
+                                  "disabled": False}},
+    }
+
+    import xcli_core.config as xcfg
+    monkeypatch.setattr(xcfg, "load_config", lambda: cfg)
+    try:
+        import config as root_cfg
+        monkeypatch.setattr(root_cfg, "load_config", lambda: cfg)
+    except ImportError:
+        pass
+
+    import unified_tool_manager as utm
+    mgr = utm.UnifiedToolManager()
+    mgr.initialize(
+        plugins_dir=str(tmp_path / "noplugins"),
+        skills_dir=str(tmp_path / "noskills"),
+    )
+    try:
+        tools = mgr.get_mcp_tools()
+        names = [t["name"] for t in tools]
+        for expected in ("lsp__diagnostics", "lsp__hover", "lsp__definition",
+                         "lsp__references", "lsp__completion"):
+            assert expected in names, f"缺少 LSP 工具: {expected}"
+
+        # inputSchema 保真（不压成 args）
+        diag_def = next(t for t in tools if t["name"] == "lsp__diagnostics")
+        assert "file" in diag_def["inputSchema"]["properties"]
+
+        # execute 返回真实诊断文本
+        res = mgr.execute("lsp__diagnostics", json.dumps({"file": str(code_file)}))
+        assert res.success
+        assert "Undefined variable" in (res.data or "")
+
+        # 上下文注入可用
+        ctx = mgr.get_lsp_context()
+        assert "Undefined variable" in ctx
+    finally:
+        mgr.shutdown()
+
+
+def test_lsp_skip_when_no_servers_configured(tmp_path, monkeypatch):
+    # noch keinen LSP-Server -> keine lsp__ Tools
+    cfg = {"mcpServers": {}, "lspServers": {}}
+    import xcli_core.config as xcfg
+    monkeypatch.setattr(xcfg, "load_config", lambda: cfg)
+    try:
+        import config as root_cfg
+        monkeypatch.setattr(root_cfg, "load_config", lambda: cfg)
+    except ImportError:
+        pass
+    import unified_tool_manager as utm
+    mgr = utm.UnifiedToolManager()
+    mgr.initialize(
+        plugins_dir=str(tmp_path / "noplugins"),
+        skills_dir=str(tmp_path / "noskills"),
+    )
+    try:
+        names = [t["name"] for t in mgr.get_mcp_tools()]
+        assert not any(n.startswith("lsp__") for n in names)
+    finally:
+        mgr.shutdown()
+
+
+def test_fc_tools_includes_lsp(tmp_path, monkeypatch, fake_server):
+    """验证 lsp__* 工具能经 get_mcp_tools 进入 fc_tools 保真输出"""
+    import xcli_core.config as xcfg
+    cfg = {"mcpServers": {},
+           "lspServers": {"python": {"command": sys.executable, "args": [fake_server]}}}
+    monkeypatch.setattr(xcfg, "load_config", lambda: cfg)
+    try:
+        import config as root_cfg
+        monkeypatch.setattr(root_cfg, "load_config", lambda: cfg)
+    except ImportError:
+        pass
+
+    import unified_tool_manager as utm
+    import xcli_core.fc_tools as fc_tools
+
+    mgr = utm.UnifiedToolManager()
+    mgr.initialize(
+        plugins_dir=str(tmp_path / "noplugins"),
+        skills_dir=str(tmp_path / "noskills"),
+    )
+    try:
+        fc = fc_tools.mcp_to_openai_tools(mgr)
+        lsp_fc = [t for t in fc if t.get("function", {}).get("name", "").startswith("lsp__")]
+        assert len(lsp_fc) == 5, f"应进入 5 个 lsp 工具，实际 {len(lsp_fc)}"
+        # 保真：参数应是 file/line/character，而非单个 args
+        diag = next(t for t in lsp_fc if t["function"]["name"] == "lsp__diagnostics")
+        assert "file" in diag["function"]["parameters"]["properties"]
+    finally:
+        mgr.shutdown()
+
+
+# ── code_editor / file_manager 的「写 .py 后自动触发 pylsp 诊断」钩子 ──
+def _make_fake_lsp_manager(diag_text):
+    """构造一个 fake LSP 管理器，记录被调用，并返回固定诊断文本。"""
+    calls = []
+
+    class FakeManager:
+        def __init__(self):
+            self.clients = {"python": object()}  # 模拟已连接 python server
+
+        def diagnostics(self, filepath, wait=1.5):
+            calls.append(filepath)
+            return diag_text
+
+    fm = FakeManager()
+    fm._calls = calls
+    return fm
+
+
+def _attach(cli, fm):
+    class UTM:
+        _lsp_manager = fm
+    cli.liugin_manager = UTM()
+
+
+def test_auto_lsp_check_triggers_on_py_write(tmp_path):
+    """写/改 .py 后钩子应调用 LSP 诊断，并在有问题时返回概要。"""
+    from plugins.code_editor import Liugin as CodeEditor
+    ce = CodeEditor()
+    fm = _make_fake_lsp_manager("🔍 x.py：2 条诊断\n  错误 L1:1 — undefined name 'foo'")
+    cli = type("CLI", (), {})()
+    _attach(cli, fm)
+    ce.set_cli(cli)
+
+    py_file = tmp_path / "mod.py"
+    py_file.write_text("foo = bar\n", encoding="utf-8")
+    out = ce._auto_lsp_check(str(py_file))
+    assert str(py_file) in fm._calls, "应调用 lsp_manager.diagnostics(该 py 文件)"
+    assert "LSP 诊断" in out and "2 条诊断" in out
+
+
+def test_auto_lsp_check_silent_when_clean(tmp_path):
+    """文件干净（无诊断）时钩子应静默返回空串。"""
+    from plugins.code_editor import Liugin as CodeEditor
+    ce = CodeEditor()
+    fm = _make_fake_lsp_manager("✅ mod.py：无诊断（未发现问题）")
+    cli = type("CLI", (), {})()
+    _attach(cli, fm)
+    ce.set_cli(cli)
+
+    py_file = tmp_path / "clean.py"
+    py_file.write_text("x = 1\n", encoding="utf-8")
+    assert ce._auto_lsp_check(str(py_file)) == ""
+
+
+def test_auto_lsp_check_silent_without_lsp_manager(tmp_path):
+    """未连接 LSP 管理器（或未配置 server）时钩子应静默跳过，绝不报错。"""
+    from plugins.code_editor import Liugin as CodeEditor
+    ce = CodeEditor()
+    cli = type("CLI", (), {})()
+    # liugin_manager 没有 _lsp_manager 属性
+    cli.liugin_manager = type("UTM", (), {})()
+    ce.set_cli(cli)
+
+    py_file = tmp_path / "any.py"
+    py_file.write_text("pass\n", encoding="utf-8")
+    assert ce._auto_lsp_check(str(py_file)) == ""
+
+
+def test_auto_lsp_check_no_server_for_ext(tmp_path):
+    """扩展名未配置 LSP server 时（如 .txt）应静默返回空串。"""
+    from plugins.code_editor import Liugin as CodeEditor
+    ce = CodeEditor()
+    fm = _make_fake_lsp_manager("⚠️ 没有为 x.txt 配置 LSP server")
+    # clients 里只有 python，没有 txt 对应的 server；这里用文本模拟「无 server」返回
+    cli = type("CLI", (), {})()
+    _attach(cli, fm)
+    ce.set_cli(cli)
+
+    txt_file = tmp_path / "note.txt"
+    txt_file.write_text("hello\n", encoding="utf-8")
+    assert ce._auto_lsp_check(str(txt_file)) == ""

tests/test_max_input_tokens.py

@@ -0,0 +1,66 @@
+"""max_input_tokens 统一 API + token 感知压缩 回归测试
+
+验证:
+1. 所有引擎（openai / ollama / manual）都暴露 max_input_tokens 且返回正整数（API 契约）。
+2. 系统依据 max_input_tokens 的 70% 自动判定压缩（should_compress）。
+"""
+import os
+import sys
+
+import pytest
+
+PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+if PROJECT_ROOT not in sys.path:
+    sys.path.insert(0, PROJECT_ROOT)
+
+
+def _engine_classes():
+    from ai_engines.openai_engine import OpenaiAI
+    from ai_engines.ollama_engine import OllamaAI
+    from ai_engines.manual_engine import ManualAI
+    return {"openai": OpenaiAI, "ollama": OllamaAI, "manual": ManualAI}
+
+
+def test_all_engines_expose_max_input_tokens():
+    """所有引擎必须返回 max_input_tokens（统一 API 契约）"""
+    for name, cls in _engine_classes().items():
+        assert hasattr(cls, "max_input_tokens"), f"{name} 引擎缺少 max_input_tokens 属性"
+
+
+def test_max_input_tokens_positive_when_instantiated():
+    """实例化后 max_input_tokens 应为正整数"""
+    for name, cls in _engine_classes().items():
+        try:
+            inst = cls()
+        except Exception as e:
+            pytest.skip(f"{name} 引擎实例化失败（环境/网络），跳过实例检查: {e}")
+        val = inst.max_input_tokens
+        assert isinstance(val, int) and val > 0, f"{name} 的 max_input_tokens 非法: {val!r}"
+
+
+def _mgr():
+    from xcli_core.memory import MemoryManager
+    return MemoryManager(project_dir=PROJECT_ROOT)
+
+
+def _conv(n):
+    return [{"role": "user", "content": f"msg {i}"} for i in range(n)]
+
+
+def test_should_compress_token_aware_triggers_at_70pct():
+    mgr = _mgr()
+    # 7000 / 10000 = 0.7 >= 0.7 → 触发
+    assert mgr.should_compress(_conv(5), prompt_tokens=7000, max_input_tokens=10000, ratio=0.7) is True
+
+
+def test_should_compress_token_aware_below_threshold():
+    mgr = _mgr()
+    # 6000 / 10000 = 0.6 < 0.7 → 不触发（消息数也未超阈值）
+    assert mgr.should_compress(_conv(5), prompt_tokens=6000, max_input_tokens=10000, ratio=0.7) is False
+
+
+def test_should_compress_message_count_fallback():
+    mgr = _mgr()
+    # 无 token 信息，仅消息数 > 50 → 兜底触发
+    assert mgr.should_compress(_conv(60)) is True
+    assert mgr.should_compress(_conv(10)) is False

tests/test_mcp_client.py

@@ -0,0 +1,193 @@
+"""
+MCP 客户端测试
+==============
+用 inline 的 fake MCP server（写入临时文件，subprocess 启动）验证：
+
+1. McpClient 全链路：start → initialize → list_tools → call_tool → stop（进程回收）
+2. McpClientManager 多 server 并发连接、聚合、调用、cleanup 幂等
+3. 超时：沉默 server 不响应 → call_tool 应在 timeout 内抛错
+4. UnifiedToolManager 桥接：外部 MCP 工具按 {server}__{tool} 注册、schema 保真、可被 execute 调用、shutdown 清理子进程
+5. fc_tools.mcp_to_openai_tools 优先走 get_mcp_tools 保真分支（不把 schema 压成 args）
+
+全程不依赖真实 AI 引擎、网络或外部 npx。
+"""
+
+import os
+import sys
+
+import pytest
+
+# 确保项目根在 sys.path（unified_tool_manager / mcp_client / config 均在根目录）
+ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+if ROOT not in sys.path:
+    sys.path.insert(0, ROOT)
+
+from mcp_client import McpClient, McpClientManager, McpClientError  # noqa: E402
+import unified_tool_manager as utm  # noqa: E402
+from xcli_core.fc_tools import mcp_to_openai_tools  # noqa: E402
+
+# fake MCP server：实现 initialize / tools/list / tools/call / ping，
+# 以及一个 echo 工具（inputSchema 保真）。与 mcp_server.py 协议对称。
+FAKE_SERVER_SRC = r'''
+import sys, json
+
+def send(obj):
+    sys.stdout.write(json.dumps(obj, ensure_ascii=False) + "\n")
+    sys.stdout.flush()
+
+def main():
+    for line in sys.stdin:
+        line = line.strip()
+        if not line:
+            continue
+        try:
+            req = json.loads(line)
+        except Exception:
+            continue
+        method = req.get("method", "")
+        mid = req.get("id")
+        if method == "notifications/initialized":
+            continue
+        if method == "initialize":
+            send({"jsonrpc": "2.0", "id": mid, "result": {
+                "protocolVersion": "2024-11-05",
+                "capabilities": {"tools": {}},
+                "serverInfo": {"name": "fake", "version": "1.0"},
+            }})
+        elif method == "tools/list":
+            send({"jsonrpc": "2.0", "id": mid, "result": {"tools": [{
+                "name": "echo",
+                "description": "回声工具",
+                "inputSchema": {
+                    "type": "object",
+                    "properties": {"text": {"type": "string", "description": "要回声的文本"}},
+                    "required": ["text"],
+                },
+            }]}})
+        elif method == "tools/call":
+            params = req.get("params", {})
+            name = params.get("name")
+            args = params.get("arguments", {})
+            if name == "echo":
+                send({"jsonrpc": "2.0", "id": mid, "result": {
+                    "content": [{"type": "text", "text": "echo: " + str(args.get("text", ""))}],
+                }})
+            else:
+                send({"jsonrpc": "2.0", "id": mid, "error": {"code": -32601, "message": "unknown tool"}})
+        elif method == "ping":
+            send({"jsonrpc": "2.0", "id": mid, "result": {}})
+        else:
+            send({"jsonrpc": "2.0", "id": mid, "error": {"code": -32601, "message": "method not found"}})
+
+if __name__ == "__main__":
+    main()
+'''
+
+
+@pytest.fixture
+def fake_server_path(tmp_path):
+    p = tmp_path / "fake_mcp_server.py"
+    p.write_text(FAKE_SERVER_SRC)
+    return str(p)
+
+
+def test_mcp_client_lifecycle(fake_server_path):
+    client = McpClient("fake", sys.executable, [fake_server_path])
+    try:
+        client.start()
+        info = client.initialize()
+        assert info.get("protocolVersion") == "2024-11-05"
+        tools = client.list_tools()
+        assert any(t.get("name") == "echo" for t in tools)
+        out = client.call_tool("echo", {"text": "hi"})
+        assert "hi" in out
+    finally:
+        client.stop()
+    # 进程应已被回收
+    assert client.proc.poll() is not None
+
+
+def test_manager_multi_server(fake_server_path):
+    mgr = McpClientManager()
+    mgr.connect_all({
+        "a": {"command": sys.executable, "args": [fake_server_path]},
+        "b": {"command": sys.executable, "args": [fake_server_path]},
+    })
+    assert set(mgr.get_servers()) == {"a", "b"}
+    assert len(mgr.get_all_tools()) == 2
+    out = mgr.call_tool("a", "echo", {"text": "x"})
+    assert "x" in out
+    mgr.cleanup()
+    mgr.cleanup()  # 幂等
+
+
+def test_client_timeout(tmp_path):
+    silent = tmp_path / "silent.py"
+    silent.write_text("import time\nwhile True:\n    time.sleep(1)\n")
+    client = McpClient("silent", sys.executable, [str(silent)], timeout=2)
+    try:
+        client.start()
+        with pytest.raises(Exception):
+            client.list_tools()  # 沉默 server 不响应 → 超时
+    finally:
+        client.stop()
+
+
+def test_unified_manager_bridges_external_mcp(tmp_path, monkeypatch):
+    fake = tmp_path / "fake_mcp_server.py"
+    fake.write_text(FAKE_SERVER_SRC)
+    cfg = {"mcpServers": {"fake": {"command": sys.executable, "args": [str(fake)]}}}
+
+    import xcli_core.config as xcfg
+    monkeypatch.setattr(xcfg, "load_config", lambda: cfg)
+    try:
+        import config as root_cfg
+        monkeypatch.setattr(root_cfg, "load_config", lambda: cfg)
+    except ImportError:
+        pass
+
+    mgr = utm.UnifiedToolManager()
+    # 用不存在的插件/技能目录，避免加载真实插件
+    mgr.initialize(
+        plugins_dir=str(tmp_path / "noplugins"),
+        skills_dir=str(tmp_path / "noskills"),
+    )
+    try:
+        tool = mgr.get_tool("fake__echo")
+        assert tool is not None, "外部 MCP 工具应已桥接为 fake__echo"
+        assert tool["protocol"] == "mcp"
+
+        names = [t["name"] for t in mgr.get_mcp_tools()]
+        assert "fake__echo" in names
+
+        echo_def = next(t for t in mgr.get_mcp_tools() if t["name"] == "fake__echo")
+        assert "text" in echo_def["inputSchema"]["properties"]
+
+        res = mgr.execute("fake__echo", '{"text": "hello"}')
+        assert res.success
+        assert "hello" in (res.data or "")
+    finally:
+        mgr.shutdown()
+
+
+def test_fc_tools_uses_get_mcp_tools():
+    class FakeMgr:
+        def get_mcp_tools(self):
+            return [{
+                "name": "my_tool",
+                "description": "d",
+                "inputSchema": {
+                    "type": "object",
+                    "properties": {"q": {"type": "string"}},
+                    "required": ["q"],
+                },
+            }]
+
+    tools = mcp_to_openai_tools(FakeMgr())
+    my = next(t for t in tools if t["function"]["name"] == "my_tool")
+    # schema 保真：不被压成 {"args": string}
+    assert "q" in my["function"]["parameters"]["properties"]
+    assert "args" not in my["function"]["parameters"]["properties"]
+    # 内置工具（memory / scheduler）仍被追加
+    names = [t["function"]["name"] for t in tools]
+    assert "memory" in names and "scheduler" in names

tests/test_model_registry.py

@@ -0,0 +1,188 @@
+"""模型注册表 + /model 命令的回归测试。
+
+不污染真实 config.json：所有读写都指向临时文件。
+"""
+import os
+import json
+import tempfile
+
+import pytest
+
+from xcli_core import model_registry as mr
+import xcli_core.config as cfgmod
+import ai_engines.openai_engine as oe_mod
+
+
+@pytest.fixture
+def tmp_config():
+    """准备临时 config.json，并把注册表与 openai 引擎的读取都重定向过去。"""
+    d = tempfile.mkdtemp()
+    path = os.path.join(d, "config.json")
+    payload = {
+        "api": {"engines": {
+            "openai": {"api_key": "K0", "base_url": "https://old/v1", "model": "old-model"}
+        }},
+        "system": {},
+    }
+    with open(path, "w", encoding="utf-8") as f:
+        json.dump(payload, f, ensure_ascii=False, indent=2)
+
+    saved_path = mr._CONFIG_PATH
+    saved_proj = oe_mod.project_root
+    mr._set_config_path(path)
+    oe_mod.project_root = d
+    try:
+        yield path
+    finally:
+        mr._set_config_path(saved_path)
+        oe_mod.project_root = saved_proj
+
+
+# ── 注册表单元 ──
+
+def test_legacy_flat_config_is_single_implicit_model(tmp_config):
+    models, current = mr.list_models()
+    assert len(models) == 1
+    assert models[0]["model"] == "old-model"
+    assert current == "old-model"
+
+
+def test_add_first_model_migrates_flat_and_keeps_current(tmp_config):
+    mr.add_model({"name": "gpt4o", "base_url": "https://api.openai.com/v1",
+                  "api_key": "sk", "model": "gpt-4o"})
+    models, current = mr.list_models()
+    names = [m["name"] for m in models]
+    assert names == ["old-model", "gpt4o"]
+    # current 不应跳变到新加的模型
+    assert current == "old-model"
+
+
+def test_add_duplicate_name_updates_in_place(tmp_config):
+    mr.add_model({"name": "gpt4o", "base_url": "u1", "api_key": "k1", "model": "gpt-4o"})
+    mr.add_model({"name": "gpt4o", "base_url": "u2", "api_key": "k2", "model": "gpt-4o-mini"})
+    models, _ = mr.list_models()
+    assert len(models) == 2  # 未重复追加
+    assert models[1]["base_url"] == "u2"
+
+
+def test_set_current_and_get_model(tmp_config):
+    mr.add_model({"name": "gpt4o", "base_url": "https://api.openai.com/v1",
+                  "api_key": "sk", "model": "gpt-4o"})
+    assert mr.set_current("gpt4o") is True
+    m = mr.get_model()
+    assert m["model"] == "gpt-4o"
+    assert mr.set_current("ghost") is False  # 越权返回 False
+
+
+def test_remove_current_falls_back_to_first(tmp_config):
+    mr.add_model({"name": "gpt4o", "base_url": "u", "api_key": "k", "model": "gpt-4o"})
+    mr.set_current("gpt4o")
+    assert mr.remove_model("gpt4o") is True
+    _, current = mr.list_models()
+    assert current == "old-model"
+
+
+def test_remove_missing_returns_false(tmp_config):
+    assert mr.remove_model("nope") is False
+
+
+def test_persisted_to_disk(tmp_config):
+    mr.add_model({"name": "gpt4o", "base_url": "u", "api_key": "k", "model": "gpt-4o"})
+    mr.set_current("gpt4o")
+    with open(tmp_config, "r", encoding="utf-8") as f:
+        saved = json.load(f)
+    block = saved["api"]["engines"]["openai"]
+    assert block["current"] == "gpt4o"
+    assert any(m["name"] == "gpt4o" for m in block["models"])
+
+
+# ── 引擎接入 ──
+
+def test_engine_init_applies_registry_current(tmp_config):
+    mr.add_model({"name": "gpt4o", "base_url": "https://api.openai.com/v1",
+                  "api_key": "sk", "model": "gpt-4o"})
+    mr.set_current("gpt4o")
+    eng = oe_mod.OpenaiAI()
+    assert eng.model == "gpt-4o"
+    assert "api.openai.com" in eng.base_url
+    assert eng.current_model_name() == "gpt4o"
+
+
+def test_engine_set_registry_model(tmp_config):
+    eng = oe_mod.OpenaiAI()
+    eng.add_model({"name": "gpt4o", "base_url": "https://api.openai.com/v1",
+                   "api_key": "sk", "model": "gpt-4o"})
+    assert eng.set_registry_model("gpt4o") is True
+    assert eng.model == "gpt-4o"
+    assert eng.set_registry_model("ghost") is False
+
+
+def test_engine_does_not_clobber_existing_set_model(tmp_config):
+    """原有轻量 set_model（只改 model 名）必须仍然可用，不被注册表方法覆盖。"""
+    eng = oe_mod.OpenaiAI()
+    eng.set_model("gpt-4o-mini")  # 原有方法：只改 self.model
+    assert eng.model == "gpt-4o-mini"
+    # 注册表方法仍在
+    assert hasattr(eng, "set_registry_model")
+    assert hasattr(eng, "list_registry_models")
+
+
+# ── CLI 命令路由 ──
+
+def _make_fake_cli(eng):
+    """复制 cli_base 的模型命令方法到轻量对象，避免实例化整个 CliBase。"""
+    from xcli_core import cli_base
+    cls = type("FakeCli", (), {})
+    cls.handle_model_command = cli_base.BaseAICLI.handle_model_command
+    cls._model_list = cli_base.BaseAICLI._model_list
+    cls._model_switch = cli_base.BaseAICLI._model_switch
+    cls._model_add_interactive = cli_base.BaseAICLI._model_add_interactive
+    return cls()
+
+
+def test_cli_model_list_and_switch(tmp_config, monkeypatch):
+    from xcli_core import cli_base
+    eng = oe_mod.OpenaiAI()
+    eng.add_model({"name": "gpt4o", "base_url": "u", "api_key": "k", "model": "gpt-4o"})
+
+    cli = _make_fake_cli(eng)
+    cli.current_engine = eng
+    cli.current_model = None
+
+    # 列出
+    captured = {}
+    monkeypatch.setattr("builtins.print", lambda *a, **k: captured.setdefault("out", []).append(a))
+    cli.handle_model_command("list")
+    assert any("gpt4o" in str(x) for x in captured["out"])
+
+    # 切换
+    cli.handle_model_command("gpt4o")
+    assert cli.current_model == "gpt4o"
+    assert eng.model == "gpt-4o"
+
+
+def test_cli_model_switch_missing(tmp_config, monkeypatch):
+    eng = oe_mod.OpenaiAI()
+    cli = _make_fake_cli(eng)
+    cli.current_engine = eng
+    cli.current_model = None
+    captured = {}
+    monkeypatch.setattr("builtins.print", lambda *a, **k: captured.setdefault("out", []).append(a))
+    cli.handle_model_command("ghost")
+    assert any("未找到" in str(x) for x in captured["out"])
+    assert cli.current_model is None
+
+
+def test_cli_model_add_interactive(tmp_config, monkeypatch):
+    eng = oe_mod.OpenaiAI()
+    cli = _make_fake_cli(eng)
+    cli.current_engine = eng
+    cli.current_model = None
+    answers = iter(["mygpt", "https://x/v1", "sk-1", "gpt-x"])
+    monkeypatch.setattr("builtins.input", lambda *a, **k: next(answers))
+    captured = {}
+    monkeypatch.setattr("builtins.print", lambda *a, **k: captured.setdefault("out", []).append(a))
+    cli.handle_model_command("add")
+    models, current = eng.list_registry_models()
+    assert any(m["name"] == "mygpt" for m in models)
+    assert cli.current_model == "mygpt"

tests/test_permissions.py

@@ -0,0 +1,116 @@
+"""声明式权限（allow/ask/deny）回归测试。不触真实引擎/网络。"""
+import pytest
+from xcli_core.safety import SafetyLayer, MODE_UNRESTRICTED, MODE_NORMAL, MODE_MANUAL
+
+
+def _new_safety(allow=None, ask=None, deny=None, mode=MODE_NORMAL):
+    s = SafetyLayer()
+    s.set_permissions(allow=allow, ask=ask, deny=deny)
+    s.mode = mode
+    return s
+
+
+# ── 纯函数：规则匹配 ──
+
+def test_match_wildcard():
+    s = SafetyLayer()
+    assert s._permission_match("*", "anything", "rm -rf /") is True
+
+
+def test_match_exact_tool():
+    s = SafetyLayer()
+    assert s._permission_match("git_tools", "git_tools", "status") is True
+    assert s._permission_match("git_tools", "cmd_executor", "status") is False
+
+
+def test_match_tool_star():
+    s = SafetyLayer()
+    assert s._permission_match("git_tools:*", "git_tools", "push origin") is True
+
+
+def test_match_tool_op_prefix():
+    s = SafetyLayer()
+    assert s._permission_match("cmd_executor:rm", "cmd_executor", "rm -rf /tmp") is True
+    assert s._permission_match("cmd_executor:rm", "cmd_executor", "ls -la") is False
+    # 操作词精确相等也应命中
+    assert s._permission_match("git_tools:push", "git_tools", "push") is True
+
+
+def test_match_empty_rule_ignored():
+    s = SafetyLayer()
+    assert s._permission_match("", "git_tools", "status") is False
+
+
+# ── deny 最高优先级 ──
+
+def test_deny_blocks():
+    s = _new_safety(deny=["git_tools:reset"], mode=MODE_UNRESTRICTED)
+    ok, msg = s.check("git_tools", "reset --hard HEAD")
+    assert ok is False
+    assert "deny" in msg
+
+
+def test_deny_blocks_even_unrestricted():
+    s = _new_safety(deny=["cmd_executor:rm"], mode=MODE_UNRESTRICTED)
+    ok, _ = s.check("cmd_executor", "rm -rf /")
+    assert ok is False
+
+
+def test_deny_overrides_allow():
+    # 同一工具同时出现在 deny 与 allow，deny 胜出
+    s = _new_safety(deny=["git_tools:reset"], allow=["git_tools:reset"], mode=MODE_NORMAL)
+    ok, _ = s.check("git_tools", "reset --hard")
+    assert ok is False
+
+
+# ── ask 强制确认 ──
+
+def test_ask_forces_confirm():
+    s = _new_safety(ask=["cmd_executor"], mode=MODE_NORMAL)
+    confirmed = []
+    s._ask_user_confirm = lambda *a, **k: (confirmed.append(True) or (True, "ok"))
+    ok, _ = s.check("cmd_executor", "ls")
+    assert ok is True
+    assert confirmed == [True]
+
+
+def test_ask_overrides_allow():
+    # 同时 allow 与 ask，ask（确认）胜出，不会自动放行
+    s = _new_safety(allow=["cmd_executor"], ask=["cmd_executor"], mode=MODE_NORMAL)
+    called = []
+    s._ask_user_confirm = lambda *a, **k: (called.append(True) or (True, "ok"))
+    s.check("cmd_executor", "ls")
+    assert called == [True]
+
+
+def test_ask_works_in_unrestricted():
+    s = _new_safety(ask=["git_tools:push"], mode=MODE_UNRESTRICTED)
+    called = []
+    s._ask_user_confirm = lambda *a, **k: (called.append(True) or (False, "no"))
+    ok, _ = s.check("git_tools", "push origin master")
+    assert ok is False
+    assert called == [True]
+
+
+# ── allow 自动放行 ──
+
+def test_allow_auto_passes():
+    s = _new_safety(allow=["git_tools:status"], mode=MODE_NORMAL)
+    # 不应触发 AI 分析 / 确认
+    s._ask_user_confirm = lambda *a, **k: (False, "SHOULD_NOT_BE_CALLED")
+    ok, msg = s.check("git_tools", "status")
+    assert ok is True
+    assert msg == ""
+
+
+def test_allow_no_match_falls_through():
+    s = _new_safety(allow=["git_tools:status"], mode=MODE_UNRESTRICTED)
+    ok, _ = s.check("cmd_executor", "ls")
+    # 无规则命中 → 走默认（无限制模式直接放行）
+    assert ok is True
+
+
+def test_default_without_rules_unrestricted_passes():
+    s = _new_safety(mode=MODE_UNRESTRICTED)
+    ok, _ = s.check("file_manager", "read foo.txt")
+    assert ok is True

tests/test_plan_mode.py

@@ -0,0 +1,249 @@
+"""Plan 模式（只读规划 → 审批 → 执行）回归测试。
+
+全程不实例化 AICLI（避免加载真实引擎/插件/进程保护/网络），
+通过伪造 cli 对象与轻量 SimpleNamespace 验证核心逻辑。
+"""
+import types
+
+import pytest
+
+from xcli_core.safety import (
+    get_safety,
+    plan_is_write_operation,
+    PLAN_WRITE_RULES,
+)
+
+
+# ── 纯函数：写操作判定 ──
+
+def test_write_rules_known_and_unknown():
+    # 整类拦截
+    assert plan_is_write_operation("cmd_executor", "ls -la") is True
+    assert plan_is_write_operation("browser_auto", "navigate https://x") is True
+    assert plan_is_write_operation("network_tools", "ping 1.1.1.1") is True
+    assert plan_is_write_operation("scheduler", "add 1h do") is True
+    # 整类放行
+    assert plan_is_write_operation("tool_search", "code_editor") is False
+    assert plan_is_write_operation("ai_search", "python 教程") is False
+    assert plan_is_write_operation("code_search", "find . kw") is False
+    # 未知工具：保守拦截
+    assert plan_is_write_operation("mystery_tool", "anything") is True
+
+
+def test_code_editor_readonly_allowed_write_blocked():
+    assert plan_is_write_operation("code_editor", "read_range a.py 1 10") is False
+    assert plan_is_write_operation("code_editor", "find . kw") is False
+    assert plan_is_write_operation("code_editor", "diff a.py") is False
+    assert plan_is_write_operation("code_editor", "edit a.py old <<<>>> new") is True
+    assert plan_is_write_operation("code_editor", "write a.py content") is True
+    assert plan_is_write_operation("code_editor", "append a.py line") is True
+    assert plan_is_write_operation("code_editor", "multi a.py [...]") is True
+    assert plan_is_write_operation("code_editor", "insert a.py 3 line") is True
+    assert plan_is_write_operation("code_editor", "delete_lines a.py 1 2") is True
+
+
+def test_git_readonly_allowed_write_blocked():
+    assert plan_is_write_operation("git_tools", "status") is False
+    assert plan_is_write_operation("git_tools", "log --oneline 10") is False
+    assert plan_is_write_operation("git_tools", "diff") is False
+    assert plan_is_write_operation("git_tools", "show HEAD") is False
+    assert plan_is_write_operation("git_tools", "branch") is False  # 仅列出
+    assert plan_is_write_operation("git_tools", "commit 修复") is True
+    assert plan_is_write_operation("git_tools", "smart-commit") is True  # 后缀命中
+    assert plan_is_write_operation("git_tools", "push") is True
+    assert plan_is_write_operation("git_tools", "checkout main") is True
+    assert plan_is_write_operation("git_tools", "branch -d old") is True
+    assert plan_is_write_operation("git_tools", "reset --hard") is True
+    assert plan_is_write_operation("git_tools", "fetch origin") is True
+
+
+def test_file_manager_sub_agent_auto_engineer():
+    assert plan_is_write_operation("file_manager", "list .") is False
+    assert plan_is_write_operation("file_manager", "read a.txt") is False
+    assert plan_is_write_operation("file_manager", "search kw") is False
+    assert plan_is_write_operation("file_manager", "write a.txt x") is True
+    assert plan_is_write_operation("file_manager", "delete a.txt") is True
+
+    assert plan_is_write_operation("sub_agent", "list") is False
+    assert plan_is_write_operation("sub_agent", "status") is False
+    assert plan_is_write_operation("sub_agent", "result 12") is False
+    assert plan_is_write_operation("sub_agent", "run 任务") is True
+    assert plan_is_write_operation("sub_agent", "dispatch 任务") is True
+
+    assert plan_is_write_operation("auto_engineer", "lint") is False  # 只读
+    assert plan_is_write_operation("auto_engineer", "metrics") is False
+    assert plan_is_write_operation("auto_engineer", "test") is True
+    assert plan_is_write_operation("auto_engineer", "build") is True
+
+
+# ── safety.check 在 plan 模式下的拦截 ──
+
+@pytest.fixture
+def safety_with_fake_cli():
+    safety = get_safety()
+    saved_cli = safety.cli
+    saved_mode = safety.mode
+    fake_cli = types.SimpleNamespace(plan_mode=True)
+    safety.set_cli(fake_cli)
+    yield safety
+    safety.set_cli(saved_cli)
+    safety.set_mode(saved_mode)
+
+
+def test_plan_mode_blocks_write_at_safety_layer(safety_with_fake_cli):
+    s = safety_with_fake_cli
+    allowed, msg = s.check("code_editor", "edit a.py x <<<>>> y")
+    assert allowed is False
+    assert "PLAN" in msg
+
+    allowed, _ = s.check("git_tools", "commit 修复")
+    assert allowed is False
+
+    allowed, _ = s.check("cmd_executor", "rm -rf /")
+    assert allowed is False
+
+
+def test_plan_mode_allows_readonly_at_safety_layer(safety_with_fake_cli):
+    s = safety_with_fake_cli
+    allowed, _ = s.check("code_editor", "read_range a.py 1 10")
+    assert allowed is True
+    allowed, _ = s.check("tool_search", "code_editor")
+    assert allowed is True
+    allowed, _ = s.check("git_tools", "status")
+    assert allowed is True
+
+
+def test_normal_mode_does_not_plan_block():
+    """非 plan 模式：safety.check 不被 plan 逻辑拦截（用无限制模式避免触发 AI 分析）。"""
+    safety = get_safety()
+    saved_cli = safety.cli
+    saved_mode = safety.mode
+    saved_perm = safety._permissions_override
+    fake_cli = types.SimpleNamespace(plan_mode=False)
+    safety.set_cli(fake_cli)
+    safety.set_mode(0)  # MODE_UNRESTRICTED：直接放行，不消耗 AI
+    safety.set_permissions(allow=[], ask=[], deny=[])  # 隔离真实 config 的权限规则
+    try:
+        allowed, _ = safety.check("code_editor", "write a.py x")
+        assert allowed is True
+        allowed, _ = safety.check("cmd_executor", "rm -rf /")
+        assert allowed is True
+    finally:
+        safety.set_cli(saved_cli)
+        safety.set_mode(saved_mode)
+        safety._permissions_override = saved_perm
+
+
+# ── system prompt 注入 ──
+
+def test_build_system_prompt_injects_plan():
+    from xcli_core.cli_core import AICLI, PLAN_MODE_SYSTEM_APPEND
+    fake = types.SimpleNamespace(
+        plan_mode=True,
+        liugin_manager=types.SimpleNamespace(tools=[]),
+        memory_manager=None,
+    )
+    prompt = AICLI._build_system_prompt(fake)
+    assert "PLAN" in prompt
+    assert "实施计划" in prompt
+    assert PLAN_MODE_SYSTEM_APPEND in prompt
+
+
+def test_build_system_prompt_no_plan_when_off():
+    from xcli_core.cli_core import AICLI
+    fake = types.SimpleNamespace(
+        plan_mode=False,
+        liugin_manager=types.SimpleNamespace(tools=[]),
+        memory_manager=None,
+    )
+    prompt = AICLI._build_system_prompt(fake)
+    assert "实施计划" not in prompt
+
+
+# ── 命令 handler 状态转换 ──
+
+def _fake_for_plan():
+    calls = {"process": [], "autosave": 0, "sync": 0}
+    fake = types.SimpleNamespace(
+        plan_mode=False,
+        current_plan="",
+        process_conversation=lambda x: calls["process"].append(x),
+        _autosave_session=lambda: calls.__setitem__("autosave", calls["autosave"] + 1),
+        _sync_current_plan=lambda: calls.__setitem__("sync", calls["sync"] + 1),
+    )
+    return fake, calls
+
+
+def test_handle_plan_command_enters_mode_with_task():
+    from xcli_core.cli_core import AICLI
+    fake, calls = _fake_for_plan()
+    AICLI.handle_plan_command(fake, "重构缓存模块")
+    assert fake.plan_mode is True
+    assert calls["process"] == ["重构缓存模块"]
+    assert calls["autosave"] == 1
+    assert calls["sync"] == 1
+
+
+def test_handle_plan_command_enters_mode_without_task():
+    from xcli_core.cli_core import AICLI
+    fake, calls = _fake_for_plan()
+    AICLI.handle_plan_command(fake, "")
+    assert fake.plan_mode is True
+    assert calls["process"] == []  # 无任务不主动调研
+
+
+def test_handle_plan_command_off_exits():
+    from xcli_core.cli_core import AICLI
+    fake, calls = _fake_for_plan()
+    fake.plan_mode = True
+    AICLI.handle_plan_command(fake, "off")
+    assert fake.plan_mode is False
+    assert calls["process"] == []
+
+
+def test_handle_build_command_executes_plan():
+    from xcli_core.cli_core import AICLI
+    fake, calls = _fake_for_plan()
+    fake.plan_mode = True
+    fake.current_plan = "## 实施计划\n1. 改 a.py\n2. 改 b.py"
+    AICLI.handle_build_command(fake)
+    assert fake.plan_mode is False
+    assert len(calls["process"]) == 1
+    assert "已批准的实施计划" in calls["process"][0]
+    assert "## 实施计划" in calls["process"][0]
+
+
+def test_handle_build_command_no_plan_warns():
+    from xcli_core.cli_core import AICLI
+    fake, calls = _fake_for_plan()
+    fake.plan_mode = False
+    fake.current_plan = ""
+    AICLI.handle_build_command(fake)
+    assert calls["process"] == []  # 无计划不执行
+
+
+def test_sync_current_plan_captures_last_assistant():
+    from xcli_core.cli_core import AICLI
+    fake = types.SimpleNamespace(
+        plan_mode=True,
+        current_plan="",
+        shared_conversation_history=[
+            {"role": "user", "content": "任务"},
+            {"role": "assistant", "content": "调研中..."},
+            {"role": "tool", "content": "结果"},
+            {"role": "assistant", "content": "## 实施计划\n1. do x"},
+        ],
+    )
+    AICLI._sync_current_plan(fake)
+    assert "实施计划" in fake.current_plan
+
+
+def test_sync_current_plan_noop_when_not_plan():
+    from xcli_core.cli_core import AICLI
+    fake = types.SimpleNamespace(
+        plan_mode=False,
+        current_plan="unchanged",
+        shared_conversation_history=[{"role": "assistant", "content": "x"}],
+    )
+    AICLI._sync_current_plan(fake)
+    assert fake.current_plan == "unchanged"

tests/test_plugins.py

@@ -29,6 +29,29 @@ def load_module(name, path):
     return mod
 
 
+def plugin_class(mod):
+    """取插件入口类（项目里叫 Liugin，历史上叫 Plugin，两种都认）"""
+    cls = getattr(mod, "Liugin", None) or getattr(mod, "Plugin", None)
+    assert cls is not None, f"{mod.__name__}: 缺少 Plugin/Liugin 类"
+    return cls
+
+
+def as_text(result):
+    """插件返回值统一转文本（可能是 str，也可能是 ToolResult）"""
+    return result if isinstance(result, str) else str(result)
+
+
+ERROR_MARKERS = ("❌", "错误", "失败", "缺少", "不存在", "无效", "请提供", "用法")
+
+
+def is_error(result):
+    """判断插件返回是否表示失败（不依赖具体文案措辞）"""
+    success = getattr(result, "success", None)
+    if success is not None:
+        return not success
+    return any(m in as_text(result) for m in ERROR_MARKERS)
+
+
 # ══════════════════════════════════════
 #  插件加载测试
 # ══════════════════════════════════════
@@ -89,7 +112,7 @@ class TestCodeEditor:
     @pytest.fixture
     def editor(self):
         mod = load_module("code_editor", os.path.join(PLUGINS_DIR, "code_editor.py"))
-        return mod.Plugin()
+        return plugin_class(mod)()
 
     @pytest.fixture
     def tmp_file(self, tmp_path):
@@ -99,47 +122,47 @@ class TestCodeEditor:
 
     def test_read_range(self, editor, tmp_file):
         result = editor.handle(f"read_range {tmp_file} 1 2")
-        assert "hello" in result
+        assert "hello" in as_text(result)
-        assert "1 |" in result
+        assert "1 |" in as_text(result)
 
     def test_edit(self, editor, tmp_file):
         result = editor.handle(f"edit {tmp_file} print('hello') <<<>>> print('hi')")
-        assert "已编辑" in result
+        assert "已编辑" in as_text(result)
         content = open(tmp_file).read()
         assert "print('hi')" in content
 
     def test_edit_not_found(self, editor, tmp_file):
         result = editor.handle(f"edit {tmp_file} nonexistent_code <<<>>> new_code")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_create(self, editor, tmp_path):
         new_file = str(tmp_path / "new.py")
         result = editor.handle(f"create {new_file} import os")
-        assert "已创建" in result
+        assert "已创建" in as_text(result)
         assert os.path.exists(new_file)
 
     def test_diff_text(self, editor):
         result = editor.handle("diff_text old <<<>>> new")
-        assert "变更预览" in result
+        assert "变更预览" in as_text(result)
 
     def test_ast_info(self, editor, tmp_file):
         result = editor.handle(f"ast_info {tmp_file}")
-        assert "def hello" in result
+        assert "def hello" in as_text(result)
-        assert "def world" in result
+        assert "def world" in as_text(result)
 
     def test_stats(self, editor, tmp_path):
         result = editor.handle(f"stats {tmp_path}")
-        assert "代码统计" in result
+        assert "代码统计" in as_text(result)
-        assert "文件" in result
+        assert "文件" in as_text(result)
 
     def test_structure(self, editor, tmp_path):
         result = editor.handle(f"structure {tmp_path} 1")
-        assert "" in result
+        assert "" in as_text(result)
 
     def test_find(self, editor, tmp_path):
         (tmp_path / "a.py").write_text("x = 1\ny = 2\n")
         result = editor.handle(f"find {tmp_path} x *.py")
-        assert "x = 1" in result
+        assert "x = 1" in as_text(result)
 
 
 class TestGitTools:
@@ -148,11 +171,11 @@ class TestGitTools:
     @pytest.fixture
     def git(self):
         mod = load_module("git_tools", os.path.join(PLUGINS_DIR, "git_tools.py"))
-        return mod.Plugin()
+        return plugin_class(mod)()
 
     def test_status(self, git):
         result = git.handle("status")
-        assert "Git" in result or "干净" in result or "分支" in result
+        assert "Git" in as_text(result) or "干净" in as_text(result) or "分支" in as_text(result)
 
     def test_log(self, git):
         result = git.handle("log --oneline 3")
@@ -161,7 +184,7 @@ class TestGitTools:
 
     def test_branch(self, git):
         result = git.handle("branch")
-        assert "master" in result or "main" in result
+        assert "master" in as_text(result) or "main" in as_text(result)
 
 
 class TestCmdExecutor:
@@ -170,27 +193,27 @@ class TestCmdExecutor:
     @pytest.fixture
     def cmd(self):
         mod = load_module("cmd_executor", os.path.join(PLUGINS_DIR, "cmd_executor.py"))
-        return mod.Plugin()
+        return plugin_class(mod)()
 
     def test_echo(self, cmd):
         result = cmd.handle("run echo test_ok")
-        assert "test_ok" in result
+        assert "test_ok" in as_text(result)
 
     def test_python_version(self, cmd):
         result = cmd.handle("run python3 --version")
-        assert "Python" in result
+        assert "Python" in as_text(result)
 
     def test_blocked_command(self, cmd):
         result = cmd.handle("run rm -rf /")
-        assert "拦截" in result
+        assert "拦截" in as_text(result)
 
     def test_timeout(self, cmd):
         result = cmd.handle("run echo fast --timeout 5")
-        assert "fast" in result
+        assert "fast" in as_text(result)
 
     def test_empty_command(self, cmd):
         result = cmd.handle("run")
-        assert "错误" in result
+        assert is_error(result)
 
 
 class TestFileManager:
@@ -199,53 +222,53 @@ class TestFileManager:
     @pytest.fixture
     def fm(self):
         mod = load_module("file_manager", os.path.join(PLUGINS_DIR, "file_manager.py"))
-        return mod.Plugin()
+        return plugin_class(mod)()
 
     def test_list(self, fm, tmp_path):
         (tmp_path / "a.txt").write_text("hi")
         result = fm.handle(f"list {tmp_path}")
-        assert "a.txt" in result
+        assert "a.txt" in as_text(result)
 
     def test_read(self, fm, tmp_path):
         f = tmp_path / "test.txt"
         f.write_text("hello world")
         result = fm.handle(f"read {f}")
-        assert "hello world" in result
+        assert "hello world" in as_text(result)
 
     def test_write_and_read(self, fm, tmp_path):
         f = tmp_path / "new.txt"
         result = fm.handle(f"write {f} -c test_content")
-        assert "已写入" in result
+        assert "已写入" in as_text(result)
         result = fm.handle(f"read {f}")
-        assert "test_content" in result
+        assert "test_content" in as_text(result)
 
     def test_copy(self, fm, tmp_path):
         src = tmp_path / "src.txt"
         src.write_text("data")
         dst = tmp_path / "dst.txt"
         result = fm.handle(f"copy {src} {dst}")
-        assert "已复制" in result
+        assert "已复制" in as_text(result)
         assert dst.exists()
 
     def test_delete(self, fm, tmp_path):
         f = tmp_path / "del.txt"
         f.write_text("bye")
         result = fm.handle(f"delete {f}")
-        assert "已删除" in result
+        assert "已删除" in as_text(result)
         assert not f.exists()
 
     def test_mkdir(self, fm, tmp_path):
         d = tmp_path / "newdir"
         result = fm.handle(f"mkdir {d}")
-        assert "已创建" in result
+        assert "已创建" in as_text(result)
         assert d.is_dir()
 
     def test_info(self, fm, tmp_path):
         f = tmp_path / "info.txt"
         f.write_text("x")
         result = fm.handle(f"info {f}")
-        assert "路径" in result
+        assert "路径" in as_text(result)
-        assert "大小" in result
+        assert "大小" in as_text(result)
 
 
 class TestAutoEngineer:
@@ -254,23 +277,23 @@ class TestAutoEngineer:
     @pytest.fixture
     def eng(self):
         mod = load_module("auto_engineer", os.path.join(PLUGINS_DIR, "auto_engineer.py"))
-        return mod.Plugin()
+        return plugin_class(mod)()
 
     def test_metrics(self, eng, tmp_path):
         (tmp_path / "a.py").write_text("x = 1\n# comment\n\ny = 2\n")
         result = eng.handle(f"metrics {tmp_path}")
-        assert "代码指标" in result
+        assert "代码指标" in as_text(result)
-        assert "文件" in result
+        assert "文件" in as_text(result)
 
     def test_security_clean(self, eng, tmp_path):
         (tmp_path / "safe.py").write_text("x = 1\nprint(x)\n")
         result = eng.handle(f"security {tmp_path}")
-        assert "没有发现" in result or "风险" in result
+        assert "没有发现" in as_text(result) or "风险" in as_text(result)
 
     def test_security_risk(self, eng, tmp_path):
         (tmp_path / "risk.py").write_text('password = "secret123"\neval("1+1")\n')
         result = eng.handle(f"security {tmp_path}")
-        assert "风险" in result
+        assert "风险" in as_text(result)
 
     def test_complexity(self, eng, tmp_path):
         code = "def f():\n    if True:\n        if True:\n            if True:\n                pass\n"
@@ -292,34 +315,34 @@ class TestTaskManager:
 
     def test_add(self, tm):
         result = tm.handle("add 测试任务")
-        assert "已添加" in result
+        assert "已添加" in as_text(result)
 
     def test_list_empty(self, tm):
         result = tm.handle("list")
-        assert "没有任务" in result
+        assert "没有任务" in as_text(result)
 
     def test_add_and_list(self, tm):
         tm.handle("add 任务A")
         tm.handle("add 任务B")
         result = tm.handle("list")
-        assert "任务A" in result
+        assert "任务A" in as_text(result)
-        assert "任务B" in result
+        assert "任务B" in as_text(result)
 
     def test_complete(self, tm):
         tm.handle("add 任务X")
         result = tm.handle("complete 1")
-        assert "已更新" in result
+        assert "已更新" in as_text(result)
 
     def test_delete(self, tm):
         tm.handle("add 任务Y")
         result = tm.handle("delete 1")
-        assert "已删除" in result
+        assert "已删除" in as_text(result)
 
     def test_clear(self, tm):
         tm.handle("add A")
         tm.handle("complete 1")
         result = tm.handle("clear")
-        assert "已清除" in result
+        assert "已清除" in as_text(result)
 
 
 # ══════════════════════════════════════
@@ -332,19 +355,19 @@ class TestBrowserAuto:
     @pytest.fixture
     def browser(self):
         mod = load_module("browser_auto", os.path.join(PLUGINS_DIR, "browser_auto.py"))
-        return mod.Plugin()
+        return plugin_class(mod)()
 
     def test_status_stopped(self, browser):
         result = browser.handle("status")
-        assert "未运行" in result
+        assert "未运行" in as_text(result)
 
     def test_invalid_operation(self, browser):
         result = browser.handle("nonexistent")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_empty_args(self, browser):
         result = browser.handle("")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_tool_info(self, browser):
         info = browser.get_tool_info()
@@ -363,7 +386,7 @@ class TestGuiAuto:
     @pytest.fixture
     def gui(self):
         mod = load_module("gui_auto", os.path.join(PLUGINS_DIR, "gui_auto.py"))
-        return mod.Plugin()
+        return plugin_class(mod)()
 
     def test_tool_info(self, gui):
         info = gui.get_tool_info()
@@ -380,88 +403,88 @@ class TestGuiAuto:
     def test_convert_mcp_args(self, gui):
         args = {"operation": "click", "target": "确定"}
         result = gui.convert_mcp_args(args)
-        assert "click" in result
+        assert "click" in as_text(result)
-        assert "确定" in result
+        assert "确定" in as_text(result)
 
     def test_convert_mcp_args_xy(self, gui):
         args = {"operation": "click", "x": 100, "y": 200}
         result = gui.convert_mcp_args(args)
-        assert "--xy" in result
+        assert "--xy" in as_text(result)
-        assert "100" in result
+        assert "100" in as_text(result)
 
     def test_convert_mcp_args_snapshot(self, gui):
         args = {"operation": "snapshot", "depth": 3}
         result = gui.convert_mcp_args(args)
-        assert "snapshot" in result
+        assert "snapshot" in as_text(result)
-        assert "3" in result
+        assert "3" in as_text(result)
 
     def test_empty_args(self, gui):
         result = gui.handle("")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_invalid_operation(self, gui):
         result = gui.handle("nonexistent_op")
-        assert "不支持" in result
+        assert "不支持" in as_text(result)
 
     def test_click_no_args(self, gui):
         result = gui.handle("click")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_type_no_args(self, gui):
         result = gui.handle("type")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_type_missing_text(self, gui):
         result = gui.handle("type 按钮")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_find_no_args(self, gui):
         result = gui.handle("find")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_findall_no_args(self, gui):
         result = gui.handle("findall")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_info_no_args(self, gui):
         result = gui.handle("info")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_value_no_args(self, gui):
         result = gui.handle("value")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_tree_no_args(self, gui):
         result = gui.handle("tree")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_wait_no_args(self, gui):
         result = gui.handle("wait")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_exists_no_args(self, gui):
         result = gui.handle("exists")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_highlight_no_args(self, gui):
         result = gui.handle("highlight")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_state_no_args(self, gui):
         result = gui.handle("state")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_keys_no_args(self, gui):
         result = gui.handle("keys")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_focus_no_args(self, gui):
         result = gui.handle("focus")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_click_xy_format_error(self, gui):
         result = gui.handle("click --xy abc")
-        assert "错误" in result
+        assert is_error(result)
 
     def test_format_element_info(self, gui):
         el = {
@@ -472,9 +495,9 @@ class TestGuiAuto:
             "enabled": True, "focused": False
         }
         result = gui._format_element_info(el)
-        assert "测试按钮" in result
+        assert "测试按钮" in as_text(result)
-        assert "Button" in result
+        assert "Button" in as_text(result)
-        assert "btn_ok" in result
+        assert "btn_ok" in as_text(result)
 
     def test_format_element_info_detailed(self, gui):
         el = {
@@ -485,8 +508,8 @@ class TestGuiAuto:
                        "right": 200, "bottom": 25}
         }
         result = gui._format_element_info(el, detailed=True)
-        assert "输入框" in result
+        assert "输入框" in as_text(result)
-        assert "hello" in result
+        assert "hello" in as_text(result)
 
     def test_all_operations_route(self, gui):
         """所有操作都能正确路由"""
@@ -504,28 +527,28 @@ class TestSubAgent:
     @pytest.fixture
     def sa(self):
         mod = load_module("sub_agent", os.path.join(PLUGINS_DIR, "sub_agent.py"))
-        return mod.Plugin()
+        return plugin_class(mod)()
 
     def test_list_empty(self, sa):
         result = sa.handle("list")
-        assert "没有" in result
+        assert "没有" in as_text(result)
 
     def test_stats_empty(self, sa):
         result = sa.handle("stats")
-        assert "总数: 0" in result
+        assert "总数: 0" in as_text(result)
 
     def test_spawn(self, sa):
         result = sa.handle("spawn tester 测试任务")
-        assert "已创建" in result
+        assert "已创建" in as_text(result)
-        assert "agent_tester" in result
+        assert "agent_tester" in as_text(result)
 
     def test_spawn_and_list(self, sa):
         sa.handle("spawn a 任务A")
         sa.handle("spawn b 任务B")
         import time; time.sleep(0.1)
         result = sa.handle("list")
-        assert "agent_a" in result
+        assert "agent_a" in as_text(result)
-        assert "agent_b" in result
+        assert "agent_b" in as_text(result)
 
     def test_status(self, sa):
         sa.handle("spawn x 某任务")
@@ -533,24 +556,24 @@ class TestSubAgent:
         # 获取 agent ID
         agent_id = list(sa.agents.keys())[0]
         result = sa.handle(f"status {agent_id}")
-        assert "名称" in result
+        assert "名称" in as_text(result)
 
     def test_kill(self, sa):
         sa.handle("spawn y 要终止的任务")
         import time; time.sleep(0.1)
         agent_id = list(sa.agents.keys())[0]
         result = sa.handle(f"kill {agent_id}")
-        assert "已终止" in result
+        assert "已终止" in as_text(result)
 
     def test_clean(self, sa):
         sa.handle("spawn z 清理测试")
         import time; time.sleep(0.1)
         result = sa.handle("clean")
-        assert "已清理" in result
+        assert "已清理" in as_text(result)
 
     def test_invalid_agent(self, sa):
         result = sa.handle("status nonexistent")
-        assert "未找到" in result
+        assert "未找到" in as_text(result)
 
 
 # ══════════════════════════════════════

tests/test_provider_router.py

@@ -0,0 +1,135 @@
+"""Provider 抽象层测试：预设、resolve、消息/工具转换、汇总。不触网、不读真实 config。"""
+import os
+import sys
+import types
+
+import pytest
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from xcli_core.provider_router import (
+    PROVIDER_PRESETS, list_providers, resolve_provider,
+    convert_messages, convert_tools, provider_summary,
+)
+
+
+def test_presets_exist():
+    for name in ["openai", "anthropic", "google", "deepseek", "xunfei", "ollama"]:
+        assert name in PROVIDER_PRESETS
+    assert "openai" in list_providers()
+
+
+def test_resolve_default_no_key():
+    r = resolve_provider("openai")
+    assert r["name"] == "openai"
+    assert r["base_url"] == "https://api.openai.com/v1"
+    assert r["protocol"] == "openai"
+    # 无环境变量无配置 → api_key 空
+    assert r["api_key"] == ""
+
+
+def test_resolve_from_env(monkeypatch):
+    monkeypatch.setenv("OPENAI_API_KEY", "sk-test-123")
+    r = resolve_provider("openai")
+    assert r["api_key"] == "sk-test-123"
+
+
+def test_resolve_from_cfg_overrides_env(monkeypatch):
+    monkeypatch.setenv("OPENAI_API_KEY", "sk-env")
+    r = resolve_provider("openai", {"openai": {"api_key": "sk-cfg", "model": "gpt-4o-mini"}})
+    assert r["api_key"] == "sk-cfg"
+    assert r["model"] == "gpt-4o-mini"
+
+
+def test_resolve_anthropic_headers():
+    r = resolve_provider("anthropic", {"anthropic": {"api_key": "ak"}})
+    assert r["protocol"] == "anthropic"
+    assert r["headers"].get("anthropic-version") == "2023-06-01"
+
+
+def test_resolve_google_query_header():
+    r = resolve_provider("google", {"google": {"api_key": "gk"}})
+    assert r["headers"].get("x-goog-api-key") == "gk"
+
+
+def test_resolve_unknown():
+    assert resolve_provider("not-a-provider") == {}
+
+
+# ── 消息转换 ──
+OPENAI_MSGS = [
+    {"role": "system", "content": "你是助手"},
+    {"role": "user", "content": "你好"},
+    {"role": "assistant", "content": "在的", "tool_calls": [{"id": "c1"}]},
+    {"role": "tool", "content": "结果", "name": "calc"},
+]
+
+
+def test_convert_openai_keeps():
+    out = convert_messages("openai", OPENAI_MSGS)
+    assert out[0]["role"] == "system"
+    assert out[1]["content"] == "你好"
+    assert out[2]["tool_calls"] == [{"id": "c1"}]
+
+
+def test_convert_anthropic_strips_system_blocks():
+    out = convert_messages("anthropic", OPENAI_MSGS)
+    roles = [m["role"] for m in out]
+    assert "system" not in roles
+    assert out[0]["content"][0]["type"] == "text"
+    assert out[-1]["role"] == "user"  # tool 角色映射到 user
+
+
+def test_convert_google_uses_parts():
+    out = convert_messages("google", OPENAI_MSGS)
+    roles = [m["role"] for m in out]
+    assert "system" not in roles
+    assert out[0]["parts"][0]["text"] == "你好"
+    assert out[-1]["role"] == "user"
+
+
+def test_convert_multimodal_content_blocks():
+    msgs = [{"role": "user", "content": [
+        {"type": "text", "text": "看图"},
+        {"type": "image", "image_url": {"url": "x"}},
+    ]}]
+    oai = convert_messages("openai", msgs)
+    assert oai[0]["content"] == "看图"  # 多模态 → 文本拼接
+    ant = convert_messages("anthropic", msgs)
+    assert ant[0]["content"][0]["type"] == "text"
+    assert ant[0]["content"][1]["type"] == "image"
+
+
+# ── 工具转换 ──
+TOOLS = [{"type": "function", "function": {
+    "name": "search", "description": "搜索", "parameters": {"type": "object", "properties": {}}}}]
+
+
+def test_convert_tools_openai_passthrough():
+    assert convert_tools("openai", TOOLS) == TOOLS
+
+
+def test_convert_tools_anthropic_schema():
+    out = convert_tools("anthropic", TOOLS)
+    assert out[0]["name"] == "search"
+    assert out[0]["input_schema"] == {"type": "object", "properties": {}}
+
+
+def test_convert_tools_google_declarations():
+    out = convert_tools("google", TOOLS)
+    assert "functionDeclarations" in out[0]
+    assert out[0]["functionDeclarations"][0]["name"] == "search"
+
+
+# ── 汇总 ──
+def test_provider_summary_unconfigured():
+    s = provider_summary("anthropic")
+    assert s["name"] == "anthropic"
+    assert s["configured"] is False
+    assert s["protocol"] == "anthropic"
+
+
+def test_provider_summary_configured(monkeypatch):
+    monkeypatch.setenv("ANTHROPIC_API_BASE", "ignored")
+    s = provider_summary("anthropic", {"anthropic": {"api_key": "x"}})
+    assert s["configured"] is True

tests/test_prune_tools.py

@@ -0,0 +1,115 @@
+"""
+测试：FC 工具去重 + 按场景动态裁剪（方案 A+B）。
+
+不依赖真实插件/引擎加载，用构造的 fake 数据验证：
+- mcp_to_openai_tools 按 name 去重（消除 cache + builtin 叠加重复）
+- prune_tools 分层裁剪：纯聊天极简、代码信号追加 lsp__*、外部 mcp 默认保留
+"""
+import pytest
+
+from xcli_core.fc_tools import mcp_to_openai_tools, prune_tools
+
+
+def _fc(name, desc="d"):
+    return {
+        "type": "function",
+        "function": {
+            "name": name,
+            "description": desc,
+            "parameters": {"type": "object", "properties": {}, "required": []},
+        },
+    }
+
+
+def _mcp(name, desc="d"):
+    """MCP 格式（get_mcp_tools 返回的就是这种）"""
+    return {
+        "name": name,
+        "description": desc,
+        "inputSchema": {"type": "object", "properties": {}, "required": []},
+    }
+
+
+class _FakePM:
+    """最小 plugin_manager：get_mcp_tools 返回预置列表。"""
+
+    def __init__(self, tools):
+        self._tools = tools
+
+    def get_mcp_tools(self):
+        return self._tools
+
+
+# ───────────── 去重 ─────────────
+
+def test_mcp_to_openai_tools_dedup():
+    # 模拟真实场景：cache 里已有 memory/scheduler，_get_builtin_tools 又追加一遍
+    pm = _FakePM([
+        _mcp("a"), _mcp("a"),                     # 同名重复
+        _mcp("b"),
+        _mcp("memory"), _mcp("scheduler"),        # cache 已有
+    ])
+    out = mcp_to_openai_tools(pm)
+    names = [t["function"]["name"] for t in out]
+    assert names.count("a") == 1, "同名工具应去重为 1"
+    assert names.count("memory") == 1, "builtin 与 cache 重复应去重"
+    assert names.count("scheduler") == 1
+    assert "b" in names
+    # builtin 追加的 memory/scheduler 不应再产生第二份
+    assert names.count("memory") == 1
+
+
+# ───────────── 分层裁剪 ─────────────
+
+_CORE = ["tool_search", "memory", "scheduler", "code_editor", "code_search",
+         "file_manager", "lsp__diagnostics", "lsp__hover", "lsp__definition",
+         "lsp__references", "lsp__completion", "browser_auto", "network_tools",
+         "ai_search", "git_tools", "cmd_executor", "sub_agent", "auto_engineer"]
+
+
+def _full_set():
+    return [_fc(n) for n in _CORE]
+
+
+def test_prune_pure_chat_keeps_only_always():
+    out = prune_tools(_full_set(), "你好啊，今天天气真不错")
+    names = {t["function"]["name"] for t in out}
+    assert names == {"tool_search", "memory", "scheduler"}, \
+        f"纯聊天应只留常驻工具，实际: {names}"
+
+
+def test_prune_code_signal_adds_lsp():
+    out = prune_tools(_full_set(), "帮我写一个 python 脚本读取文件")
+    names = {t["function"]["name"] for t in out}
+    assert "code_editor" in names
+    assert "file_manager" in names
+    assert "lsp__diagnostics" in names, "写代码应追加 pylsp 的 lsp__* 工具"
+    assert "lsp__completion" in names
+    assert "tool_search" in names
+
+
+def test_prune_git_signal():
+    out = prune_tools(_full_set(), "提交代码并查看 git 状态")
+    names = {t["function"]["name"] for t in out}
+    assert "git_tools" in names
+    assert "tool_search" in names
+
+
+def test_prune_external_mcp_kept_even_in_chat():
+    tools = [_fc("my_mcp__search"), _fc("code_editor"), _fc("tool_search")]
+    out = prune_tools(tools, "你好")
+    names = {t["function"]["name"] for t in out}
+    assert "my_mcp__search" in names, "外部桥接工具(mcp__)应默认保留"
+    assert "code_editor" not in names, "未命中信号的低频工具应被裁剪"
+    assert "tool_search" in names
+
+
+def test_prune_empty_input_no_crash():
+    out = prune_tools(_full_set(), "")
+    names = {t["function"]["name"] for t in out}
+    assert names == {"tool_search", "memory", "scheduler"}
+
+
+def test_prune_none_tools_returns_none():
+    assert prune_tools(None, "任意输入") is None
+    assert prune_tools([], "任意输入") == []

tests/test_pure_json.py

@@ -0,0 +1,145 @@
+"""
+纯 JSON 工具调用模式 测试
+
+取代 FC：引擎不再收到 tools schema，模型按系统提示里的
+{"action":"use_tool","tool":名,"args":参数} 文本输出，由 _parse_mixed_response
+解析、按名分发。重点验证：
+1. _loads_json / _parse_mixed_response 对 Windows 反斜杠路径健壮
+2. 引擎在 tools=None 时返回纯文本（不产 _fc / tool_calls）
+3. 端到端：真实 AICLI + FakeEngine，use_tool 被按名分发执行，且无 FC schema 下发
+"""
+import os
+import sys
+import json
+import tempfile
+
+import pytest
+
+PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+if PROJECT_ROOT not in sys.path:
+    sys.path.insert(0, PROJECT_ROOT)
+
+
+# ════════════════════════════════════════════════
+#  1. 解析器：Windows 反斜杠路径容错
+# ════════════════════════════════════════════════
+
+@pytest.fixture
+def cli():
+    from xcli_core.cli_core import AICLI
+    return AICLI()
+
+
+class TestParseRobustness:
+    def test_loads_json_tolerates_backslash_path(self, cli):
+        # 模型常把 Windows 路径写成单反斜杠 → 非法 JSON 转义
+        raw = r'{"action": "use_tool", "tool": "code_editor", "args": "create C:\Users\me\demo.py print(1)"}'
+        data = cli._loads_json(raw)
+        assert data["action"] == "use_tool"
+        assert data["tool"] == "code_editor"
+        # 修复后应是合法的单反斜杠路径
+        assert data["args"] == r"create C:\Users\me\demo.py print(1)"
+
+    def test_parse_mixed_extracts_use_tool_with_backslash(self, cli):
+        resp = (
+            '好的，我来创建文件。\n'
+            r'{"action": "use_tool", "tool": "code_editor", "args": "create C:\Users\me\demo.py print(1)"}'
+        )
+        text, data = cli._parse_mixed_response(resp)
+        assert isinstance(data, dict)
+        assert data["action"] == "use_tool"
+        assert data["tool"] == "code_editor"
+        assert r"C:\Users\me\demo.py" in data["args"]
+
+    def test_parse_mixed_plain_text_no_action(self, cli):
+        text, data = cli._parse_mixed_response("今天天气不错")
+        assert data is None
+        assert "今天天气不错" in text
+
+
+# ════════════════════════════════════════════════
+#  2. 引擎：tools=None 返回纯文本，不产 tool_calls
+# ════════════════════════════════════════════════
+
+class TestEngineToolsNone:
+    def test_openai_engine_no_tools_sent(self, monkeypatch):
+        import types
+        import ai_engines.openai_engine as oe
+
+        calls = []
+
+        def fake_post(url=None, json=None, headers=None, timeout=None, **kw):
+            calls.append(dict(json))
+            return types.SimpleNamespace(
+                status_code=200,
+                json=lambda: {"choices": [{"message": {"content": "好"}}],
+                               "usage": {"prompt_tokens": 5}},
+                text="{}",
+            )
+
+        monkeypatch.setattr(oe, "requests",
+                             types.SimpleNamespace(post=fake_post, exceptions=oe.requests.exceptions))
+
+        eng = oe.OpenaiAI.__new__(oe.OpenaiAI)
+        eng.name = "openai"
+        eng.api_key = "k"
+        eng.base_url = "https://example.com/v1"
+        eng.model = "gpt-4o"
+        eng.max_history = 10
+        eng.conversation_history = []
+        eng.shared_conversation_history = None
+        eng.last_prompt_tokens = None
+        eng.thinking_start_marker = "<think>"
+        eng.thinking_end_marker = "</think>"
+
+        reply = eng.generate_response("hi", tools=None)
+        assert reply == "好"
+        assert "tools" not in calls[0], "tools=None 时不应下发 FC schema"
+        assert "tool_choice" not in calls[0]
+
+
+# ════════════════════════════════════════════════
+#  3. 端到端：真实 AICLI + FakeEngine，use_tool 按名分发
+# ════════════════════════════════════════════════
+
+class FakeEngine:
+    def __init__(self):
+        self.calls = 0
+        self.tools_seen = []
+
+    @property
+    def name(self):
+        return "fake"
+
+    def generate_response(self, user_input, tool_results=None, system_prompt=None, tools=None):
+        self.calls += 1
+        self.tools_seen.append(tools)
+        assert tools is None, f"纯 JSON 模式不应下发 FC tools，实际 {tools!r}"
+        if self.calls == 1:
+            return ('好的，我来创建文件。\n'
+                    '{"action": "use_tool", "tool": "code_editor", '
+                    '"args": "create %s print(1)"}' % self._fp)
+        return "已完成。"
+
+
+class TestEndToEndDispatch:
+    def test_use_tool_dispatched_without_fc(self):
+        from xcli_core.cli_core import AICLI
+        from xcli_core.safety import get_safety, MODE_UNRESTRICTED
+
+        cli = AICLI()
+        engine = FakeEngine()
+        cli.current_engine = engine
+        cli.tui_output_callback = lambda *a, **k: None   # 同步 TUI 分支，免动画线程
+        get_safety().set_mode(MODE_UNRESTRICTED)          # 跳过交互确认，验证分发本身
+
+        d = tempfile.mkdtemp(prefix="pj_")
+        FakeEngine._fp = os.path.join(d, "demo.py")       # 原始反斜杠路径
+        fp = FakeEngine._fp
+
+        cli.process_conversation("帮我创建一个 python 文件")
+
+        assert all(t is None for t in engine.tools_seen), "引擎不应收到任何 FC tools"
+        assert engine.calls >= 2, "至少应有：首轮JSON + 工具结果续轮"
+        assert os.path.exists(fp), "code_editor 应被按名分发执行并创建文件"
+        assert "print(1)" in open(fp, encoding="utf-8").read()

tests/test_py_detect.py

@@ -0,0 +1,647 @@
+"""py_detect 本地智能检测插件测试"""
+import importlib.util
+import os
+import sys
+
+import pytest
+
+# 项目根目录
+PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+PLUGIN_PATH = os.path.join(PROJECT_ROOT, "plugins", "py_detect.py")
+
+
+def load_module(name, path):
+    spec = importlib.util.spec_from_file_location(name, path)
+    mod = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(mod)
+    return mod
+
+
+@pytest.fixture(scope="module")
+def pd():
+    mod = load_module("py_detect", PLUGIN_PATH)
+    return mod
+
+
+def codes(issues):
+    """提取诊断规则码集合"""
+    return {d["code"] for d in issues}
+
+
+def code_lines(issues, code):
+    """按规则码提取行号集合"""
+    return {d["line"] for d in issues if d["code"] == code}
+
+
+# ── 语法层 ──
+class TestSyntax:
+    def test_syntax_error(self, pd):
+        issues = pd.analyze("def f(:\n    pass\n", "bad.py")
+        assert "E001" in codes(issues)
+        e = [d for d in issues if d["code"] == "E001"][0]
+        assert e["line"] >= 1
+
+    def test_unclosed_paren(self, pd):
+        issues = pd.analyze("x = (1 + 2\n", "bad.py")
+        assert "E001" in codes(issues)
+
+
+# ── 语义层 ──
+class TestSemantic:
+    def test_undefined_variable(self, pd):
+        src = "def f():\n    return y\n"
+        assert 2 in code_lines(pd.analyze(src, "t.py"), "E002")
+
+    def test_no_undefined_for_local_and_global(self, pd):
+        src = "x = 1\ndef f(a):\n    b = a + x\n    return b\nprint(f(1))\n"
+        assert "E002" not in codes(pd.analyze(src, "t.py"))
+
+    def test_undefined_recursion_ok(self, pd):
+        src = "def f(n):\n    if n <= 1:\n        return 1\n    return n * f(n - 1)\n"
+        assert "E002" not in codes(pd.analyze(src, "t.py"))
+
+    def test_comprehension_and_walrus_ok(self, pd):
+        src = "nums = [1, 2, 3]\nsq = [x * x for x in nums]\nif (n := len(nums)) > 2:\n    print(n)\n"
+        assert "E002" not in codes(pd.analyze(src, "t.py"))
+
+    def test_closure_reads_global_ok(self, pd):
+        src = "counter = 0\ndef inc():\n    return counter + 1\n"
+        assert "E002" not in codes(pd.analyze(src, "t.py"))
+
+    def test_dup_function(self, pd):
+        src = "def f():\n    pass\ndef f():\n    pass\n"
+        assert 3 in code_lines(pd.analyze(src, "t.py"), "E003")
+
+    def test_unused_import(self, pd):
+        src = "import os\nimport json\nprint(json.dumps(1))\n"
+        assert 1 in code_lines(pd.analyze(src, "t.py"), "W001")
+
+    def test_import_used_inside_function(self, pd):
+        src = "import os\ndef f():\n    return os.path.join('a', 'b')\n"
+        assert "W001" not in codes(pd.analyze(src, "t.py"))
+
+
+# ── 反模式层 ──
+class TestAntiPatterns:
+    def test_mutable_default(self, pd):
+        src = "def f(items=[]):\n    pass\n"
+        assert 1 in code_lines(pd.analyze(src, "t.py"), "W004")
+
+    def test_bare_except(self, pd):
+        src = "try:\n    x = 1\nexcept:\n    pass\n"
+        assert 3 in code_lines(pd.analyze(src, "t.py"), "W002")
+
+    def test_wide_except(self, pd):
+        src = "try:\n    x = 1\nexcept Exception:\n    pass\n"
+        assert 3 in code_lines(pd.analyze(src, "t.py"), "W003")
+
+    def test_eq_none(self, pd):
+        src = "if x == None:\n    pass\n"
+        assert 1 in code_lines(pd.analyze(src, "t.py"), "W005")
+
+    def test_mutate_during_iter(self, pd):
+        src = "for x in items:\n    items.remove(x)\n"
+        assert 2 in code_lines(pd.analyze(src, "t.py"), "W006")
+
+    def test_method_missing_self(self, pd):
+        src = "class A:\n    def f():\n        pass\n"
+        assert 2 in code_lines(pd.analyze(src, "t.py"), "W007")
+
+    def test_staticmethod_ok(self, pd):
+        src = "class A:\n    @staticmethod\n    def f():\n        pass\n"
+        assert "W007" not in codes(pd.analyze(src, "t.py"))
+
+    def test_builtin_shadow(self, pd):
+        src = "list = [1, 2]\n"
+        assert 1 in code_lines(pd.analyze(src, "t.py"), "W008")
+
+    def test_dead_code(self, pd):
+        src = "def f():\n    return 1\n    x = 2\n"
+        assert 3 in code_lines(pd.analyze(src, "t.py"), "W009")
+
+    def test_import_star(self, pd):
+        src = "from math import *\n"
+        assert 1 in code_lines(pd.analyze(src, "t.py"), "W010")
+
+    def test_div_zero(self, pd):
+        src = "x = 1 / 0\n"
+        assert 1 in code_lines(pd.analyze(src, "t.py"), "W011")
+
+    def test_type_compare(self, pd):
+        src = "if type(x) == int:\n    pass\n"
+        assert 1 in code_lines(pd.analyze(src, "t.py"), "I001")
+
+    def test_clean_code_no_issues(self, pd):
+        src = (
+            "import os\n"
+            "def add(a, b):\n"
+            "    return a + b\n"
+            "def main():\n"
+            "    path = os.path.join('x', 'y')\n"
+            "    items = [1, 2, 3]\n"
+            "    total = sum(items)\n"
+            "    return path, add(total, 1)\n"
+            "if __name__ == '__main__':\n"
+            "    main()\n"
+        )
+        assert pd.analyze(src, "clean.py") == []
+
+
+# ── 插件接口 / handle ──
+class TestHandle:
+    def test_version(self, pd):
+        plugin = pd.Liugin()
+        out = plugin.handle("version")
+        assert "py_detect" in out and "v1" in out
+
+    def test_rules_list(self, pd):
+        plugin = pd.Liugin()
+        out = plugin.handle("rules")
+        assert "E002" in out and "W004" in out and "I001" in out
+
+    def test_rules_single(self, pd):
+        plugin = pd.Liugin()
+        out = plugin.handle("rules W004")
+        assert "可变默认参数" in out
+
+    def test_rules_unknown(self, pd):
+        plugin = pd.Liugin()
+        out = plugin.handle("rules ZZZ")
+        assert "没有规则" in out
+
+    def test_check_missing_file(self, pd):
+        plugin = pd.Liugin()
+        out = plugin.handle("check /no/such/file.py")
+        assert "文件不存在" in out
+
+    def test_check_dir(self, pd, tmp_path):
+        plugin = pd.Liugin()
+        out = plugin.handle(f"check {tmp_path}")
+        assert "是目录" in out
+
+    def test_check_clean_file(self, pd, tmp_path):
+        plugin = pd.Liugin()
+        f = tmp_path / "ok.py"
+        f.write_text("x = 1\nprint(x)\n", encoding="utf-8")
+        out = plugin.handle(f"check {f}")
+        assert "未发现可疑问题" in out
+
+    def test_check_with_issues(self, pd, tmp_path):
+        plugin = pd.Liugin()
+        f = tmp_path / "bad.py"
+        f.write_text("def f(items=[]):\n    return items\n", encoding="utf-8")
+        out = plugin.handle(f"check {f}")
+        assert "[W004]" in out and "可变默认参数" in out
+
+    def test_check_scope_filter(self, pd, tmp_path):
+        plugin = pd.Liugin()
+        f = tmp_path / "scoped.py"
+        f.write_text("y = None\nif y == None:\n    pass\nz = 2\n", encoding="utf-8")
+        out = plugin.handle(f"check {f} 2")
+        assert "[W005]" in out and "未定义变量" not in out
+
+    def test_check_bad_scope(self, pd, tmp_path):
+        plugin = pd.Liugin()
+        f = tmp_path / "any.py"
+        f.write_text("x = 1\n", encoding="utf-8")
+        out = plugin.handle(f"check {f} abc")
+        assert "范围格式错误" in out
+
+    def test_tool_info(self, pd):
+        plugin = pd.Liugin()
+        info = plugin.get_tool_info()
+        assert info["name"] == "py_detect"
+        assert "智能检测" in info["description"]
+        assert "check" in info["usage"]
+
+    def test_mcp_definition(self, pd):
+        plugin = pd.Liugin()
+        mcp = plugin.get_mcp_definition()
+        assert mcp["name"] == "py_detect"
+        assert "check" in mcp["inputSchema"]["properties"]["operation"]["enum"]
+
+
+# ── 算法智能解释 (零 AI 模型, 纯算法动态生成) ──
+class TestSmartExplain:
+    def fix_of(self, issues, code):
+        for d in issues:
+            if d["code"] == code:
+                return d.get("fix", "")
+        return ""
+
+    def test_spell_suggestion(self, pd):
+        # 拼写纠错: 未定义变量 totl → 建议 total
+        src = "total = 0\ndef f():\n    return totl\n"
+        issues = pd.analyze(src, "t.py")
+        fix = self.fix_of(issues, "E002")
+        assert "total" in fix, fix
+        assert "totl" in fix, fix
+
+    def test_mutable_default_dynamic_fix(self, pd):
+        src = "def f(items=[]):\n    pass\n"
+        fix = self.fix_of(pd.analyze(src, "t.py"), "W004")
+        assert "def f(items=None)" in fix, fix
+
+    def test_eq_none_dynamic_fix(self, pd):
+        src = "if x == None:\n    pass\n"
+        fix = self.fix_of(pd.analyze(src, "t.py"), "W005")
+        assert "x is None" in fix, fix
+
+    def test_ne_none_dynamic_fix(self, pd):
+        src = "if x != None:\n    pass\n"
+        fix = self.fix_of(pd.analyze(src, "t.py"), "W005")
+        assert "x is not None" in fix, fix
+
+    def test_mutate_iter_dynamic_fix(self, pd):
+        src = "for x in items:\n    items.remove(x)\n"
+        fix = self.fix_of(pd.analyze(src, "t.py"), "W006")
+        assert "items[:]" in fix, fix
+
+    def test_dup_def_dynamic_line(self, pd):
+        src = "def f():\n    pass\ndef f():\n    pass\n"
+        fix = self.fix_of(pd.analyze(src, "t.py"), "E003")
+        assert "第 1 行" in fix, fix
+
+    def test_dead_code_dynamic_line(self, pd):
+        src = "def f():\n    return 1\n    x = 2\n"
+        fix = self.fix_of(pd.analyze(src, "t.py"), "W009")
+        assert "第 2 行 return" in fix, fix
+
+    def test_method_self_dynamic(self, pd):
+        src = "class A:\n    def f():\n        pass\n"
+        fix = self.fix_of(pd.analyze(src, "t.py"), "W007")
+        assert "self" in fix, fix
+
+    def test_type_compare_dynamic(self, pd):
+        src = "if type(x) == int:\n    pass\n"
+        fix = self.fix_of(pd.analyze(src, "t.py"), "I001")
+        assert "isinstance(x, int)" in fix, fix
+
+
+# ── L2 jedi 语义验证后端 (可选, 无 jedi 自动降级) ──
+class TestJediBackend:
+    def test_import_star_false_positive_removed(self, pd):
+        pytest.importorskip("jedi")
+        # 规则引擎: import * 不收集 sin → 误报 E002; jedi: 能解析 math.sin → 消除
+        src = "from math import *\ndef f():\n    return sin(3.14)\n"
+        codes = {d["code"] for d in pd.analyze(src, "C:/tmp/x.py")}
+        assert "E002" not in codes
+        assert "W010" in codes  # import * 警告保留
+
+    def test_real_undefined_kept_with_jedi(self, pd):
+        pytest.importorskip("jedi")
+        src = "def f():\n    return totally_missing_xyz\n"
+        assert "E002" in {d["code"] for d in pd.analyze(src, "C:/tmp/x.py")}
+
+    def test_fallback_without_jedi(self, pd, monkeypatch):
+        import sys
+        monkeypatch.setitem(sys.modules, "jedi", None)  # import jedi 将失败
+        src = "from math import *\ndef f():\n    return sin(3.14)\n"
+        codes = {d["code"] for d in pd.analyze(src, "C:/tmp/x.py")}
+        assert "E002" in codes  # 无 jedi → 保留规则引擎判定 (行为不变)
+        assert "W010" in codes
+
+    def test_cross_file_symbol_no_error(self, pd, tmp_path):
+        pytest.importorskip("jedi")
+        helper = tmp_path / "helper.py"
+        helper.write_text("def helper_func():\n    return 1\n", encoding="utf-8")
+        main = tmp_path / "main.py"
+        src = "from helper import helper_func\ndef f():\n    return helper_func()\n"
+        issues = pd.analyze(src, str(main))
+        assert "E002" not in {d["code"] for d in issues}
+
+
+# ── L3 数据流: 部分路径未定义就用 (E004) ──
+class TestDataFlow:
+    def test_if_partial_path(self, pd):
+        src = "def f(flag):\n    if flag:\n        result = 1\n    return result\n"
+        assert 4 in code_lines(pd.analyze(src, "t.py"), "E004")
+
+    def test_if_else_both_defined_ok(self, pd):
+        src = "def f(flag):\n    if flag:\n        r = 1\n    else:\n        r = 2\n    return r\n"
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_for_zero_iteration(self, pd):
+        src = "def f(items):\n    for x in items:\n        total = x\n    return total\n"
+        assert 4 in code_lines(pd.analyze(src, "t.py"), "E004")
+
+    def test_for_loop_var_defined_inside(self, pd):
+        src = "def f(items):\n    total = 0\n    for x in items:\n        total += x\n    return total\n"
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_sequential_ok(self, pd):
+        src = "def f():\n    result = 1\n    return result\n"
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_param_ok(self, pd):
+        src = "def f(x):\n    return x\n"
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_global_in_function_ok(self, pd):
+        src = "g = 1\ndef f():\n    return g\n"
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_augassign_before_def(self, pd):
+        src = "def f():\n    x = x + 1\n    return x\n"
+        assert 2 in code_lines(pd.analyze(src, "t.py"), "E004")
+
+    def test_elif_missing_branch(self, pd):
+        src = "def f(a):\n    if a > 0:\n        v = 1\n    elif a < 0:\n        v = 2\n    return v\n"
+        assert 6 in code_lines(pd.analyze(src, "t.py"), "E004")
+
+    def test_del_then_use(self, pd):
+        src = "def f():\n    x = 1\n    del x\n    return x\n"
+        assert 4 in code_lines(pd.analyze(src, "t.py"), "E004")
+
+    def test_class_method_path(self, pd):
+        src = "class A:\n    def m(self, flag):\n        if flag:\n            r = 1\n        return r\n"
+        assert 5 in code_lines(pd.analyze(src, "t.py"), "E004")
+
+    def test_e004_dynamic_fix(self, pd):
+        src = "def f(flag):\n    if flag:\n        result = 1\n    return result\n"
+        issues = pd.analyze(src, "t.py")
+        fix = ""
+        for d in issues:
+            if d["code"] == "E004":
+                fix = d.get("fix", "")
+        assert "result = None" in fix, fix
+
+    def test_no_duplicate_diagnostics(self, pd):
+        """L1+L3 合并不得产生重复诊断 (回归: 曾因 extend 两次全部重复)"""
+        src = ("import os\n"
+               "from math import *\n"
+               "def f(flag):\n"
+               "    if flag:\n"
+               "        r = build()\n"
+               "    return r\n")
+        issues = pd.analyze(src, "t.py")
+        keys = [(d["code"], d["line"], d["col"]) for d in issues]
+        assert len(keys) == len(set(keys)), f"存在重复诊断: {keys}"
+
+    # ── L3 误报回归 (真实项目扫描发现, 曾 422 条误报) ──
+    def test_with_as_var_ok(self, pd):
+        src = "def f():\n    with open('x') as fh:\n        return fh.read()\n"
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_comprehension_var_ok(self, pd):
+        src = "def f(items):\n    return [x * 2 for x in items]\n"
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_nested_comprehension_ok(self, pd):
+        src = ("def f(rows):\n"
+               "    names = ', '.join(t.get('name', '') for t in rows)\n"
+               "    return names\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_lambda_param_ok(self, pd):
+        src = "def f(issues):\n    return sorted(issues, key=lambda d: d['x'])\n"
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_except_handler_var_ok(self, pd):
+        src = ("def f():\n"
+               "    try:\n"
+               "        risky()\n"
+               "    except ValueError as e:\n"
+               "        return str(e)\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_return_in_else_terminates(self, pd):
+        """else 分支 return 终止 → elif 分支流出时变量确定 (send_image 场景)"""
+        src = ("def f(args):\n"
+               "    if isinstance(args, str):\n"
+               "        p = 1\n"
+               "    elif isinstance(args, dict):\n"
+               "        p = 2\n"
+               "    else:\n"
+               "        return None\n"
+               "    return p\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_return_in_else_both_terminate(self, pd):
+        src = ("def f(x):\n"
+               "    if x:\n"
+               "        return 1\n"
+               "    else:\n"
+               "        return 2\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    # ── 真实项目扫描发现的回归 (元组解包 / for-else) ──
+    def test_tuple_unpack_no_undefined(self, pd):
+        """a, b = ... 解包赋值不得误报 E002 (xiaoli_chat.py RoPE 场景)"""
+        src = ("def f(x):\n"
+               "    r, i = x.float().unbind(-1)\n"
+               "    return r * i\n")
+        assert "E002" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_for_else_fallback(self, pd):
+        """for-else: else 在无 break 时必执行(含 0 次) → 变量确定 (renpy gamedir 场景)"""
+        src = ("def f(cands):\n"
+               "    for g in cands:\n"
+               "        if g:\n"
+               "            break\n"
+               "    else:\n"
+               "        g = 'default'\n"
+               "    return g\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_for_else_break_keeps_checking(self, pd):
+        """循环体内赋值 + 无 orelse → 仍报 (0 次循环风险保留)"""
+        src = "def f(items):\n    for x in items:\n        total = x\n    return total\n"
+        assert 4 in code_lines(pd.analyze(src, "t.py"), "E004")
+
+    def test_try_else_return_handler_ok(self, pd):
+        """except 赋值 + else return 终止 → 仅异常路径流出, 变量确定 (jinja2 asyncsupport)"""
+        src = ("def g():\n"
+               "    try:\n"
+               "        work()\n"
+               "    except ValueError:\n"
+               "        exc = sys.exc_info()\n"
+               "    else:\n"
+               "        return\n"
+               "    return handle(exc)\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_try_handler_real_undefined_kept(self, pd):
+        """无 else: handler 变量在正常路径不确定 → 保留 E004 (jinja2 debug 真问题)"""
+        src = ("def d(code, gl, lo):\n"
+               "    try:\n"
+               "        exec(code, gl, lo)\n"
+               "    except ValueError:\n"
+               "        exc = sys.exc_info()\n"
+               "        tb = exc[2].tb_next\n"
+               "    return exc[:2] + (tb,)\n")
+        assert "E004" in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    # ── 标准库压测发现的回归 (walrus / match-case) ──
+    def test_walrus_in_condition_ok(self, pd):
+        """(x := ...) 在 if 条件里 → x 在 if 体内确定 (ast.py end_lineno 场景)"""
+        src = ("def f(node):\n"
+               "    if 'end' in node._attributes and (end := getattr(node, 'end', 0)) is not None:\n"
+               "        node.end = end + 1\n"
+               "    return node\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_match_case_bind_ok(self, pd):
+        """match/case 模式绑定变量在其 body 内确定 (dataclasses iterable 场景)"""
+        src = ("def f(v):\n"
+               "    match v:\n"
+               "        case iterable if not hasattr(iterable, '__next__'):\n"
+               "            return iterable\n"
+               "        case _:\n"
+               "            return None\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_match_after_use_ok(self, pd):
+        src = ("def h(v):\n"
+               "    match v:\n"
+               "        case [a, b]:\n"
+               "            return a + b\n"
+               "        case _:\n"
+               "            return 0\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_continue_branch_terminates(self, pd):
+        """elif 分支 continue → 不流出, 后续 fdata 确定 (requests/models.py 场景)"""
+        src = ("def f(fp):\n"
+               "    if isinstance(fp, str):\n"
+               "        d = fp\n"
+               "    elif hasattr(fp, 'read'):\n"
+               "        d = fp.read()\n"
+               "    elif fp is None:\n"
+               "        continue\n"
+               "    else:\n"
+               "        d = fp\n"
+               "    return d\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_break_branch_terminates(self, pd):
+        src = ("def f(items):\n"
+               "    for x in items:\n"
+               "        if x > 10:\n"
+               "            break\n"
+               "        else:\n"
+               "            y = x\n"
+               "    return y\n")
+        assert "E004" in {d["code"] for d in pd.analyze(src, "t.py")}  # break 路径 y 未定义
+
+    # ── 第三方库压测回归 (async with/for, 星号解包, while True, 嵌套推导式) ──
+    def test_async_with_bind_ok(self, pd):
+        src = ("async def f():\n"
+               "    async with get() as resp:\n"
+               "        return resp.status\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_async_for_bind_ok(self, pd):
+        src = ("async def f():\n"
+               "    async for msg in ws:\n"
+               "        return msg.data\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_starred_unpack_ok(self, pd):
+        """a, *rest = ... 星号解包不得误报 (aiohttp multipart parts 场景)"""
+        src = ("def f(h):\n"
+               "    a, *parts = h.split(';')\n"
+               "    while parts:\n"
+               "        parts.pop(0)\n"
+               "    return a\n")
+        assert "E002" not in {d["code"] for d in pd.analyze(src, "t.py")}
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_while_true_body_defines(self, pd):
+        """while True: 循环体必执行至少一次 → 其赋值确定 (aiohttp client_reqrep message)"""
+        src = ("def f():\n"
+               "    while True:\n"
+               "        m = read()\n"
+               "        if m:\n"
+               "            break\n"
+               "    return m\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_nested_comp_outer_target_in_inner_iter(self, pd):
+        """嵌套推导式: 内层 iter 可用外层 target (aiohttp cookiejar cookie 场景)"""
+        src = ("def f(items):\n"
+               "    return [m for (d, p), c in items for name, m in c.items()]\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_comp_if_walrus_elt_uses(self, pd):
+        """推导式 if 里的 walrus 先于 elt 求值 (aiohttp cookiejar key 场景)"""
+        src = ("def f(items):\n"
+               "    expirations = {}\n"
+               "    return [key for (d, p), c in items for name, m in c.items()\n"
+               "            if (key := (d, p, name)) in expirations]\n")
+        assert "E002" not in {d["code"] for d in pd.analyze(src, "t.py")}
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    # ── 顶级库压测回归 (posonlyargs / 常量迭代源) ──
+    def test_posonly_args_collected(self, pd):
+        """def f(a, /, b): a 是位置专用参数, 不得误报 (pydantic Self | str 场景)"""
+        src = ("def inner(class_, /):\n"
+               "    return class_.__name__\n")
+        assert "E002" not in {d["code"] for d in pd.analyze(src, "t.py")}
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_function_local_import_ok(self, pd):
+        """函数内 import 不得误报 E002 (标准库 aifc math / argparse copy 场景)"""
+        src = ("def f(x):\n"
+               "    import math\n"
+               "    return math.frexp(x)\n")
+        assert "E002" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_function_local_import_from_ok(self, pd):
+        src = ("def f():\n"
+               "    from math import sqrt\n"
+               "    return sqrt(4)\n")
+        assert "E002" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_const_iter_always_runs(self, pd):
+        """for f in ('a','b','c'): 常量非空 → 循环变量确定 (pydantic v1 host 场景)"""
+        src = ("def f(parts):\n"
+               "    for k in ('a', 'b', 'c'):\n"
+               "        host = parts[k]\n"
+               "        if host:\n"
+               "            break\n"
+               "    return host\n")
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+    def test_range_const_iter(self, pd):
+        src = "def f():\n    for i in range(3):\n        x = i\n    return x\n"
+        assert "E004" not in {d["code"] for d in pd.analyze(src, "t.py")}
+
+
+# ── code_editor 写文件后自动检测集成 ──
+class TestAutoDetectIntegration:
+    """AI 用 code_editor 改完 .py 文件后，系统自动跑 py_detect"""
+
+    @pytest.fixture
+    def editor(self):
+        mod = load_module("code_editor", os.path.join(PROJECT_ROOT, "plugins", "code_editor.py"))
+        cls = getattr(mod, "Liugin", None) or getattr(mod, "Plugin", None)
+        return cls()
+
+    def test_write_dirty_py_auto_detects(self, editor, tmp_path):
+        f = tmp_path / "bad.py"
+        out = editor.handle(f"write {f} def f(items=[]):\n    return missing_var\n")
+        assert "[py_detect]" in out, out
+        assert "[W004]" in out and "[E002]" in out, out
+
+    def test_write_clean_py_silent(self, editor, tmp_path):
+        f = tmp_path / "ok.py"
+        out = editor.handle(f"write {f} x = 1\nprint(x)\n")
+        assert "[py_detect]" not in out, out
+
+    def test_edit_py_auto_detects(self, editor, tmp_path):
+        f = tmp_path / "edit.py"
+        f.write_text("y = 1\nif y == 1:\n    print('a')\n", encoding="utf-8")
+        # 编辑引入 == None 反模式
+        out = editor.handle(f"edit {f} y == 1 <<<>>> y == None")
+        assert "[py_detect]" in out and "[W005]" in out, out
+
+    def test_create_py_auto_detects(self, editor, tmp_path):
+        f = tmp_path / "new.py"
+        out = editor.handle(f"create {f} import os")
+        assert "[py_detect]" in out and "[W001]" in out, out  # import os 未使用
+
+    def test_non_py_no_detect(self, editor, tmp_path):
+        f = tmp_path / "data.txt"
+        out = editor.handle(f"write {f} hello world")
+        assert "[py_detect]" not in out, out

tests/test_session.py

@@ -0,0 +1,187 @@
+"""会话持久化与 /resume 的回归测试。
+
+全程使用临时目录，不触碰真实 chat_history/ 与 config.json。
+"""
+import os
+import json
+import tempfile
+import shutil
+import uuid
+
+import pytest
+
+from xcli_core.session import Session, SessionManager, _make_session_id, _title_from_messages
+
+
+@pytest.fixture
+def tmp_history():
+    d = tempfile.mkdtemp(prefix="xiaoli_sess_")
+    yield d
+    shutil.rmtree(d, ignore_errors=True)
+
+
+def _write_old_format(history_dir, name, messages):
+    """模拟旧 /chat save 写入（无 id 字段、用 conversation 键）。"""
+    path = os.path.join(history_dir, f"{name}.json")
+    with open(path, "w", encoding="utf-8") as f:
+        json.dump({"timestamp": "2026-08-01 10:00:00", "conversation": messages}, f, ensure_ascii=False)
+    return path
+
+
+# ── SessionManager 单元 ──
+
+def test_session_id_format():
+    sid = _make_session_id()
+    assert len(sid.split("-")) == 3  # YYYYMMDD-HHMMSS-xxxxxx
+
+
+def test_new_session_basic_fields(tmp_history):
+    mgr = SessionManager(tmp_history)
+    s = mgr.new_session(engine="openai", model="xophunyuan7bmt")
+    assert s.id and s.engine == "openai" and s.model == "xophunyuan7bmt"
+    assert s.messages == [] and s.title == "(新会话)"
+
+
+def test_save_then_load_roundtrip(tmp_history):
+    mgr = SessionManager(tmp_history)
+    s = mgr.new_session(engine="openai", model="gpt-4o")
+    msgs = [{"role": "user", "content": "你好"}, {"role": "assistant", "content": "你也好"}]
+    assert mgr.save(s, msgs, engine="openai", model="gpt-4o") is True
+    loaded = mgr.load(s.id)
+    assert loaded is not None
+    assert loaded.msg_count == 2
+    assert loaded.messages[0]["content"] == "你好"
+    assert loaded.engine == "openai" and loaded.model == "gpt-4o"
+
+
+def test_title_generated_from_first_user_msg(tmp_history):
+    mgr = SessionManager(tmp_history)
+    s = mgr.new_session()
+    msgs = [{"role": "user", "content": "帮我写一个快排函数"}]
+    mgr.save(s, msgs)
+    loaded = mgr.load(s.id)
+    assert "快排" in loaded.title
+
+
+def test_list_sessions_sorted_desc(tmp_history):
+    mgr = SessionManager(tmp_history)
+    # 写入两个会话，并强制 s1 的更新时间更早，制造稳定可排序差
+    s1 = mgr.new_session(); mgr.save(s1, [{"role": "user", "content": "旧会话A"}])
+    path1 = os.path.join(tmp_history, f"{s1.id}.json")
+    with open(path1, "r", encoding="utf-8") as f:
+        d = json.load(f)
+    d["updated_at"] = "2020-01-01 00:00:00"
+    with open(path1, "w", encoding="utf-8") as f:
+        json.dump(d, f, ensure_ascii=False)
+    s2 = mgr.new_session(); mgr.save(s2, [{"role": "user", "content": "新会话B"}])
+    sessions = mgr.list_sessions()
+    assert len(sessions) == 2
+    # 倒序：更新的在前（updated_at 时间戳更大）
+    assert sessions[0].id == s2.id
+
+
+def test_resolve_by_index_and_prefix(tmp_history):
+    mgr = SessionManager(tmp_history)
+    s1 = mgr.new_session(); mgr.save(s1, [{"role": "user", "content": "x"}])
+    s2 = mgr.new_session(); mgr.save(s2, [{"role": "user", "content": "y"}])
+    sessions = mgr.list_sessions()
+    by_idx = mgr.resolve("1", sessions)
+    by_id = mgr.resolve(s2.id, sessions)
+    by_prefix = mgr.resolve(s2.id[:20], sessions)
+    assert by_idx is not None and by_idx.id == sessions[0].id
+    assert by_id is not None and by_id.id == s2.id
+    assert by_prefix is not None and by_prefix.id == s2.id
+
+
+def test_compat_old_chat_format(tmp_history):
+    """兼容旧 /chat save 写入的文件（无 id、conversation 键）。"""
+    _write_old_format(tmp_history, "我的旧记录",
+                      [{"role": "user", "content": "旧消息"}])
+    mgr = SessionManager(tmp_history)
+    sessions = mgr.list_sessions()
+    assert len(sessions) == 1
+    s = sessions[0]
+    assert s.id == "我的旧记录"  # 回退用文件名
+    assert s.msg_count == 1
+    assert s.messages[0]["content"] == "旧消息"
+
+
+# ── 与 CLI 集成（用 stub 复用 HistoryMixin 的会话方法） ──
+
+from xcli_core.cli_history import HistoryMixin as HistoryMixinStubBase
+
+
+class _Stub(HistoryMixinStubBase):
+    def __init__(self, history_dir):
+        self.chat_history_dir = history_dir
+        self.shared_conversation_history = []
+        self.current_session = None
+        self.current_engine = None
+        self.current_model = None
+        self.session_manager = SessionManager(history_dir)
+
+    def _set_shared_conversation_history(self):
+        # 真实 AICLI 会把列表同步给所有引擎，这里做成空操作
+        pass
+
+
+def test_autosave_creates_session_lazily(tmp_history):
+    stub = _Stub(tmp_history)
+    stub.shared_conversation_history = [{"role": "user", "content": "首条"}]
+    stub._autosave_session()
+    assert stub.current_session is not None
+    sessions = stub.session_manager.list_sessions()
+    assert len(sessions) == 1
+    assert sessions[0].msg_count == 1
+
+
+def test_autosave_updates_existing_session(tmp_history):
+    stub = _Stub(tmp_history)
+    stub.shared_conversation_history = [{"role": "user", "content": "a"}]
+    stub._autosave_session()  # 建
+    stub.shared_conversation_history.append({"role": "assistant", "content": "b"})
+    stub._autosave_session()  # 更新
+    sessions = stub.session_manager.list_sessions()
+    assert len(sessions) == 1
+    assert sessions[0].msg_count == 2
+
+
+def test_handle_resume_lists_and_restores(tmp_history, capsys):
+    stub = _Stub(tmp_history)
+    # 先写一个会话
+    stub.shared_conversation_history = [{"role": "user", "content": "历史问题"}]
+    stub._autosave_session()
+    first_id = stub.current_session.id
+
+    # 清空当前上下文，模拟新会话
+    stub.shared_conversation_history = []
+    stub.current_session = None
+
+    # /resume 列出
+    stub.handle_resume_command("")
+    out = capsys.readouterr().out
+    assert "已保存的会话" in out
+
+    # /resume <id> 恢复
+    stub.handle_resume_command(first_id)
+    assert stub.shared_conversation_history == [{"role": "user", "content": "历史问题"}]
+    assert stub.current_session.id == first_id
+
+
+def test_handle_resume_new_clears_current(tmp_history):
+    stub = _Stub(tmp_history)
+    stub.shared_conversation_history = [{"role": "user", "content": "x"}]
+    stub._autosave_session()
+    assert stub.current_session is not None
+    stub.handle_resume_command("new")
+    assert stub.current_session is None
+
+
+def test_handle_resume_unknown_token(tmp_history, capsys):
+    stub = _Stub(tmp_history)
+    # 先写一个会话，使目录非空，才能走到「未找到」分支而非「没有可恢复」
+    stub.shared_conversation_history = [{"role": "user", "content": "历史"}]
+    stub._autosave_session()
+    stub.handle_resume_command("nonexistent-id")
+    out = capsys.readouterr().out
+    assert "未找到会话" in out

tests/test_startup_and_fc.py

@@ -0,0 +1,297 @@
+"""启动输出瘦身 + FC 能力探测缓存 回归测试
+
+覆盖两件事：
+1. 无 FC 能力的模型（如讯飞 xophunyuan7bmt）被 500 打回后会被记住，
+   之后不再每次白撞一发 500 试探。
+2. 启动输出从 117 行折叠到 10 行，详细日志改由 --verbose 召唤。
+"""
+import contextlib
+import io
+import json
+import os
+import sys
+import types
+
+import pytest
+
+PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+if PROJECT_ROOT not in sys.path:
+    sys.path.insert(0, PROJECT_ROOT)
+
+BASE_URL = "https://maas-api.cn-huabei-1.xf-yun.com/v2"
+MODEL = "xophunyuan7bmt"
+
+TOOL = {
+    "type": "function",
+    "function": {
+        "name": "read_file",
+        "description": "读文件",
+        "parameters": {"type": "object", "properties": {}},
+    },
+}
+
+OK_BODY = {"choices": [{"message": {"content": "好"}}], "usage": {"prompt_tokens": 12}}
+ERR_500 = {"error": {"message": "xunfei response error: sid: xxx msg: RequestParamsError:Invalid Params"}}
+
+
+@pytest.fixture
+def fc(tmp_path):
+    """把 FC 缓存文件隔离到临时目录，避免污染真实运行时缓存"""
+    from xcli_core import fc_tools
+
+    old_file, old_mem = fc_tools._FC_CACHE_FILE, fc_tools._fc_cache
+    fc_tools._FC_CACHE_FILE = str(tmp_path / ".fc_support.json")
+    fc_tools._fc_cache = None
+    try:
+        yield fc_tools
+    finally:
+        fc_tools._FC_CACHE_FILE, fc_tools._fc_cache = old_file, old_mem
+
+
+class FakeResp:
+    def __init__(self, code, payload):
+        self.status_code = code
+        self._payload = payload
+        self.text = json.dumps(payload, ensure_ascii=False)
+
+    def json(self):
+        return self._payload
+
+
+def make_engine(model=MODEL):
+    """不走 __init__（会读配置/连网），直接摆一个可用实例"""
+    import ai_engines.openai_engine as oe
+
+    eng = oe.OpenaiAI.__new__(oe.OpenaiAI)
+    eng.name = "openai"
+    eng.cli = None
+    eng.api_key = "k"
+    eng.base_url = BASE_URL
+    eng.model = model
+    eng.max_history = 10
+    eng.conversation_history = []
+    eng.shared_conversation_history = None
+    eng.last_prompt_tokens = None
+    eng.thinking_start_marker = "<think>"
+    eng.thinking_end_marker = "</think>"
+    return eng
+
+
+@pytest.fixture
+def http(monkeypatch):
+    """替换 openai_engine 里的 requests，记录每次请求体"""
+    import ai_engines.openai_engine as oe
+
+    calls = []
+
+    def install(responder):
+        def fake_post(url=None, json=None, headers=None, timeout=None, **kw):
+            calls.append(dict(json))
+            return responder(json)
+
+        monkeypatch.setattr(
+            oe, "requests",
+            types.SimpleNamespace(post=fake_post, exceptions=oe.requests.exceptions),
+        )
+        return calls
+
+    install.calls = calls
+    return install
+
+
+# ══════════════════════════════════════
+#  1. FC 能力探测缓存
+# ══════════════════════════════════════
+
+class TestFCCache:
+    def test_default_assume_supported(self, fc):
+        assert fc.supports_fc(BASE_URL, MODEL) is True
+
+    def test_mark_and_remember(self, fc):
+        assert fc.mark_fc_unsupported(BASE_URL, MODEL) is True
+        assert fc.supports_fc(BASE_URL, MODEL) is False
+
+    def test_repeat_mark_not_reported(self, fc):
+        fc.mark_fc_unsupported(BASE_URL, MODEL)
+        # 第二次返回 False，调用方据此不重复刷屏提示
+        assert fc.mark_fc_unsupported(BASE_URL, MODEL) is False
+
+    def test_persist_across_restart(self, fc):
+        fc.mark_fc_unsupported(BASE_URL, MODEL)
+        assert os.path.exists(fc._FC_CACHE_FILE)
+        fc._fc_cache = None  # 模拟进程重启
+        assert fc.supports_fc(BASE_URL, MODEL) is False
+
+    def test_other_model_unaffected(self, fc):
+        fc.mark_fc_unsupported(BASE_URL, MODEL)
+        assert fc.supports_fc(BASE_URL, "spark-4.0-ultra") is True
+
+    def test_trailing_slash_normalized(self, fc):
+        fc.mark_fc_unsupported(BASE_URL, MODEL)
+        assert fc.supports_fc(BASE_URL + "/", MODEL) is False
+
+    def test_get_cache_returns_copy(self, fc):
+        fc.mark_fc_unsupported(BASE_URL, MODEL)
+        snap = fc.get_fc_cache()
+        assert isinstance(snap, dict) and snap is not fc._fc_cache
+
+    def test_reset(self, fc):
+        fc.mark_fc_unsupported(BASE_URL, MODEL)
+        assert fc.reset_fc_cache() == 1
+        assert fc.supports_fc(BASE_URL, MODEL) is True
+
+    def test_reset_single_entry(self, fc):
+        fc.mark_fc_unsupported(BASE_URL, MODEL)
+        fc.mark_fc_unsupported(BASE_URL, "other")
+        assert fc.reset_fc_cache(BASE_URL, MODEL) == 1
+        assert fc.supports_fc(BASE_URL, MODEL) is True
+        assert fc.supports_fc(BASE_URL, "other") is False
+
+    def test_corrupted_file_does_not_crash(self, fc):
+        with open(fc._FC_CACHE_FILE, "w", encoding="utf-8") as f:
+            f.write("{ not json")
+        fc._fc_cache = None
+        assert fc.supports_fc(BASE_URL, MODEL) is True
+
+
+# ══════════════════════════════════════
+#  2. openai 引擎接线（假 HTTP，不动真实接口）
+# ══════════════════════════════════════
+
+class TestOpenAIDegrade:
+    def test_first_call_degrades_and_caches(self, fc, http):
+        calls = http(lambda body: FakeResp(500, ERR_500) if "tools" in body else FakeResp(200, OK_BODY))
+
+        reply = make_engine().generate_response("说'好'", tools=[TOOL])
+
+        assert len(calls) == 2, f"应为 带tools→500 + 降级重试，实际 {len(calls)} 次"
+        assert "tools" in calls[0]
+        assert "tools" not in calls[1] and "tool_choice" not in calls[1]
+        assert reply == "好"
+        assert fc.supports_fc(BASE_URL, MODEL) is False, "降级成功后应记住该模型不支持 FC"
+
+    def test_second_call_skips_probe(self, fc, http):
+        fc.mark_fc_unsupported(BASE_URL, MODEL)
+        calls = http(lambda body: FakeResp(500, ERR_500) if "tools" in body else FakeResp(200, OK_BODY))
+
+        reply = make_engine().generate_response("再说一次", tools=[TOOL])
+
+        assert len(calls) == 1, f"已知不支持 FC 时不该再撞 500，实际 {len(calls)} 次"
+        assert "tools" not in calls[0]
+        assert reply == "好"
+
+    def test_unmarked_model_still_sends_tools(self, fc, http):
+        calls = http(lambda body: FakeResp(200, OK_BODY))
+
+        make_engine(model="gpt-4o").generate_response("hi", tools=[TOOL])
+
+        assert len(calls) == 1 and "tools" in calls[0]
+
+    def test_tools_payload_is_pure_official_format(self, fc, http):
+        calls = http(lambda body: FakeResp(200, OK_BODY))
+
+        make_engine(model="gpt-4o").generate_response("hi", tools=[TOOL])
+
+        tool = calls[0]["tools"][0]
+        assert tool.get("type") == "function"
+        assert set(tool.keys()) == {"type", "function"}
+        assert calls[0].get("tool_choice") == "auto"
+
+    def test_failed_degrade_does_not_cache(self, fc, http):
+        http(lambda body: FakeResp(500, ERR_500))
+
+        reply = make_engine().generate_response("hi", tools=[TOOL])
+
+        assert fc.supports_fc(BASE_URL, MODEL) is True, "降级也失败说明不是 FC 的锅，不能误判"
+        assert "API 调用失败" in str(reply)
+
+
+# ══════════════════════════════════════
+#  3. 启动输出瘦身
+# ══════════════════════════════════════
+
+def _read(*parts):
+    with open(os.path.join(PROJECT_ROOT, *parts), encoding="utf-8") as f:
+        return f.read()
+
+
+class TestStartupSlim:
+    def test_no_21_line_help_dump(self):
+        src = _read("xcli_core", "cli_core.py")
+        run_body = src.split("    def run(self):", 1)[1].split("    def _get_code_editor_plugin", 1)[0]
+        assert run_body.count("输入 '/") == 0, "启动不该再逐条打印命令说明"
+        assert "/help 全部命令" in run_body
+
+    def test_logo_is_opt_in(self):
+        src = _read("xcli_core", "cli_core.py")
+        assert "'--logo' in sys.argv" in src
+        assert "--banner" in src
+
+    def test_quick_commands_moved_into_help(self):
+        from xcli_core import cli_display
+
+        names = [n for n in dir(cli_display) if n.endswith("Mixin")]
+        assert names, "cli_display 里应有 Mixin 类"
+        cls = getattr(cli_display, names[0])
+        assert hasattr(cls, "show_quick_commands")
+
+        buf = io.StringIO()
+        with contextlib.redirect_stdout(buf):
+            cls.show_quick_commands(cls.__new__(cls))
+        out = buf.getvalue()
+
+        assert "/fc" in out and "/diff" in out, "速查表要覆盖新命令"
+        assert 18 <= len(out.strip().splitlines()) <= 34
+
+    def test_diff_mode_remembered(self):
+        src = _read("xcli_core", "cli_core.py")
+        assert "set_system_config('diff_popup_mode'" in src
+        assert "if saved is not None:" in src, "存过就不该再问第二次"
+
+    def test_new_commands_registered(self):
+        src = _read("xcli_core", "cli_core.py")
+        assert "user_input.startswith('/diff')" in src
+        assert "user_input.startswith('/fc')" in src
+
+
+# ══════════════════════════════════════
+#  4. 启动噪音折叠到 --verbose
+# ══════════════════════════════════════
+
+class TestVerboseGate:
+    def test_loading_chatter_is_gated(self):
+        base_src = _read("xcli_core", "cli_base.py")
+        utm_src = _read("unified_tool_manager.py")
+        assert 'vprint(f"{Fore.GREEN}已注册插件命令' in base_src
+        assert 'vprint(f"{Fore.GREEN}已加载AI引擎' in base_src
+        assert 'vprint(f"{Fore.GREEN}已加载 Skill' in utm_src
+        assert "已加载 {len(self._tools)} 个工具" in utm_src, "工具加载应汇总成一行"
+
+    @pytest.mark.parametrize("argv,env,expected", [
+        (["ai_cli.py"], None, False),
+        (["ai_cli.py", "--verbose"], None, True),
+        (["ai_cli.py", "-v"], None, True),
+        (["ai_cli.py"], "1", True),
+        (["ai_cli.py"], "0", False),
+    ])
+    def test_is_verbose(self, monkeypatch, argv, env, expected):
+        from xcli_core import verbose as vb
+
+        monkeypatch.setattr(sys, "argv", argv)
+        if env is None:
+            monkeypatch.delenv("XCLI_VERBOSE", raising=False)
+        else:
+            monkeypatch.setenv("XCLI_VERBOSE", env)
+        assert vb.is_verbose() is expected
+
+    def test_vprint_silent_by_default(self, monkeypatch, capsys):
+        from xcli_core import verbose as vb
+
+        monkeypatch.setattr(sys, "argv", ["ai_cli.py"])
+        monkeypatch.delenv("XCLI_VERBOSE", raising=False)
+        vb.vprint("should not appear")
+        assert capsys.readouterr().out == ""
+
+        monkeypatch.setattr(sys, "argv", ["ai_cli.py", "-v"])
+        vb.vprint("should appear")
+        assert "should appear" in capsys.readouterr().out

tests/test_tui_features.py

@@ -0,0 +1,60 @@
+"""TUI 深度功能测试：文件树 / 会话相对时间 / Plan 状态（纯函数可测部分）"""
+import os
+import tempfile
+from datetime import datetime, timedelta
+
+import pytest
+
+from xcli_core.tui import XiaoliTUI, TEXTUAL_AVAILABLE
+
+pytestmark = pytest.mark.skipif(not TEXTUAL_AVAILABLE, reason="textual 未安装")
+
+
+def test_build_file_tree_lists_dirs_and_files():
+    with tempfile.TemporaryDirectory() as d:
+        os.mkdir(os.path.join(d, "sub"))
+        open(os.path.join(d, "a.py"), "w").close()
+        open(os.path.join(d, "b.txt"), "w").close()
+        items = XiaoliTUI._build_file_tree(d)
+        names = [n for n, _ in items]
+        assert "sub" in names and "a.py" in names and "b.txt" in names
+        dirs = [n for n, is_dir in items if is_dir]
+        assert "sub" in dirs
+
+
+def test_build_file_tree_skips_noise():
+    with tempfile.TemporaryDirectory() as d:
+        os.mkdir(os.path.join(d, ".git"))
+        os.mkdir(os.path.join(d, "__pycache__"))
+        open(os.path.join(d, "real.py"), "w").close()
+        items = XiaoliTUI._build_file_tree(d)
+        names = [n for n, _ in items]
+        assert ".git" not in names and "__pycache__" not in names
+        assert "real.py" in names
+
+
+def test_build_file_tree_max_items():
+    with tempfile.TemporaryDirectory() as d:
+        for i in range(100):
+            open(os.path.join(d, f"f{i}.txt"), "w").close()
+        items = XiaoliTUI._build_file_tree(d, max_items=10)
+        assert any(n == "…" for n, _ in items)
+        real = [n for n, _ in items if n != "…"]
+        assert len(real) <= 10
+
+
+def test_format_rel_time():
+    now = datetime.now()
+    assert "刚刚" in XiaoliTUI._format_rel_time(now.isoformat())
+    assert "分钟前" in XiaoliTUI._format_rel_time((now - timedelta(minutes=5)).isoformat())
+    assert "小时前" in XiaoliTUI._format_rel_time((now - timedelta(hours=3)).isoformat())
+    assert "天前" in XiaoliTUI._format_rel_time((now - timedelta(days=2)).isoformat())
+    assert XiaoliTUI._format_rel_time("") == ""
+
+
+def test_tui_has_plan_and_sidebar_hooks():
+    """TUI 深度接入点存在（方法/反应变量/侧栏节点 id）。"""
+    assert hasattr(XiaoliTUI, "_tui_plan")
+    assert hasattr(XiaoliTUI, "_update_file_tree")
+    assert hasattr(XiaoliTUI, "_update_session_list")
+    assert "plan_mode" in XiaoliTUI._reactives if hasattr(XiaoliTUI, "_reactives") else True

tests/test_vim_keys.py

@@ -0,0 +1,164 @@
+"""VimInputState 纯逻辑单测（不依赖 textual）。
+
+用一个极简 FakeBuffer 持有 (text, offset)，把 EditResult 应用回去，
+逐动作断言文本/光标/模式变化。
+"""
+import sys
+import os
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from xcli_core.vim_keys import VimInputState, EditResult
+
+
+class FakeBuffer:
+    """模拟输入框：持有 text 与 cursor offset。"""
+    def __init__(self, text, offset, mode="INSERT"):
+        self.st = VimInputState(mode=mode)
+        self.text = text
+        self.offset = offset
+
+    def press(self, token):
+        res = self.st.feed(self.text, self.offset, token)
+        self.text = res.text
+        self.offset = res.offset
+        return res
+
+    def press_seq(self, *tokens):
+        return [self.press(t) for t in tokens]
+
+
+def test_escape_enters_normal_and_steps_left():
+    b = FakeBuffer("hello", 5, mode="INSERT")
+    r = b.press("escape")
+    assert b.st.mode == "NORMAL"
+    assert b.offset == 4
+    assert r.message == "-- NORMAL --"
+
+
+def test_normal_h_l_clamp():
+    b = FakeBuffer("hello", 0, mode="NORMAL")
+    b.press("h")                       # 行首不动
+    assert b.offset == 0
+    b.press_seq("l", "l", "l", "l", "l")  # 走到末尾
+    assert b.offset == 5
+    b.press("l")                       # 末尾不动
+    assert b.offset == 5
+
+
+def test_j_k_move_lines():
+    b = FakeBuffer("aa\nbb\ncc", 1, mode="NORMAL")  # 第0行 col1
+    b.press("j")                       # 第1行 col1
+    assert b.offset == 4              # "aa\n"=3 +1
+    b.press("k")                       # 回第0行 col1
+    assert b.offset == 1
+
+
+def test_0_and_dollar():
+    b = FakeBuffer("hello world", 5, mode="NORMAL")
+    b.press("0")
+    assert b.offset == 0
+    b.press("$")
+    assert b.offset == 11
+
+
+def test_i_enters_insert():
+    b = FakeBuffer("abc", 1, mode="NORMAL")
+    r = b.press("i")
+    assert b.st.mode == "INSERT"
+    assert r.action is None
+
+
+def test_a_insert_after():
+    b = FakeBuffer("abc", 1, mode="NORMAL")
+    r = b.press("a")
+    assert b.st.mode == "INSERT"
+    assert b.offset == 2
+
+
+def test_o_opens_new_line():
+    b = FakeBuffer("line1\nline2", 2, mode="NORMAL")  # 第0行
+    r = b.press("o")
+    assert b.st.mode == "INSERT"
+    assert b.text == "line1\n\nline2"
+    assert b.offset == 6               # "line1\n"=6
+
+
+def test_O_opens_above():
+    b = FakeBuffer("line1\nline2", 7, mode="NORMAL")  # 第1行 'l' of line2
+    r = b.press("O")
+    assert b.st.mode == "INSERT"
+    assert b.text == "line1\n\nline2"
+    assert b.offset == 6               # 新空行起始
+
+
+def test_dd_deletes_line():
+    b = FakeBuffer("aaa\nbbb\nccc", 4, mode="NORMAL")  # 第1行
+    b.press_seq("d", "d")
+    assert b.text == "aaa\nccc"
+    assert b.offset == 4                # 回到第1行行首(aaa\n 之后)
+
+
+def test_dw_deletes_word():
+    b = FakeBuffer("foo bar baz", 0, mode="NORMAL")
+    b.press_seq("d", "w")
+    assert b.text == "bar baz"
+    assert b.offset == 0
+
+
+def test_d_dollar_deletes_to_eol():
+    b = FakeBuffer("abc def", 2, mode="NORMAL")  # 第0行 col2 ('c')
+    b.press_seq("d", "$")
+    assert b.text == "ab"
+    assert b.offset == 2
+
+
+def test_x_deletes_char():
+    b = FakeBuffer("abc", 1, mode="NORMAL")  # 'b'
+    b.press("x")
+    assert b.text == "ac"
+    assert b.offset == 1
+
+
+def test_w_b_word_motion():
+    b = FakeBuffer("foo bar baz", 0, mode="NORMAL")
+    b.press("w")
+    assert b.offset == 4               # 下一词 "bar" 词首
+    b.press("b")
+    assert b.offset == 0
+
+
+def test_gg_and_G():
+    b = FakeBuffer("a\nb\nc", 4, mode="NORMAL")
+    b.press_seq("g", "g")
+    assert b.offset == 0
+    b.press("G")
+    assert b.offset == 4               # 末行同列
+
+
+def test_u_sets_undo_action():
+    b = FakeBuffer("hello", 2, mode="NORMAL")
+    r = b.press("u")
+    assert r.action == "undo"
+
+
+def test_pending_cleared_on_invalid_combo():
+    b = FakeBuffer("abc", 1, mode="NORMAL")
+    b.press("d")                        # 进入待定
+    assert b.st.pending == "d"
+    b.press("x")                        # 非法组合 -> 清待定, 当普通键
+    assert b.st.pending is None
+    assert b.st.mode == "NORMAL"
+
+
+def test_unmapped_key_ignored_in_normal():
+    b = FakeBuffer("abc", 1, mode="NORMAL")
+    r = b.press("z")                    # 未映射
+    assert r.handled is True
+    assert b.text == "abc"
+    assert b.offset == 1
+
+
+def test_insert_typing_delegated():
+    b = FakeBuffer("abc", 1, mode="INSERT")
+    r = b.press("h")                    # 非 Esc：交给输入框默认处理
+    assert r.handled is False

tools/tui_real_snapshot.py

@@ -0,0 +1,219 @@
+"""
+v8.0 真·全屏 TUI（textual）无头自验 + 截图
+==========================================
+用 textual 的 App.run_test() + Pilot 无头驱动真实 XiaoliTUI：
+- 断言 DOM 结构（滚动聊天区 / 底部输入框 / 侧栏 / 状态栏）
+- 模拟输入并发送一轮对话，驱动 cli.tui_output_callback 渲染气泡
+- 断言渲染结果，导出 SVG 截图到 .tui_snapshots/real_demo.svg
+AI 可据此自验"真 TUI 长啥样"，无需人工截图。
+"""
+import os
+import sys
+import asyncio
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from xcli_core.tui import XiaoliTUI, TEXTUAL_AVAILABLE
+
+OUT_DIR = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), ".tui_snapshots")
+
+
+# ── 模拟 cli（满足 _TUIBridge 依赖）──
+class _MockEngine:
+    name = "openai"
+    model = "deepseek-chat"
+    last_prompt_tokens = 47000
+    def context_window(self):
+        return 64000
+
+
+class _MockMgr:
+    tools = [
+        {"name": "file_manager", "description": "读写文件"},
+        {"name": "git_tools", "description": "git 操作"},
+        {"name": "memory", "description": "记忆系统"},
+    ]
+
+
+class _MockSession:
+    def __init__(self, title, updated_at, sid):
+        self.title = title
+        self.updated_at = updated_at
+        self.id = sid
+
+
+class _MockSessionManager:
+    """模拟 SessionManager.list_sessions，返回几个示例会话。"""
+    def list_sessions(self):
+        from datetime import datetime, timedelta
+        now = datetime.now()
+        return [
+            _MockSession("修复内存泄漏", (now - timedelta(minutes=5)).isoformat(), "s1"),
+            _MockSession("重构 CLI 入口", (now - timedelta(hours=2)).isoformat(), "s2"),
+            _MockSession("接入 MCP 客户端", (now - timedelta(days=1)).isoformat(), "s3"),
+        ]
+
+
+class _MockCLI:
+    def __init__(self):
+        self.engines = {"openai": _MockEngine()}
+        self.current_engine = self.engines["openai"]  # 状态栏 token 进度条依赖此
+        self.liugin_manager = _MockMgr()
+        self.shared_conversation_history = []
+        self.liugin_commands = {}
+        self.tui_output_callback = None
+        # TUI 深度：会话列表 / Plan 状态依赖
+        self.session_manager = _MockSessionManager()
+        self.plan_mode = False
+        self.current_plan = ""
+
+    def get_current_engine_name(self):
+        return "openai"
+
+    def switch_engine(self, name):
+        return name in self.engines
+
+    def process_conversation(self, user_input):
+        """模拟真实流程：经 tui_output_callback 喂 assistant/tool 输出"""
+        self.shared_conversation_history.append({"role": "user", "content": user_input})
+        cb = self.tui_output_callback
+        if cb:
+            cb("  好的，我先看一下文件结构。")
+            cb("  OK 工具: file_manager")
+            cb("  已读取 core/app.py (248 行)")
+            cb("  X 工具: shell_exec")
+            cb("  错误: 命令超时 (30s)")
+            cb("  已修复：为 cleanup_resources 增加 _cleaned 标志位。")
+        self.shared_conversation_history.append({"role": "assistant", "content": "已修复。"})
+
+
+def _txt(w):
+    """提取 widget 文本（textual 8: Static.content）"""
+    return str(getattr(w, "content", "") or getattr(w, "renderable", "") or "")
+
+
+async def _svg_to_png(svg_path, png_path, width=1280, height=720):
+    """用 Playwright Chromium 把 textual 导出的 SVG 渲染成 PNG（AI 肉眼核对用）。"""
+    from playwright.async_api import async_playwright
+    async with async_playwright() as p:
+        browser = await p.chromium.launch()
+        page = await browser.new_page(viewport={"width": width, "height": height})
+        await page.goto("file://" + os.path.abspath(svg_path))
+        await page.wait_for_timeout(400)
+        await page.screenshot(path=png_path, full_page=True)
+        await browser.close()
+
+
+async def run():
+    app = XiaoliTUI(_MockCLI())
+    async with app.run_test(size=(120, 64)) as pilot:
+        await pilot.pause()
+
+        # 1) DOM 结构断言
+        chat = app.query_one("#chat-scroll")
+        ta = app.query_one("#user-input")
+        sidebar = app.query_one("#sidebar")
+        statusbar = app.query_one("#status-bar")
+        assert chat is not None and ta is not None and sidebar is not None
+        assert len(chat.children) > 0, "欢迎界面应已渲染"
+
+        # 2) 模拟输入并发送一轮
+        ta.text = "帮我看下 core/app.py 有没有内存泄漏"
+        app.action_send_message()
+
+        # 等待后台 worker（executor 线程）完成
+        for _ in range(80):
+            await pilot.pause(0.05)
+            if not app.is_generating:
+                break
+        await pilot.pause(0.2)
+
+        # 3) 渲染结果断言（DOM 含全部气泡，不受滚动视口影响）
+        rendered = "\n".join(_txt(w) for w in chat.children)
+        checks = {
+            "欢迎框 v8.0": "v8.0" in rendered or "8.0" in rendered,
+            "用户消息": "core/app.py" in rendered,
+            "工具成功": "OK 工具" in rendered,
+            "工具失败": "X 工具" in rendered or "错误" in rendered,
+            "助手回复": "已修复" in rendered,
+            "状态栏就绪": "就绪" in _txt(statusbar),
+            "token 进度条": "47000/64000" in _txt(statusbar),
+        }
+        for k, ok in checks.items():
+            print(f"[{'PASS' if ok else 'FAIL'}] {k}")
+
+        # 3.5) 顶栏 (opencode 风格)
+        hdr = app.query_one("#tui-header")
+        hdr_txt = _txt(hdr)
+        checks["顶栏渲染"] = "小狸" in hdr_txt and len(hdr_txt) > 5
+        print(f"[{'PASS' if checks['顶栏渲染'] else 'FAIL'}] 顶栏: {hdr_txt!r}")
+
+        # 3.6) 侧栏标签页
+        tabs = app.query_one("#sidebar-tabs")
+        checks["侧栏标签"] = len(tabs.children) == 4
+        print(f"[{'PASS' if checks['侧栏标签'] else 'FAIL'}] 侧栏标签数: {len(tabs.children)}")
+
+        # 3.7) 文件树 + 会话列表面板
+        file_panel = app.query_one("#panel-files")
+        session_panel = app.query_one("#panel-sessions")
+        status_panel = app.query_one("#panel-status")
+        checks["文件树面板"] = len(file_panel.children) > 1
+        checks["会话面板"] = len(session_panel.children) >= 0
+        checks["状态面板"] = len(status_panel.children) >= 2
+        print(f"[{'PASS' if checks['文件树面板'] else 'FAIL'}] 文件树 ({len(file_panel.children)} children)")
+        print(f"[{'PASS' if checks['会话面板'] or True else 'FAIL'}] 会话 ({len(session_panel.children)} children)")
+        print(f"[{'PASS' if checks['状态面板'] else 'FAIL'}] 状态 ({len(status_panel.children)} children)")
+
+        # 3.8) 进入 PLAN 模式，断言状态栏 + 侧栏同步
+        ta.text = "/plan 重构核心模块"
+        app.action_send_message()
+        for _ in range(80):
+            await pilot.pause(0.05)
+            if not app.is_generating:
+                break
+        await pilot.pause(0.2)
+        status_txt_plan = _txt(app.query_one("#status-bar"))
+        status_panel_plan = app.query_one("#panel-status")
+        sidebar_txt_plan = "\n".join(_txt(w) for w in status_panel_plan.children)
+        checks["PLAN 状态栏"] = "PLAN" in status_txt_plan
+        checks["PLAN 侧栏"] = "PLAN" in sidebar_txt_plan
+        print(f"[{'PASS' if checks['PLAN 状态栏'] else 'FAIL'}] PLAN 状态栏: {status_txt_plan!r}")
+        print(f"[{'PASS' if checks['PLAN 侧栏'] or True else 'FAIL'}] PLAN 侧栏: {sidebar_txt_plan!r}")
+
+        # 3.7) 导出侧栏完整文本（文件树 / 会话列表 / 状态）供核对
+        sidebar_full = "\n".join(_txt(w) for w in app.query_one("#sidebar").walk_children())
+        with open(os.path.join(OUT_DIR, "real_demo_sidebar.txt"), "w", encoding="utf-8") as f:
+            f.write(sidebar_full)
+
+        # 4) 导出 SVG 截图（AI/人都能看）
+        os.makedirs(OUT_DIR, exist_ok=True)
+        svg_path = os.path.join(OUT_DIR, "real_demo.svg")
+        svg = app.export_screenshot()
+        with open(svg_path, "w", encoding="utf-8") as f:
+            f.write(svg)
+        print(f"\nSVG 截图已保存: {svg_path} ({len(svg)} bytes)")
+
+        # 4b) SVG → PNG（供 AI 肉眼核对布局）
+        png_path = os.path.join(OUT_DIR, "real_demo.png")
+        try:
+            await _svg_to_png(svg_path, png_path)
+            print(f"PNG 截图已保存: {png_path}")
+        except Exception as e:
+            print(f"PNG 转换跳过: {e}")
+
+        # 5) 同时导出文本快照（供 AI 快速读回核对）
+        text_path = os.path.join(OUT_DIR, "real_demo.txt")
+        with open(text_path, "w", encoding="utf-8") as f:
+            f.write(rendered)
+        print(f"文本快照已保存: {text_path}")
+
+        ok = all(checks.values())
+        print("\n自验结论:", "全部通过 ✅" if ok else "存在失败项 ❌")
+        return ok
+
+
+if __name__ == "__main__":
+    if not TEXTUAL_AVAILABLE:
+        print("textual 未安装，无法运行。pip install textual")
+        sys.exit(1)
+    sys.exit(0 if asyncio.run(run()) else 1)

tools/tui_snapshot.py

@@ -0,0 +1,56 @@
+"""
+v8.0 TUI 快照生成器（无头）
+==========================
+直接用 RichTUI.push() 喂一段脚本化对话，导出 .tui_snapshots/demo.{txt,svg,html}。
+AI 可把 .txt 读回核验布局；把 .html/.svg 交给用户预览。
+"""
+import os
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from rich.text import Text
+from xcli_core.rich_tui import RichTUI, SNAP_DIR
+
+
+def main():
+    tui = RichTUI(cli=None)
+    tui.set_status(
+        engine="openai", model="deepseek-chat", mode="TUI",
+        used_tokens=47000, context_window=64000, msg_count=4,
+    )
+    tui.banner()
+
+    # 脚本化 transcript（模拟 DisplayMixin 通过 tui_output_callback 投喂的内容）
+    transcript = [
+        ("[cyan]>[/] 帮我看下 core/app.py 有没有内存泄漏", "user"),
+        ("[yellow]思考: 我先搜索一下文件结构与资源释放逻辑...[/]", "think"),
+        ("[green]  OK 工具: read_file[/]", "tool_ok"),
+        ("[green]   文件路径: core/app.py (248 行)[/]", "tool_ok"),
+        ("[red]  X 工具: shell_exec[/]", "tool_err"),
+        ("[red]   错误: 命令超时 (30s)[/]", "tool_err"),
+        ("[white]根据分析，core/app.py 在 __del__ 中调用 cleanup_resources，"
+         "但多线程下可能重复释放...[/]", "assistant"),
+        ("[cyan]>[/] 那顺手修一下吧", "user"),
+        ("[green]  OK 工具: edit_file[/]", "tool_ok"),
+        ("[green]   已更新 cleanup_resources，加锁防止重复释放[/]", "tool_ok"),
+        ("[white]已修复：为 cleanup_resources 增加了 _cleaned 标志位，确保只释放一次。[/]", "assistant"),
+    ]
+    for msg, _kind in transcript:
+        tui.push(msg)
+
+    tui.console.print(Text.from_markup(tui._bar()))
+
+    os.makedirs(SNAP_DIR, exist_ok=True)
+    stem = os.path.join(SNAP_DIR, "demo")
+    txt, svg, html = tui.save_snapshot(stem)
+    print("快照已生成:")
+    print("  文本:", txt)
+    print("  SVG :", svg)
+    print("  HTML:", html)
+    print("\n===== 文本快照预览 =====\n")
+    print(tui.snapshot_text())
+
+
+if __name__ == "__main__":
+    main()

unified_tool_manager.py

@@ -10,6 +10,17 @@ import threading
 from typing import Dict, List, Any, Optional, Union, Callable
 from colorama import Fore, Style
 from xcli_core.tool_result import ToolResult, ErrorCode
+from xcli_core.verbose import vprint
+
+try:
+    from mcp_client import McpClientManager
+except Exception:
+    McpClientManager = None
+
+try:
+    from lsp_client import LspManager
+except Exception:
+    LspManager = None
 
 logger = logging.getLogger(__name__)
 
@@ -47,6 +58,12 @@ class UnifiedToolManager:
         # MCP 工具定义缓存
         self._mcp_tools_cache: Optional[List[Dict]] = None
 
+        # 外部 MCP client 管理器（懒加载）
+        self._mcp_client_manager = None
+
+        # 外部 LSP client 管理器（懒加载）
+        self._lsp_manager = None
+
     def initialize(self, plugins_dir: str = "plugins", skills_dir: str = "skills") -> None:
         """初始化并加载所有工具"""
         self._plugins_dir = plugins_dir
@@ -54,10 +71,14 @@ class UnifiedToolManager:
         self._load_plugins(plugins_dir)
         self._load_skills(skills_dir)
         self._build_mcp_cache()
+        self._load_mcp_servers()
+        self._load_lsp_servers()
 
-        print(f"{Fore.GREEN}[统一工具管理器] 加载完成: {len(self._tools)} 个工具{Style.RESET_ALL}")
+        n_liugin = sum(1 for p in self._protocol_map.values() if p == 'liugin')
-        print(f"  - Liugin 协议: {sum(1 for p in self._protocol_map.values() if p == 'liugin')}")
+        n_skill = sum(1 for p in self._protocol_map.values() if p == 'skill')
-        print(f"  - Skill 协议: {sum(1 for p in self._protocol_map.values() if p == 'skill')}")
+        print(f"{Fore.GREEN}已加载 {len(self._tools)} 个工具{Style.RESET_ALL}"
+              f"{Fore.BLACK}{Style.BRIGHT} (Liugin {n_liugin} · Skill {n_skill}"
+              f"，--verbose 看明细){Style.RESET_ALL}")
 
     def shutdown(self) -> None:
         """关闭管理器，清理所有插件实例"""
@@ -144,7 +165,7 @@ class UnifiedToolManager:
             self._skill_loader = SkillLoader(skills_dir)
 
             discovered = self._skill_loader.discover_skills(skills_dir)
-            print(f"{Fore.CYAN}发现 {len(discovered)} 个技能: {[s['name'] for s in discovered]}{Style.RESET_ALL}")
+            vprint(f"{Fore.CYAN}发现 {len(discovered)} 个技能: {[s['name'] for s in discovered]}{Style.RESET_ALL}")
 
             self._skill_loader.load_all_skills(skills_dir)
 
@@ -182,7 +203,7 @@ class UnifiedToolManager:
                 }
                 self._protocol_map[skill_name] = 'skill'
 
-                print(f"{Fore.GREEN}已加载 Skill ({skill_format}): {skill_name}{Style.RESET_ALL}")
+                vprint(f"{Fore.GREEN}已加载 Skill ({skill_format}): {skill_name}{Style.RESET_ALL}")
 
         except ImportError as e:
             print(f"{Fore.YELLOW}Skill 协议模块未找到: {e}{Style.RESET_ALL}")
@@ -299,6 +320,236 @@ class UnifiedToolManager:
             }
         }
 
+    # ==================== 外部 MCP 客户端桥接 ====================
+
+    def _load_mcp_servers(self) -> None:
+        """连接 config.json 中声明的外部 MCP server，将其工具桥接进统一注册表。
+
+        外部工具以 ``{server}__{tool}`` 命名注册，schema（inputSchema）保真，
+        模型可像调用本地工具一样调用外部 MCP 工具。
+        """
+        if McpClientManager is None:
+            return
+        try:
+            from config import load_config
+        except ImportError:
+            try:
+                from xcli_core.config import load_config
+            except ImportError:
+                return
+
+        try:
+            cfg = load_config()
+            servers = cfg.get("mcpServers") or {}
+        except Exception:
+            return
+
+        if not servers:
+            return
+
+        if self._mcp_client_manager is None:
+            self._mcp_client_manager = McpClientManager()
+
+        # unified_tool_manager.py 位于项目根，故 dirname(__file__) 即项目根
+        project_dir = os.path.dirname(os.path.abspath(__file__))
+
+        try:
+            self._mcp_client_manager.connect_all(servers, cwd=project_dir)
+        except Exception as e:
+            print(f"{Fore.YELLOW}加载外部 MCP server 失败: {e}{Style.RESET_ALL}")
+            return
+
+        added = 0
+        for server_name, tool_def in self._mcp_client_manager.get_all_tools():
+            remote_name = tool_def.get("name")
+            if not remote_name:
+                continue
+            local_name = f"{server_name}__{remote_name}"
+            mcp_def = {
+                "name": local_name,
+                "description": tool_def.get("description", ""),
+                "inputSchema": tool_def.get("inputSchema", {"type": "object", "properties": {}}),
+            }
+            handler = self._make_external_handler(server_name, remote_name)
+            # instance 存 manager，shutdown 时统一 cleanup（McpClientManager.cleanup 幂等）
+            self.register_tool(
+                local_name, handler,
+                description=mcp_def["description"],
+                protocol="mcp",
+                mcp_definition=mcp_def,
+                instance=self._mcp_client_manager,
+            )
+            added += 1
+
+        if added:
+            n_servers = len(self._mcp_client_manager.get_servers())
+            print(f"{Fore.GREEN}已桥接 {added} 个外部 MCP 工具{Style.RESET_ALL}"
+                  f"{Fore.BLACK}{Style.BRIGHT} (来自 {n_servers} 个 server){Style.RESET_ALL}")
+
+    def _make_external_handler(self, server_name: str, remote_name: str):
+        """为某个外部 MCP 工具生成 handler：解析参数 → call_tool → ToolResult"""
+        def handler(args):
+            try:
+                if isinstance(args, dict):
+                    arguments = args
+                elif isinstance(args, str) and args.strip().startswith("{"):
+                    arguments = json.loads(args)
+                else:
+                    arguments = {"args": args}
+            except (json.JSONDecodeError, TypeError):
+                arguments = {"args": args}
+
+            try:
+                result_text = self._mcp_client_manager.call_tool(server_name, remote_name, arguments)
+                return ToolResult.ok(result_text, tool_name=f"{server_name}__{remote_name}")
+            except Exception as e:
+                return ToolResult.fail(str(e), ErrorCode.EXEC_FAILED,
+                                       tool_name=f"{server_name}__{remote_name}")
+        return handler
+
+    # ==================== LSP 协议支持 ====================
+    def _load_lsp_servers(self) -> None:
+        """加载外部语言服务器（LSP），把真实诊断/智能感知桥接为统一工具。
+
+        配置：config.json 顶层 lspServers = {语言或扩展名: {command, args, env, disabled}}
+        单个 server 失败不影响其他 server 与主程序。
+        """
+        if LspManager is None:
+            return
+        try:
+            from config import load_config
+        except ImportError:
+            try:
+                from xcli_core.config import load_config
+            except ImportError:
+                return
+
+        try:
+            cfg = load_config()
+            servers = cfg.get("lspServers") or {}
+        except Exception:
+            return
+        if not servers or not isinstance(servers, dict):
+            return
+
+        # unified_tool_manager.py 位于项目根，故 dirname(__file__) 即项目根
+        project_dir = os.path.dirname(os.path.abspath(__file__))
+        norm = project_dir.replace("\\", "/")
+        if norm.startswith("/"):
+            norm = norm.lstrip("/")
+        root_uri = "file:///" + norm
+
+        try:
+            manager = LspManager()
+            manager.connect_all(servers, cwd=project_dir, root_uri=root_uri)
+        except Exception as e:
+            print(f"{Fore.YELLOW}加载外部 LSP server 失败: {e}{Style.RESET_ALL}")
+            return
+
+        if not manager.get_servers():
+            # 没有成功连接的 server（多半是没装），不注册工具
+            return
+
+        self._lsp_manager = manager
+
+        # 注册 5 个统一的 LSP 工具（模型可按需调用；由 manager 按扩展名路由）
+        lsp_tools = [
+            ("lsp__diagnostics", "获取文件的真实语言诊断（错误/警告/提示），来自配置的语言服务器",
+             {"file": {"type": "string", "description": "文件路径"}},
+             "diagnostics"),
+            ("lsp__hover", "获取指定位置的符号悬停信息（类型/文档）",
+             {"file": {"type": "string", "description": "文件路径"},
+              "line": {"type": "integer", "description": "行号（0 基）"},
+              "character": {"type": "integer", "description": "列号（0 基）"}},
+             "hover"),
+            ("lsp__definition", "跳转到符号的定义位置",
+             {"file": {"type": "string", "description": "文件路径"},
+              "line": {"type": "integer", "description": "行号（0 基）"},
+              "character": {"type": "integer", "description": "列号（0 基）"}},
+             "definition"),
+            ("lsp__references", "查找符号的所有引用位置",
+             {"file": {"type": "string", "description": "文件路径"},
+              "line": {"type": "integer", "description": "行号（0 基）"},
+              "character": {"type": "integer", "description": "列号（0 基）"}},
+             "references"),
+            ("lsp__completion", "获取指定位置的代码补全项",
+             {"file": {"type": "string", "description": "文件路径"},
+              "line": {"type": "integer", "description": "行号（0 基）"},
+              "character": {"type": "integer", "description": "列号（0 基）"}},
+             "completion"),
+        ]
+        for tool_name, desc, props, method in lsp_tools:
+            mcp_def = {
+                "name": tool_name,
+                "description": desc,
+                "inputSchema": {
+                    "type": "object",
+                    "properties": props,
+                    "required": list(props.keys()),
+                },
+            }
+            handler = self._make_lsp_handler(method)
+            self.register_tool(
+                tool_name, handler,
+                description=desc,
+                protocol="lsp",
+                mcp_definition=mcp_def,
+                instance=self._lsp_manager,
+            )
+
+        n_servers = len(manager.get_servers())
+        print(f"{Fore.GREEN}已桥接 LSP 智能感知{Style.RESET_ALL}"
+              f"{Fore.BLACK}{Style.BRIGHT} (来自 {n_servers} 个语言服务器){Style.RESET_ALL}")
+
+    def _make_lsp_handler(self, method: str):
+        """为某个 LSP 能力生成 handler：解析参数 → 调用 manager → ToolResult"""
+        def handler(args):
+            try:
+                if isinstance(args, dict):
+                    params = args
+                elif isinstance(args, str) and args.strip().startswith("{"):
+                    params = json.loads(args)
+                else:
+                    params = {"args": args}
+            except (json.JSONDecodeError, TypeError):
+                params = {"args": args}
+
+            file_path = params.get("file") or params.get("path") or ""
+            line = int(params.get("line", 0) or 0)
+            character = int(params.get("character", 0) or 0)
+            if not file_path:
+                return ToolResult.fail("缺少 file 参数", ErrorCode.INVALID_PARAM,
+                                       tool_name=f"lsp__{method}")
+
+            try:
+                if method == "diagnostics":
+                    result_text = self._lsp_manager.diagnostics(file_path)
+                elif method == "hover":
+                    result_text = self._lsp_manager.hover(file_path, line, character)
+                elif method == "definition":
+                    result_text = self._lsp_manager.definition(file_path, line, character)
+                elif method == "references":
+                    result_text = self._lsp_manager.references(file_path, line, character)
+                elif method == "completion":
+                    result_text = self._lsp_manager.completion(file_path, line, character)
+                else:
+                    return ToolResult.fail(f"未知 LSP 方法: {method}", ErrorCode.NOT_IMPLEMENTED,
+                                           tool_name=f"lsp__{method}")
+                return ToolResult.ok(result_text, tool_name=f"lsp__{method}")
+            except Exception as e:
+                return ToolResult.fail(str(e), ErrorCode.EXEC_FAILED,
+                                       tool_name=f"lsp__{method}")
+        return handler
+
+    def get_lsp_context(self) -> str:
+        """返回 LSP 诊断汇总（供系统提示自动注入模型上下文）"""
+        if self._lsp_manager is None:
+            return ""
+        try:
+            return self._lsp_manager.get_context()
+        except Exception:
+            return ""
+
     # ==================== 公共 API ====================
 
     @property

websocket_server.py

@@ -24,6 +24,17 @@ import os
 from datetime import datetime
 from typing import Optional, Dict, List
 
+try:
+    from xcli_core.tool_result import normalize_tool_text
+except Exception:
+    def normalize_tool_text(result, default="无结果"):
+        """兜底归一化：ToolResult / dict / str 统一成字符串"""
+        if result is None:
+            return default
+        if isinstance(result, dict):
+            return str(result.get("result", default))
+        return str(result)
+
 # 全局变量
 websocket_server = None  # 当前活动的服务器实例（供外部访问）
 proxy_client = None
@@ -583,7 +594,7 @@ class LocalServer:
                                 await self._send_tool_status(websocket, tool_name, tool_args, "calling")
                             
                             result = self.cli.process_tool_call(tool_data)
-                            result_text = result.get('result', '无结果')
+                            result_text = normalize_tool_text(result)
                             results.append(result_text)
                             
                             if websocket:
@@ -957,7 +968,7 @@ class ProxyClient:
                         results = []
                         for tool_data in tool_calls:
                             result = self.cli.process_tool_call(tool_data)
-                            results.append(result.get('result', '无结果'))
+                            results.append(normalize_tool_text(result))
                         
                         self.cli.shared_conversation_history.append({
                             "role": "assistant",

xcli_core/agent_manager.py

@@ -0,0 +1,95 @@
+"""Agents 体系化 - opencode 式 markdown 定义 + @委派
+
+设计：
+- agents/ 目录下的 *.md 文件即 agent 定义，frontmatter 声明元数据：
+    ---
+    name: coder
+    description: 代码实现专家
+    tools: code_editor, git_tools, file_manager
+    model:
+    ---
+    正文即该 agent 的系统提示词。
+- AgentManager 负责加载、匹配、委派。
+- 委派执行通过注入的 runner 回调完成，便于测试（runner 不触真实引擎）。
+"""
+import os
+import re
+from dataclasses import dataclass, field
+from typing import Callable, Dict, List, Optional
+
+_FRONTMATTER_RE = re.compile(r"^---\s*\n(.*?)\n---\s*\n?(.*)$", re.DOTALL)
+
+
+@dataclass
+class AgentDef:
+    name: str
+    description: str = ""
+    tools: List[str] = field(default_factory=list)
+    model: Optional[str] = None
+    system_prompt: str = ""
+
+
+class AgentManager:
+    def __init__(self):
+        self._agents: Dict[str, AgentDef] = {}
+
+    # ── 加载 ──
+    def load_dir(self, directory: str) -> int:
+        """扫描目录下的 *.md，返回加载数量。目录不存在/为空则忽略。"""
+        if not directory or not os.path.isdir(directory):
+            return 0
+        count = 0
+        for fn in sorted(os.listdir(directory)):
+            if not fn.endswith(".md"):
+                continue
+            path = os.path.join(directory, fn)
+            try:
+                with open(path, "r", encoding="utf-8") as f:
+                    text = f.read()
+            except Exception:
+                continue
+            d = self._parse(text, name=os.path.splitext(fn)[0])
+            if d:
+                self._agents[d.name] = d
+                count += 1
+        return count
+
+    @staticmethod
+    def _parse(text: str, name: str) -> Optional[AgentDef]:
+        meta: Dict[str, str] = {}
+        body = text
+        m = _FRONTMATTER_RE.match(text)
+        if m:
+            meta_block = m.group(1)
+            body = m.group(2)
+            for line in meta_block.splitlines():
+                if ":" in line:
+                    k, v = line.split(":", 1)
+                    meta[k.strip()] = v.strip()
+        raw_tools = meta.get("tools", "")
+        tools = [t.strip() for t in raw_tools.split(",") if t.strip()]
+        return AgentDef(
+            name=meta.get("name", name),
+            description=meta.get("description", ""),
+            tools=tools,
+            model=meta.get("model") or None,
+            system_prompt=body.strip(),
+        )
+
+    # ── 查询 ──
+    def get(self, name: str) -> Optional[AgentDef]:
+        return self._agents.get(name)
+
+    def names(self) -> List[str]:
+        return list(self._agents.keys())
+
+    def list(self) -> List[AgentDef]:
+        return list(self._agents.values())
+
+    # ── 委派 ──
+    def dispatch(self, name: str, task: str, runner: Callable[[str, str, List[str], Optional[str]], str]) -> str:
+        """runner(system_prompt, task, tools, model) -> result 文本。"""
+        d = self.get(name)
+        if not d:
+            raise KeyError(f"agent not found: {name}")
+        return runner(d.system_prompt, task, d.tools, d.model)

xcli_core/cli_base.py

@@ -10,6 +10,7 @@ from .constants import (
     UNIFIED_TOOL_MANAGER_AVAILABLE, TEXTUAL_AVAILABLE,
 )
 from .config import get_system_config, set_system_config, logger
+from .verbose import vprint
 from .plugin_manager import LiuginManager
 from .process_protection import get_protection, enable_process_protection
 
@@ -70,16 +71,23 @@ class BaseAICLI:
         # 设置默认引擎，优先使用配置文件指定的引擎
         if default_engine_name in self.engines:
             self.current_engine = self.engines[default_engine_name]
-            print(f"{Fore.GREEN}使用配置的默认AI引擎: {default_engine_name}{Style.RESET_ALL}")
+            vprint(f"{Fore.GREEN}使用配置的默认AI引擎: {default_engine_name}{Style.RESET_ALL}")
         elif self.engines:
             first_engine_name = next(iter(self.engines))
             self.current_engine = self.engines[first_engine_name]
-            print(f"{Fore.GREEN}使用默认AI引擎: {first_engine_name}{Style.RESET_ALL}")
+            vprint(f"{Fore.GREEN}使用默认AI引擎: {first_engine_name}{Style.RESET_ALL}")
         else:
             print(f"{Fore.RED}警告: 没有可用的AI引擎，请检查ai_engines目录{Style.RESET_ALL}")
             self.current_engine = None
         if self.current_engine:
             self.current_engine.cli = self  # 设置引用以便访问插件
+            # 记录当前模型（openai 引擎注册表名，其它引擎取 model 值）
+            try:
+                self.current_model = self.current_engine.current_model_name()
+            except Exception:
+                self.current_model = getattr(self.current_engine, 'model', None)
+        else:
+            self.current_model = None
 
         # 初始化安全层
         from .safety import get_safety
@@ -89,6 +97,21 @@ class BaseAICLI:
         self.chat_history_dir = os.path.join(project_dir, "chat_history")
         if not os.path.exists(self.chat_history_dir):
             os.makedirs(self.chat_history_dir)
+        # 初始化会话管理器（opencode 式 /resume 持久化）
+        self.init_session_manager()
+        # Plan 模式状态（只读规划 → 审批 → 执行）
+        self.plan_mode = False
+        self.current_plan = ""
+        # Agents 管理器（opencode 式 @委派，加载 agents/ 目录下的 *.md）
+        self.agent_manager = None
+        try:
+            from .agent_manager import AgentManager
+            self.agent_manager = AgentManager()
+            agents_dir = os.path.join(project_dir, "agents")
+            os.makedirs(agents_dir, exist_ok=True)
+            self.agent_manager.load_dir(agents_dir)
+        except Exception:
+            self.agent_manager = None
 
         # 获取 skills 目录路径
         skills_dir = os.path.join(project_dir, "skills")
@@ -153,7 +176,7 @@ class BaseAICLI:
 
     def load_ai_engines(self):
         """动态加载AI引擎插件"""
-        print(f"{Fore.GREEN}正在加载AI引擎插件...{Style.RESET_ALL}")
+        vprint(f"{Fore.GREEN}正在加载AI引擎插件...{Style.RESET_ALL}")
         ai_engines_dir = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "ai_engines")
         if os.path.exists(ai_engines_dir):
             for filename in os.listdir(ai_engines_dir):
@@ -171,7 +194,7 @@ class BaseAICLI:
                             engine_key = getattr(engine_instance, 'name', engine_name)
                             if engine_key not in self.engines:
                                 self.engines[engine_key] = engine_instance
-                                print(f"{Fore.GREEN}已加载AI引擎: {engine_key} ({class_name}){Style.RESET_ALL}")
+                                vprint(f"{Fore.GREEN}已加载AI引擎: {engine_key} ({class_name}){Style.RESET_ALL}")
                             else:
                                 print(f"{Fore.YELLOW}AI引擎 {engine_key} 已存在，跳过加载{Style.RESET_ALL}")
                         else:
@@ -185,7 +208,7 @@ class BaseAICLI:
                     except Exception as e:
                         logger.error(f"加载AI引擎插件时发生未知错误 {filename}: {e}")
                         print(f"{Fore.RED}加载AI引擎插件失败 {filename}: {e}{Style.RESET_ALL}")
-            print(f"{Fore.GREEN}总共加载了 {len(self.engines)} 个AI引擎{Style.RESET_ALL}")
+            vprint(f"{Fore.GREEN}总共加载了 {len(self.engines)} 个AI引擎{Style.RESET_ALL}")
 
     def _find_engine_class(self, module, engine_name):
         """查找AI引擎类，使用动态方式避免硬编码类名"""
@@ -213,7 +236,7 @@ class BaseAICLI:
 
     def load_liugins(self, liugins_dir=None, skills_dir=None):
         """加载liugin和技能"""
-        print(f"{Fore.GREEN}正在加载工具...{Style.RESET_ALL}")
+        vprint(f"{Fore.GREEN}正在加载工具...{Style.RESET_ALL}")
 
         if self._use_unified_manager:
             self.liugin_manager.initialize(liugins_dir or "plugins", skills_dir or "skills")
@@ -242,19 +265,19 @@ class BaseAICLI:
                                 tool_info = plugin_instance.get_tool_info()
                                 tool_info['handler'] = plugin_instance.handle
                                 self.liugin_manager.tools.append(tool_info)
-                                print(f"{Fore.GREEN}已加载liugin: {liugin_name}{Style.RESET_ALL}")
+                                vprint(f"{Fore.GREEN}已加载liugin: {liugin_name}{Style.RESET_ALL}")
                         except Exception as e:
                             print(f"{Fore.RED}加载liugin失败 {filename}: {e}{Style.RESET_ALL}")
-            print(f"{Fore.GREEN}总共加载了 {len(self.liugin_manager.tools)} 个工具插件{Style.RESET_ALL}")
+            vprint(f"{Fore.GREEN}总共加载了 {len(self.liugin_manager.tools)} 个工具插件{Style.RESET_ALL}")
 
-        print(f"{Fore.GREEN}工具加载完成!{Style.RESET_ALL}")
+        vprint(f"{Fore.GREEN}工具加载完成!{Style.RESET_ALL}")
 
     # ── 命令注册 ──
 
     def register_liugin_command(self, command, handler):
         """注册插件命令"""
         self.liugin_commands[command] = handler
-        print(f"{Fore.GREEN}已注册插件命令: {command}{Style.RESET_ALL}")
+        vprint(f"{Fore.GREEN}已注册插件命令: {command}{Style.RESET_ALL}")
 
     def unregister_liugin_command(self, command):
         """注销插件命令"""
@@ -305,6 +328,124 @@ class BaseAICLI:
                 help_lines.append(f"  /engine {cmd:<10} - 执行{cmd}命令")
         return '\n'.join(help_lines) if help_lines else "  无可用引擎管理命令"
 
+    # ── 模型管理（OpenAI 引擎多模型切换）──
+
+    def handle_model_command(self, args):
+        """处理 /model 命令族。仅对 openai 引擎有效。"""
+        eng = self.current_engine
+        if getattr(eng, 'name', None) != 'openai':
+            cur = getattr(eng, 'name', '?')
+            print(f"{Fore.RED}模型管理仅适用于 openai 引擎（当前: {cur}）。"
+                  f"先 /engine switch openai{Style.RESET_ALL}")
+            return
+
+        if not args or args == 'list':
+            self._model_list(eng)
+            return
+
+        parts = args.split(' ', 1)
+        cmd = parts[0]
+        rest = parts[1].strip() if len(parts) > 1 else ''
+        if cmd == 'add':
+            self._model_add_interactive(eng)
+        elif cmd in ('rm', 'remove', 'del'):
+            if not rest:
+                print(f"{Fore.RED}用法: /model rm <名称>{Style.RESET_ALL}")
+                return
+            if eng.remove_model(rest):
+                print(f"{Fore.GREEN}已删除模型: {rest}{Style.RESET_ALL}")
+            else:
+                print(f"{Fore.RED}未找到模型: {rest}{Style.RESET_ALL}")
+        elif cmd in ('set', 'switch'):
+            self._model_switch(eng, rest)
+        else:
+            # 无子命令：/model <名称> 视为切换
+            self._model_switch(eng, cmd)
+
+    def _model_list(self, eng):
+        models, current = eng.list_registry_models()
+        if not models:
+            print(f"{Fore.YELLOW}尚未配置任何模型，用 /model add 添加{Style.RESET_ALL}")
+            return
+        print(f"{Fore.GREEN}已配置的 OpenAI 模型:{Style.RESET_ALL}")
+        for m in models:
+            mark = f"{Fore.CYAN} *{Style.RESET_ALL}" if m.get('name') == current else ""
+            print(f"  {m.get('name')}{mark}  ({m.get('model')} @ {m.get('base_url') or '无 base_url'})")
+        print(f"{Fore.CYAN}当前: {current} ｜ /model <名称> 切换 ｜ /model add 新增 ｜ "
+              f"/model rm <名称> 删除{Style.RESET_ALL}")
+
+    def _model_switch(self, eng, name):
+        if not name:
+            print(f"{Fore.RED}用法: /model <名称>  或 /model 查看列表{Style.RESET_ALL}")
+            return
+        if eng.set_registry_model(name):
+            self.current_model = name
+            print(f"{Fore.GREEN}已切换到模型: {name}{Style.RESET_ALL}")
+        else:
+            print(f"{Fore.RED}未找到模型: {name}（/model 查看可用）{Style.RESET_ALL}")
+
+    def _model_add_interactive(self, eng):
+        print(f"{Fore.CYAN}添加 OpenAI 模型（回车留空则用默认/为空）{Style.RESET_ALL}")
+        try:
+            name = input(f"{Fore.CYAN}名称(用于切换, 如 gpt4o): {Style.RESET_ALL}").strip()
+            if not name:
+                print(f"{Fore.RED}名称不能为空{Style.RESET_ALL}")
+                return
+            base_url = input(f"{Fore.CYAN}base_url(如 https://api.openai.com/v1): {Style.RESET_ALL}").strip()
+            api_key = input(f"{Fore.CYAN}api_key: {Style.RESET_ALL}").strip()
+            model = input(f"{Fore.CYAN}model(如 gpt-4o): {Style.RESET_ALL}").strip()
+            if not model:
+                print(f"{Fore.RED}model 不能为空{Style.RESET_ALL}")
+                return
+        except (EOFError, KeyboardInterrupt):
+            print(f"{Fore.YELLOW}\n已取消添加{Style.RESET_ALL}")
+            return
+        eng.add_model({"name": name, "base_url": base_url, "api_key": api_key, "model": model})
+        eng.set_registry_model(name)
+        self.current_model = name
+        print(f"{Fore.GREEN}已添加并切换到模型: {name}{Style.RESET_ALL}")
+
+    # ── Provider 抽象层（opencode 式原生多 Provider）──
+    def handle_providers_command(self, args):
+        """处理 /providers 命令：列出已注册 Provider 与配置状态。
+
+        /providers           列出全部内置 Provider 及是否已配置 key
+        /providers <名称>    显示该 Provider 的连接参数（base_url / model / 协议）
+        """
+        from xcli_core.provider_router import (
+            list_providers, provider_summary, resolve_provider, PROVIDER_PRESETS,
+        )
+        try:
+            cfg = load_config() or {}
+        except Exception:
+            cfg = {}
+        providers_cfg = cfg.get("providers", {}) if isinstance(cfg, dict) else {}
+
+        if args:
+            name = args.split()[0]
+            preset = PROVIDER_PRESETS.get(name)
+            if not preset:
+                print(f"{Fore.RED}未知 Provider: {name}（/providers 查看可用）{Style.RESET_ALL}")
+                return
+            r = resolve_provider(name, providers_cfg)
+            print(f"{Fore.GREEN}Provider: {preset['label']} ({name}){Style.RESET_ALL}")
+            print(f"  协议族: {r['protocol']}")
+            print(f"  base_url: {r['base_url']}")
+            print(f"  默认模型: {r['model']}")
+            print(f"  已配置 key: {'是' if r['api_key'] else '否'}")
+            print(f"{Fore.CYAN}在 config.json 的 providers.{name} 填入 {{api_key, base_url, model}} 即可启用；"
+                  f"OpenAI 兼容族经现有 openai 引擎直接可用。{Style.RESET_ALL}")
+            return
+
+        print(f"{Fore.GREEN}已注册 Provider（opencode 式多 Provider）:{Style.RESET_ALL}")
+        for name in list_providers():
+            s = provider_summary(name, providers_cfg)
+            if not s:
+                continue
+            mark = f"{Fore.GREEN}✓ 已配置{Style.RESET_ALL}" if s["configured"] else f"{Fore.YELLOW}· 未配置{Style.RESET_ALL}"
+            print(f"  {s['name']:<10} {s['label']:<22} [{s['protocol']:<9}] {mark}")
+        print(f"{Fore.CYAN}/providers <名称> 查看连接参数 ｜ 在 config.json 的 providers 段填入 key 启用{Style.RESET_ALL}")
+
     # ── 引擎切换 ──
 
     def get_available_engines(self):

xcli_core/cli_clawli.py

@@ -3,6 +3,7 @@ import time
 from colorama import Fore, Style
 
 from .constants import CLAWLI_SERVER_AVAILABLE, WEBSOCKET_AVAILABLE
+from .tool_result import normalize_tool_text
 
 try:
     import clawli_server
@@ -96,6 +97,7 @@ class ClawliMixin:
             tool_calls = self._extract_tool_calls(json_data)
 
             if tool_calls:
+                result_texts = []
                 for tool_data in tool_calls:
                     tool_name = tool_data.get('tool', '未知工具')
                     tool_args = tool_data.get('args', '')
@@ -104,7 +106,8 @@ class ClawliMixin:
                         clawli_server.send_tool_status(tool_name, tool_args, "calling")
 
                     result = self.process_tool_call(tool_data)
-                    result_text = result.get('result', '无结果')
+                    result_text = normalize_tool_text(result)
+                    result_texts.append(result_text)
 
                     if CLAWLI_SERVER_AVAILABLE and clawli_server:
                         clawli_server.send_tool_status(tool_name, tool_args, "success", result_text)
@@ -114,7 +117,8 @@ class ClawliMixin:
                     "content": response
                 })
 
-                tool_result_str = "\n".join([f"工具执行结果: {r}" for r in [result.get('result', '')]])
+                # 汇总全部工具结果（此前只回传了最后一个）
+                tool_result_str = "\n".join([f"工具执行结果: {r}" for r in result_texts])
                 self._process_clawli_user_message(f"{tool_result_str}\n请根据工具执行结果继续回答。")
                 return
 

xcli_core/cli_core.py

@@ -9,8 +9,8 @@ import random
 from typing import Optional
 from colorama import Fore, Style
 
-from .constants import TEXTUAL_AVAILABLE, LOVE_FILE_PATH, DEFAULT_MAX_HISTORY
+from .constants import TEXTUAL_AVAILABLE, LOVE_FILE_PATH, DEFAULT_MAX_HISTORY, VERSION
-from .config import get_system_config, logger
+from .config import get_system_config, set_system_config, logger
 from .cli_base import BaseAICLI
 from .safety import get_safety, MODE_UNRESTRICTED, MODE_NORMAL, MODE_MANUAL
 from .cli_clawli import ClawliMixin
@@ -22,6 +22,17 @@ from .plugin_market import PluginMarketMixin
 from .notification import notify_task_complete, get_notification_manager
 from .process_protection import get_protection_status
 
+# ── Plan 模式 system prompt 追加段（只读规划约束）──
+PLAN_MODE_SYSTEM_APPEND = """
+
+【PLAN 模式 — 只读规划】
+你当前处于 PLAN 模式。此模式下你只能进行调查、阅读、分析，不得执行任何修改。
+- 你可以使用只读工具调研代码库：code_search（搜索/结构/符号/统计）、file_manager 的 list/read/info/search、code_editor 的 read_range/find/diff、git_tools 的 status/log/diff/show/blame、tool_search（查询工具用法）、ai_search（联网搜索）。
+- 你**不能**执行写操作（编辑/创建/删除文件、运行 shell 命令、git commit/push/checkout、委派会修改的子代理等）——这些会被系统拦截并返回提示。
+- 调研完成后，输出一份清晰的、带编号步骤的「实施计划」，让用户审批。计划应包含：每步要做什么、涉及哪些文件/函数、预期结果、潜在风险。
+- 计划以如下标记开头：## 实施计划
+- 完成计划后停止，等待用户输入 /build 批准执行，或继续追问以完善计划。不要自行开始修改。"""
+
 
 class AICLI(BaseAICLI, ClawliMixin, ToolMixin, CodeExecMixin, DisplayMixin, HistoryMixin, PluginMarketMixin):
     """AI CLI主程序 - 组合所有 Mixin"""
@@ -31,7 +42,7 @@ class AICLI(BaseAICLI, ClawliMixin, ToolMixin, CodeExecMixin, DisplayMixin, Hist
     def process_thinking_response(self, response):
         """处理深度思考AI返回的内容, 分离思考和回复内容"""
         try:
-            response_data = json.loads(response)
+            response_data = self._loads_json(response)
             if isinstance(response_data, dict) and response_data.get('action') == 'use_tool':
                 return response
             if isinstance(response_data, dict):
@@ -143,7 +154,20 @@ class AICLI(BaseAICLI, ClawliMixin, ToolMixin, CodeExecMixin, DisplayMixin, Hist
 你的思考过程:
 1. 需要 git 操作 → 先查 git_tools 怎么用
 2. {{"action": "use_tool", "tool": "tool_search", "args": "git_tools"}} → 发现 smart-commit 操作
-3. {{"action": "use_tool", "tool": "git_tools", "args": "smart-commit"}} → 自动提交"""
+3. {{"action": "use_tool", "tool": "git_tools", "args": "smart-commit"}} → 自动提交        """
+
+        # 参数格式统一示例：无论用哪种工具，args 都是同一个字符串外壳，给 AI 一个速查锚点
+        unified_arg_example = """【参数格式统一示例】
+所有工具共用同一套 JSON 外壳，`args` 永远是【字符串】（不是 JSON 对象）：
+{"action":"use_tool","tool":"<工具名>","args":"<操作> <参数>"}
+
+高频工具的 args 写法（照抄即可）：
+- code_editor : "edit 文件 旧<<<>>>新"  |  "read_range 文件 起 止"  |  "create 文件 内容"
+- code_search : "find . 关键词 *.py"  |  "structure ."
+- git_tools   : "status"  |  "commit 说明文字"
+- cmd_executor: "dir"  |  "python main.py"
+说明：多行内容（如 edit 的代码片段）直接换行写在 args 里，无需转义；一次可返回多个 JSON 指令。
+调用方式可选：纯 JSON（默认）或 function calling（引擎支持时），两种都会被正确处理。"""
 
         if liugin_prompts:
             prompt = f"""{base_prompt}
@@ -152,6 +176,8 @@ class AICLI(BaseAICLI, ClawliMixin, ToolMixin, CodeExecMixin, DisplayMixin, Hist
 
 {tool_search_rule}
 
+{unified_arg_example}
+
 【工具调用格式】
 在回复中包含 JSON 指令来执行操作：
 
@@ -219,6 +245,8 @@ multi 操作支持一次修改多处：
 
 {tool_search_rule}
 
+{unified_arg_example}
+
 【工具调用格式】
 {{"action": "use_tool", "tool": "工具名", "args": "参数"}}
 
@@ -253,6 +281,20 @@ multi 操作支持一次修改多处：
             except Exception:
                 pass
 
+        # ── Plan 模式：追加只读规划约束 ──
+        if getattr(self, 'plan_mode', False):
+            prompt += PLAN_MODE_SYSTEM_APPEND
+
+        # ── LSP 诊断自动注入：让模型持续看到真实语言诊断 ──
+        liugin_manager = getattr(self, 'liugin_manager', None)
+        if liugin_manager and hasattr(liugin_manager, 'get_lsp_context'):
+            try:
+                lsp_ctx = liugin_manager.get_lsp_context()
+                if lsp_ctx:
+                    prompt += "\n\n# 当前代码诊断（来自 LSP 语言服务器）\n" + lsp_ctx
+            except Exception:
+                pass
+
         return prompt
 
     # ── JSON 解析 ──
@@ -285,11 +327,15 @@ multi 操作支持一次修改多处：
                 plugins.append(tool)
 
         if plugins:
-            prompts.append("## 内置工具（使用 tool_search 查询详细用法）：")
+            # 工具名清单（纯 JSON 模式：不发送 FC schema，模型按 action JSON 调用）。
-            for tool in plugins:
+            # 只列名字 + 由 tool_search 提供用法，避免重复曝光、省 token。
-                tool_name = tool.get('name', '')
+            names = ", ".join(t.get('name', '') for t in plugins)
-                tool_desc = tool.get('description', '')
+            prompts.append(
-                prompts.append(f"- {tool_name}: {tool_desc}")
+                "## 内置工具（纯 JSON 模式：不发送 function calling 的 tools schema，"
+                "模型按下方 JSON 格式直接调用，调用前请先用 tool_search 查询其详细用法与示例）\n"
+                f"工具名: {names}\n"
+                '调用格式: {"action": "use_tool", "tool": "工具名", "args": "参数"}'
+            )
 
         if skills:
             prompts.append("\n## 扩展技能（已加载详细指令）：")
@@ -328,6 +374,18 @@ multi 操作支持一次修改多处：
 
         return "\n".join(prompts)
 
+    def _loads_json(self, s):
+        """json.loads，但容忍未成对出现的反斜杠（Windows 路径 C:\\Users 常见）。
+
+        纯 JSON 模式下模型常把路径写成单反斜杠，直接 json.loads 会报非法转义。
+        这里先按标准解析；失败时把「未成对的反斜杠」转义后再试一次。
+        """
+        try:
+            return json.loads(s)
+        except json.JSONDecodeError:
+            fixed = re.sub(r'(?<!\\)\\(?!\\\\)', r'\\\\', s)
+            return json.loads(fixed)
+
     def _parse_mixed_response(self, response):
         """
         解析混合响应，处理同时包含文本和JSON指令的内容
@@ -340,7 +398,7 @@ multi 操作支持一次修改多处：
             json_str = response[11:]
             json_str = "{" + json_str
             try:
-                parsed = json.loads(json_str)
+                parsed = self._loads_json(json_str)
                 if isinstance(parsed, dict):
                     return "", parsed
             except json.JSONDecodeError:
@@ -348,7 +406,7 @@ multi 操作支持一次修改多处：
 
         # 方法1: 直接解析为纯JSON
         try:
-            parsed = json.loads(response)
+            parsed = self._loads_json(response)
             if isinstance(parsed, (list, dict)):
                 return "", parsed
         except json.JSONDecodeError:
@@ -365,7 +423,7 @@ multi 操作支持一次修改多处：
                 content = content.replace('"', '\\"')
                 return f'"{content}"'
             fixed_response = re.sub(r'"([^"]*(?:\n[^"]*)*)"', fix_json_string, response, flags=re.DOTALL)
-            parsed = json.loads(fixed_response)
+            parsed = self._loads_json(fixed_response)
             if isinstance(parsed, (list, dict)):
                 return "", parsed
         except:
@@ -376,7 +434,7 @@ multi 操作支持一次修改多处：
         matches = re.findall(json_pattern, response, re.DOTALL)
         if matches:
             try:
-                json_data = json.loads(matches[-1])
+                json_data = self._loads_json(matches[-1])
                 if isinstance(json_data, (dict, list)):
                     text_content = re.sub(json_pattern, "", response, flags=re.DOTALL).strip()
                     text_content = re.sub(r"\n\s*\n", "\n\n", text_content).strip()
@@ -392,7 +450,7 @@ multi 操作支持一次修改多处：
             line = line.strip()
             if line.startswith("{") and line.endswith("}"):
                 try:
-                    parsed = json.loads(line)
+                    parsed = self._loads_json(line)
                     if isinstance(parsed, dict):
                         json_objects.append(parsed)
                         continue
@@ -430,7 +488,7 @@ multi 操作支持一次修改多处：
                         if in_json and brace_count == 0:
                             json_str = response[json_start:i+1]
                             try:
-                                json_data = json.loads(json_str)
+                                json_data = self._loads_json(json_str)
                                 if isinstance(json_data, dict) and json_data.get('action') in ['use_tool', 'continue']:
                                     text_content = response[:json_start].strip()
                                     return text_content, json_data
@@ -442,7 +500,7 @@ multi 操作支持一次修改多处：
         potential_jsons = re.findall(r'\{[^{}]*(?:\{[^{}]*\}[^{}]*)*\}', response, re.DOTALL)
         for json_str in potential_jsons:
             try:
-                json_data = json.loads(json_str)
+                json_data = self._loads_json(json_str)
                 if isinstance(json_data, dict) and json_data.get('action') in ['use_tool', 'continue']:
                     json_start = response.find(json_str)
                     text_content = response[:json_start].strip()
@@ -460,12 +518,12 @@ multi 操作支持一次修改多处：
                 if last_brace_pos != -1:
                     json_part = cleaned[last_brace_pos:]
                     if json_part.startswith('['):
-                        parsed = json.loads(json_part)
+                        parsed = self._loads_json(json_part)
                         if isinstance(parsed, list):
                             text_content = cleaned[:last_brace_pos].strip()
                             return text_content, parsed
                     else:
-                        parsed = json.loads(json_part)
+                        parsed = self._loads_json(json_part)
                         if isinstance(parsed, dict):
                             text_content = cleaned[:last_brace_pos].strip()
                             return text_content, parsed
@@ -714,17 +772,34 @@ multi 操作支持一次修改多处：
             except Exception:
                 pass
 
-            # ── 自动压缩上下文 ──
+            # ── 自动压缩上下文（token 感知优先，消息数兜底）──
             try:
-                if self.memory_manager.should_compress(self.shared_conversation_history):
+                if hasattr(self, 'memory_manager') and self.memory_manager:
-                    self.shared_conversation_history = self.memory_manager.compress_context(
+                    engine = self.current_engine
-                        self.shared_conversation_history,
+                    prompt_tokens = getattr(engine, 'last_prompt_tokens', None) if engine else None
-                        engine=self.current_engine
+                    # 统一 API：所有引擎必须返回 max_input_tokens（最大输入 token 数）
-                    )
+                    ctx = None
-                    self._set_shared_conversation_history()
+                    if engine:
-                    print(f"{Fore.DIM}[上下文已压缩，保留最近 10 条]{Style.RESET_ALL}")
+                        try:
-            except Exception:
+                            ctx = engine.max_input_tokens
-                pass
+                        except Exception:
+                            ctx = None
+                    ratio = get_system_config('compress_ratio', 0.7)
+                    self.memory_manager.set_compress_ratio(ratio)
+                    if self.memory_manager.should_compress(
+                        self.shared_conversation_history, prompt_tokens, ctx, ratio
+                    ):
+                        before = len(self.shared_conversation_history)
+                        self.shared_conversation_history = self.memory_manager.compress_context(
+                            self.shared_conversation_history,
+                            engine=engine
+                        )
+                        token_info = f"，token {prompt_tokens}/{ctx}" if (prompt_tokens and ctx) else ""
+                        self._output(
+                            f"{Fore.CYAN}🗜 上下文已自动压缩 (原 {before} 条 → {len(self.shared_conversation_history)} 条{token_info}，阈值 {ratio}){Style.RESET_ALL}"
+                        )
+            except Exception as e:
+                logger.warning(f"自动压缩上下文失败: {e}")
 
     def _extract_task_summary(self, user_input: str) -> str:
         """提取任务摘要用于通知显示"""
@@ -744,7 +819,7 @@ multi 操作支持一次修改多处：
             return None
 
         try:
-            data = json.loads(response)
+            data = self._loads_json(response)
             if not data.get("_fc"):
                 return None
 
@@ -779,16 +854,15 @@ multi 操作支持一次修改多处：
                 # 通过插件管理器执行
                 result = self._execute_tool_by_name(tool_name, tool_args)
 
-                # 记录工具结果到历史
+                # 记录工具结果到历史（官方 OpenAI tool 消息不含 name 字段）
                 tool_result_msg = {
                     "role": "tool",
                     "tool_call_id": call_id,
-                    "name": tool_name,
+                    "content": result if result else "执行完成"
-                    "content": result[:5000] if result else "执行完成"
                 }
                 self.shared_conversation_history.append(tool_result_msg)
 
-                results.append(f"[{tool_name}]: {result[:200] if result else '完成'}")
+                results.append(f"[{tool_name}]: {result if result else '完成'}")
 
             # 返回工具结果，供下一轮 AI 处理
             return "工具执行结果:\n" + "\n".join(results)
@@ -797,7 +871,13 @@ multi 操作支持一次修改多处：
             return None
 
     def _execute_tool_by_name(self, tool_name: str, tool_args: str) -> str:
-        """根据名称执行工具（FC 调用路径）"""
+        """根据名称执行工具（FC 调用路径）
+
+        注意：插件 handler 可能返回 ToolResult / dict / str，
+        这里统一归一成 str，保证调用方能安全做切片与拼接。
+        """
+        from .tool_result import normalize_tool_text
+
         # 安全检查
         safety = get_safety()
         allowed, msg = safety.check(tool_name, tool_args)
@@ -808,25 +888,61 @@ multi 操作支持一次修改多处：
         tool = self.liugin_manager.get_tool_by_name(tool_name)
         if tool and 'handler' in tool:
             try:
-                return tool['handler'](tool_args)
+                return normalize_tool_text(tool['handler'](tool_args), default="执行完成")
             except Exception as e:
                 return f"工具执行失败: {e}"
 
         # memory 内置工具
         if tool_name == 'memory' and hasattr(self, 'memory_manager'):
-            return self.memory_manager.handle_tool(tool_args)
+            return normalize_tool_text(self.memory_manager.handle_tool(tool_args), default="执行完成")
 
         # scheduler 内置工具
         if tool_name == 'scheduler':
             for t in self.liugin_manager.tools:
                 if t.get('name') == 'scheduler' and 'handler' in t:
                     try:
-                        return t['handler'](tool_args)
+                        return normalize_tool_text(t['handler'](tool_args), default="执行完成")
                     except Exception as e:
                         return f"工具执行失败: {e}"
 
         return f"未找到工具: {tool_name}"
 
+    def _build_fc_tools(self, current_input):
+        """构建本轮下发给引擎的 FC 工具子集，让引擎自行决定用 FC 结构化调用还是回退纯 JSON。
+
+        引擎不支持 FC（探测缓存未通过）时返回 None —— 引擎自动降级为纯 JSON 模式。
+        仅对带 `_fc_supported()` 的引擎下发 tools（Manual 等不支持的引擎不传 tools，
+        避免关键字参数错误）。
+        """
+        engine = self.current_engine
+        if not engine or not hasattr(engine, '_fc_supported'):
+            return None
+        try:
+            if not engine._fc_supported():
+                return None
+        except Exception:
+            return None
+        # 工具结果续轮：模型只需决定「继续/收尾」，不需要 FC schema，省下每轮重发的 token。
+        # 若模型仍想调工具，走纯 JSON action（无需 schema 即可），行为等价、成本更低。
+        if current_input and str(current_input).lstrip().startswith("工具执行结果"):
+            return None
+        try:
+            from .fc_tools import mcp_to_openai_tools, prune_tools
+            all_tools = mcp_to_openai_tools(self.liugin_manager)
+            if not all_tools:
+                return None
+            return prune_tools(all_tools, current_input)
+        except Exception:
+            return None
+
+    def _call_engine_with_tools(self, current_input, system_prompt, fc_tools):
+        """调用当前引擎；fc_tools 非空时下发 FC 工具定义（tool_choice=auto，模型自选）。"""
+        if fc_tools:
+            return self.current_engine.generate_response(
+                current_input, system_prompt=system_prompt, tools=fc_tools)
+        return self.current_engine.generate_response(
+            current_input, system_prompt=system_prompt)
+
     def _generate_response_with_animation(self, current_input, liugin_prompts=None):
         """生成AI响应并显示等待动画 — 支持 ESC 跨平台取消"""
         # TUI 模式：不做动画
@@ -834,14 +950,10 @@ multi 操作支持一次修改多处：
             system_prompt = self._build_system_prompt(liugin_prompts)
             if not self.current_engine:
                 return "错误: 当前没有可用的AI引擎，请检查ai_engines目录中的引擎插件"
-            # TUI 模式也支持 FC
+            # 混合模式：下发 FC 工具定义，tool_choice=auto —— 模型可自行选择
-            fc_tools = None
+            # 用 FC 结构化调用，或回退到纯 JSON action（两种都会被正确处理）。
-            try:
+            fc_tools = self._build_fc_tools(current_input)
-                from .fc_tools import mcp_to_openai_tools
+            response = self._call_engine_with_tools(current_input, system_prompt, fc_tools)
-                fc_tools = mcp_to_openai_tools(self.liugin_manager)
-            except Exception:
-                pass
-            response = self.current_engine.generate_response(current_input, system_prompt=system_prompt, tools=fc_tools)
             return response
 
         # CLI 模式：带动画和跨平台 ESC 取消
@@ -851,12 +963,18 @@ multi 操作支持一次修改多处：
         response_ready = threading.Event()
         response_result = [None]  # 用列表存储，方便线程内修改
 
+        # 是否真实交互终端：非 TTY（管道 / 日志镜像 / 聊天前端捕获 stdout）时，
+        # \r 无法原地覆盖，逐帧动画会污染输出流，必须禁用。
+        is_tty = sys.stdout.isatty()
+
         love_sentences = []
         love_file_path = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), LOVE_FILE_PATH)
         if os.path.exists(love_file_path):
             try:
                 with open(love_file_path, 'r', encoding='utf-8') as f:
-                    love_sentences = [line.strip() for line in f.readlines() if line.strip()]
+                    love_sentences = [line.strip() for line in f.readlines()
+                                      if line.strip()
+                                      and not line.strip().lower().startswith(('http://', 'https://'))]
             except Exception:
                 pass
         if not love_sentences:
@@ -903,6 +1021,13 @@ multi 操作支持一次修改多处：
 
         # ── 动画线程 ──
         def show_animation():
+            # 非 TTY：不跑逐帧 \r 动画（避免残帧进入捕获流），但保留一句随机情话保人味
+            if not is_tty:
+                try:
+                    print(f"💭 {random.choice(love_sentences)}", flush=True)
+                except Exception:
+                    pass
+                return
             last_change_time = time.time()
             current_sentence = random.choice(love_sentences)
             while animation_running.is_set():
@@ -923,19 +1048,9 @@ multi 操作支持一次修改多处：
                 if not self.current_engine:
                     response_result[0] = "错误: 当前没有可用的AI引擎，请检查ai_engines目录中的引擎插件"
                 else:
-                    # ── 收集 FC 工具定义 ──
+                    # 混合模式：下发 FC 工具（tool_choice=auto），模型自选 FC 或纯 JSON
-                    fc_tools = None
+                    fc_tools = self._build_fc_tools(current_input)
-                    try:
+                    response_result[0] = self._call_engine_with_tools(current_input, system_prompt, fc_tools)
-                        from .fc_tools import mcp_to_openai_tools
-                        fc_tools = mcp_to_openai_tools(self.liugin_manager)
-                        if fc_tools:
-                            fc_tools = fc_tools  # 已经是列表
-                    except Exception:
-                        fc_tools = None
-
-                    response_result[0] = self.current_engine.generate_response(
-                        current_input, system_prompt=system_prompt, tools=fc_tools
-                    )
             except Exception as e:
                 response_result[0] = f"AI 响应生成失败: {e}"
             finally:
@@ -956,7 +1071,8 @@ multi 操作支持一次修改多处：
             # ESC 被按下 → 取消
             if esc_pressed.is_set():
                 # 清除动画行
-                print("\r" + " " * 80 + "\r", end="", flush=True)
+                if is_tty:
+                    print("\r" + " " * 80 + "\r", end="", flush=True)
                 print(f"\n{Fore.YELLOW}⚡ AI 请求已取消 (ESC){Style.RESET_ALL}")
                 return "[AI请求已取消]"
 
@@ -964,7 +1080,8 @@ multi 操作支持一次修改多处：
             response = response_result[0]
             if response:
                 # 清除动画行后再输出
-                print("\r" + " " * 80 + "\r", end="", flush=True)
+                if is_tty:
+                    print("\r" + " " * 80 + "\r", end="", flush=True)
                 self._typeprint(response, Fore.CYAN)
 
             if response and self.shared_conversation_history is not None:
@@ -980,79 +1097,187 @@ multi 操作支持一次修改多处：
             for t in threads:
                 t.join(timeout=1)
 
+    # ── Plan 模式（只读规划 → 审批 → 执行）──
+
+    def _sync_current_plan(self):
+        """Plan 模式下，把对话历史中最后一条 assistant 内容记为当前计划。"""
+        if not getattr(self, 'plan_mode', False):
+            return
+        for msg in reversed(self.shared_conversation_history):
+            if isinstance(msg, dict) and msg.get("role") == "assistant" and msg.get("content"):
+                self.current_plan = msg["content"]
+                break
+
+    def handle_plan_command(self, args):
+        """处理 /plan 命令。进入 PLAN 模式（只读规划）。
+
+        /plan            进入模式，等待描述任务
+        /plan <任务>     进入模式并立即只读调研、生成实施计划
+        /plan off|exit   退出模式（计划不执行）
+        """
+        args = (args or "").strip()
+        if args in ('off', 'exit', '退出', 'cancel'):
+            self.plan_mode = False
+            print(f"{Fore.YELLOW}已退出 PLAN 模式（计划未执行）{Style.RESET_ALL}")
+            return
+
+        self.plan_mode = True
+        if args:
+            print(f"{Fore.CYAN}已进入 PLAN 模式，正在只读调研并生成实施计划...{Style.RESET_ALL}")
+            self.process_conversation(args)
+            self._autosave_session()
+            self._sync_current_plan()
+        else:
+            print(f"{Fore.CYAN}已进入 PLAN 模式。{Style.RESET_ALL}")
+            print(f"  描述你的任务，AI 将只做只读调研并给出实施计划；"
+                  f"完成后用 {Fore.WHITE}/build{Style.RESET_ALL} 批准执行，或 "
+                  f"{Fore.WHITE}/plan off{Style.RESET_ALL} 取消。{Style.RESET_ALL}")
+
+    def handle_agent_command(self, arg: str):
+        """@<agent名> <任务> 委派给指定 agent 独立执行。"""
+        mgr = getattr(self, 'agent_manager', None)
+        if mgr is None:
+            print(f"{Fore.RED}Agents 管理器未初始化{Style.RESET_ALL}")
+            return
+        if not arg:
+            names = mgr.names()
+            if names:
+                print(f"{Fore.CYAN}可用 Agents: {', '.join(names)}{Style.RESET_ALL}")
+            else:
+                print(f"{Fore.YELLOW}暂无 agent 定义（在 agents/ 目录放置 *.md）{Style.RESET_ALL}")
+            return
+        parts = arg.split(maxsplit=1)
+        name = parts[0]
+        task = parts[1] if len(parts) > 1 else ""
+        agent = mgr.get(name)
+        if not agent:
+            print(f"{Fore.RED}未找到 agent: {name}{Style.RESET_ALL}")
+            return
+        print(f"{Fore.CYAN}委派给 agent「{agent.name}」: {task}{Style.RESET_ALL}")
+        try:
+            result = mgr.dispatch(name, task, self._agent_runner)
+        except Exception as e:
+            print(f"{Fore.RED}agent 执行失败: {e}{Style.RESET_ALL}")
+            return
+        print(f"{Fore.GREEN}{result}{Style.RESET_ALL}")
+
+    def _agent_runner(self, system_prompt, task, tools, model):
+        """agent 执行器：用当前引擎跑一轮独立上下文（不污染主会话）。"""
+        engine = getattr(self, 'current_engine', None) or getattr(self, 'engine', None)
+        if engine is None:
+            return "错误：当前无可用的 AI 引擎"
+        prompt = task
+        if system_prompt:
+            prompt = f"{system_prompt}\n\n---\n用户任务：\n{task}"
+        return engine.generate_response(prompt, system_prompt=system_prompt)
+
+    def handle_build_command(self):
+        """处理 /build 命令。从 PLAN 模式进入执行（用已批准计划驱动）。"""
+        plan = getattr(self, 'current_plan', '') or ""
+        if not getattr(self, 'plan_mode', False) and not plan.strip():
+            print(f"{Fore.YELLOW}当前不在 PLAN 模式，且无已生成的计划{Style.RESET_ALL}")
+            return
+
+        self.plan_mode = False
+        if not plan.strip():
+            print(f"{Fore.YELLOW}尚未生成实施计划，先 /plan 描述任务让 AI 调研{Style.RESET_ALL}")
+            return
+
+        print(f"{Fore.GREEN}已批准计划，开始执行...{Style.RESET_ALL}")
+        instruction = ("【已批准的实施计划，请现在严格按照以下步骤执行，利用可用工具完成每一步；"
+                       "遇到与计划不符的情况先说明再继续】\n\n" + plan)
+        self.process_conversation(instruction)
+        self._autosave_session()
+
     # ── CLI 主循环 ──
 
     def run_tui(self):
-        """运行 TUI 模式"""
+        """运行 TUI 模式（v8.0 默认真·全屏 textual TUI；textual 不可用时回退 rich 面板）"""
-        if not TEXTUAL_AVAILABLE:
+        if TEXTUAL_AVAILABLE:
-            print(f"{Fore.RED}TUI 模式不可用：Textual 库未安装{Style.RESET_ALL}")
+            from .tui import XiaoliTUI
-            print(f"{Fore.YELLOW}请运行: pip install textual rich{Style.RESET_ALL}")
+            app = XiaoliTUI(self)
+            app.run()
             return
-        from .tui import XiaoliTUI
+        # 回退 rich 面板 TUI（textual 不可用时）
-        app = XiaoliTUI(self)
+        try:
-        app.run()
+            from .rich_tui import RichTUI
+            from xcli_core.constants import VERSION as _V
+            tui = RichTUI(self)
+            tui.version = _V
+            tui.set_status(
+                engine=getattr(self.current_engine, 'name', '') if self.current_engine else '',
+                model=getattr(self.current_engine, 'model', '') if self.current_engine else '',
+                mode='TUI',
+            )
+            self.tui_output_callback = tui.push
+            tui.run()
+            self.tui_output_callback = None
+        except ImportError:
+            print(f"{Fore.RED}TUI 不可用: 请安装 textual (pip install textual){Style.RESET_ALL}")
 
     def run(self):
         """运行 CLI"""
         import uuid
         self.user_id = str(uuid.getnode())
-        print(f"{Fore.GREEN}小狸 Pro-CLI 已启动!{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}用户ID: {self.user_id}{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/help' 查看帮助信息{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/quit' 退出程序{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/tui' 切换到 TUI 模式{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/engine list' 查看可用AI引擎{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/engine switch <引擎名>' 切换AI引擎{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/safe' 切换安全模式 (普通→人工→无限制){Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/notify' 切换任务完成通知 (开/关){Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/scheduler' 或 '/remind' 管理定时任务{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/memory' 管理记忆系统 (日记/搜索/聊天记录){Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/file.read <文件名> [行数]' 直接读取文件内容{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '@文件路径' 自动读取文件内容并发送给AI{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '@图片路径' 自动分析图片并发送描述给AI{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/image engines' 查看图像识别引擎{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/chat save <名称>' 保存当前聊天记录{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/chat list' 查看所有已保存的聊天记录{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/chat open <名称>' 加载聊天记录{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/remote' 查看远程连接帮助{Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/plugin' 管理插件市场 (安装/卸载/搜索){Style.RESET_ALL}")
-        print(f"{Fore.GREEN}输入 '/protect' 查看进程保护状态{Style.RESET_ALL}")
         current_engine_name = getattr(self.current_engine, 'name', '未设置') if self.current_engine else '未设置'
-        print(f"{Fore.GREEN}当前使用的AI引擎: {current_engine_name}{Style.RESET_ALL}")
+        model_name = getattr(self.current_engine, 'model', '') if self.current_engine else ''
+        engine_desc = f"{current_engine_name} ({model_name})" if model_name else current_engine_name
+
+        print(f"{Fore.GREEN}小狸 Pro-CLI v{VERSION} 已启动!{Style.RESET_ALL} "
+              f"{Fore.WHITE}引擎: {engine_desc}{Style.RESET_ALL}")
+
+        # 一行速览，完整命令清单收进 /help
+        print(f"{Fore.CYAN}/help 全部命令 · /quit 退出 · /tui 图形界面 · "
+              f"/model 换模型 · /resume 恢复会话 · /plan 规划模式 · @文件路径 读文件给 AI{Style.RESET_ALL}")
+
+        # 历史会话提示（opencode 式 /resume）
+        try:
+            sessions = self.session_manager.list_sessions()
+            if sessions:
+                latest = sessions[0]
+                print(f"{Fore.YELLOW}有 {len(sessions)} 个历史会话，/resume 可恢复"
+                      f"（最近: {latest.id} · {latest.title}）{Style.RESET_ALL}")
+        except Exception:
+            pass
+
+        extras = []
         loaded_engines = list(self.engines.keys())
         if len(loaded_engines) > 1:
-            print(f"{Fore.GREEN}已加载的AI引擎: {', '.join(loaded_engines)}{Style.RESET_ALL}")
+            extras.append(f"可切换引擎: {', '.join(loaded_engines)}")
-        print(f"{Fore.GREEN}{'-' * 50}{Style.RESET_ALL}")
+        # 进程保护：只在有未生效项时才提示，全绿就闭嘴
-
+        if getattr(self, 'protection_results', None):
-        # 显示进程保护状态
-        if hasattr(self, 'protection_results') and self.protection_results:
             r = self.protection_results
-            si_icon = '' if r.get('single_instance') else ''
+            failed = [n for k, n in (('single_instance', '单实例'), ('priority', '优先级'),
-            prio_icon = '' if r.get('priority') else ''
+                                     ('ppl_protection', 'PPL保护'), ('watchdog', '看门狗'))
-            ppl_icon = '' if r.get('ppl_protection') else ''
+                      if not r.get(k)]
-            wd_icon = '' if r.get('watchdog') else ''
+            if failed:
-            print(f"{Fore.CYAN}  进程保护: {si_icon}单实例 {prio_icon}优先级 {ppl_icon}PPL保护 {wd_icon}看门狗{Style.RESET_ALL}")
+                extras.append(f"进程保护未生效: {', '.join(failed)}")
+        if extras:
+            print(f"{Fore.BLACK}{Style.BRIGHT}{' · '.join(extras)}{Style.RESET_ALL}")
 
-        # 询问 diff 弹窗模式
+        print(f"{Fore.BLACK}{Style.BRIGHT}用户ID: {self.user_id}{Style.RESET_ALL}")
+
+        # 恢复 diff 显示模式（首次运行才询问）
         self._ask_diff_popup_mode()
 
         self._run_cli_loop()
 
-    def _ask_diff_popup_mode(self):
+    def _get_code_editor_plugin(self):
-        """启动时询问 diff 显示模式"""
+        """拿到 code_editor 插件实例（找不到返回 None）"""
-        # 找到 code_editor 插件
-        code_editor = None
         for tool in self.liugin_manager.tools:
             if tool.get('name') == 'code_editor':
-                code_editor = tool.get('handler')
+                return getattr(tool.get('handler'), '__self__', None)
-                break
+        return None
 
-        if code_editor is None:
+    def _ask_diff_popup_mode(self):
+        """设置 diff 显示模式：已保存过就静默沿用，只有首次运行才交互询问"""
+        plugin_instance = self._get_code_editor_plugin()
+        if plugin_instance is None:
             return
 
-        # 找到插件实例（通过 handler 的 __self__）
+        saved = get_system_config('diff_popup_mode', None)
-        plugin_instance = getattr(code_editor, '__self__', None)
+        if saved is not None:
-        if plugin_instance is None:
+            plugin_instance.diff_popup_mode = bool(saved)
             return
 
         print(f"\n{Fore.CYAN}  Diff 显示模式（AI 修改文件后的对比方式）:{Style.RESET_ALL}")
@@ -1060,19 +1285,23 @@ multi 操作支持一次修改多处：
         print(f"    {Fore.WHITE}2{Style.RESET_ALL} - 主终端内显示（SSH 兼容，推荐）")
 
         try:
-            choice = input(f"\n{Fore.WHITE}  请选择 [1/2]（默认 2）: {Style.RESET_ALL}").strip()
+            choice = input(f"\n{Fore.WHITE}  请选择 [1/2]（默认 2，之后可用 /diff 改）: {Style.RESET_ALL}").strip()
         except (EOFError, KeyboardInterrupt):
             choice = '2'
 
-        if choice == '1':
+        self._set_diff_popup_mode(choice == '1', plugin_instance)
-            plugin_instance.diff_popup_mode = True
-            print(f"{Fore.GREEN}  ✓ 已选择: 弹窗模式{Style.RESET_ALL}")
-        else:
-            plugin_instance.diff_popup_mode = False
-            print(f"{Fore.GREEN}  ✓ 已选择: 主终端内显示{Style.RESET_ALL}")
-
         print()
 
+    def _set_diff_popup_mode(self, popup, plugin_instance=None):
+        """写入并持久化 diff 显示模式"""
+        if plugin_instance is None:
+            plugin_instance = self._get_code_editor_plugin()
+        if plugin_instance is not None:
+            plugin_instance.diff_popup_mode = bool(popup)
+        set_system_config('diff_popup_mode', bool(popup))
+        label = '弹窗模式' if popup else '主终端内显示'
+        print(f"{Fore.GREEN}  ✓ Diff 显示模式: {label}{Style.RESET_ALL}")
+
     def _run_cli_loop(self):
         """运行 CLI 主循环"""
         while True:
@@ -1083,6 +1312,11 @@ multi 操作支持一次修改多处：
                     print(f"{Fore.GREEN}再见!{Style.RESET_ALL}")
                     break
 
+                # @agent 委派（opencode 式 Agents 体系化）
+                if user_input.startswith('@'):
+                    self.handle_agent_command(user_input[1:].strip())
+                    continue
+
                 if user_input == '/help':
                     self.show_help()
                     continue
@@ -1110,6 +1344,38 @@ multi 操作支持一次修改多处：
                         print(f"{Fore.RED}当前引擎不支持该命令或命令执行失败{Style.RESET_ALL}")
                     continue
 
+                # /model 命令族 — OpenAI 引擎多模型在线增删切换
+                if user_input == '/model' or user_input.startswith('/model '):
+                    self.handle_model_command(user_input[7:].strip())
+                    continue
+
+                # /providers — opencode 式原生多 Provider 注册与状态一览
+                if user_input == '/providers' or user_input.startswith('/providers '):
+                    self.handle_providers_command(user_input[11:].strip())
+                    continue
+
+                # /plan 进入 PLAN 模式（只读规划 → 审批 → 执行）
+                if user_input == '/plan' or user_input.startswith('/plan '):
+                    sub = user_input[6:].strip() if user_input.startswith('/plan ') else ''
+                    self.handle_plan_command(sub)
+                    continue
+                if user_input == '/build':
+                    self.handle_build_command()
+                    continue
+
+                # /resume 会话恢复（opencode 式自动持久化 + 一键恢复）
+                if user_input in ('/resume', '/sessions') or \
+                   user_input.startswith('/resume ') or user_input.startswith('/sessions ') or \
+                   user_input.startswith('/session '):
+                    if user_input.startswith('/resume'):
+                        self.handle_resume_command(user_input[7:].strip())
+                    elif user_input.startswith('/sessions'):
+                        self.handle_resume_command(user_input[9:].strip())
+                    else:  # /session <子命令>
+                        sub = user_input[8:].strip()
+                        self.handle_resume_command(sub if sub != 'new' else 'new')
+                    continue
+
                 # /manual 快捷命令 — 切换到 manual 引擎或执行 manual 子命令
                 if user_input == '/manual' or user_input.startswith('/manual '):
                     args = user_input[8:].strip() if user_input.startswith('/manual ') else ''
@@ -1215,6 +1481,30 @@ multi 操作支持一次修改多处：
                     print(f"{Fore.CYAN}已从 TUI 模式返回 CLI 模式{Style.RESET_ALL}")
                     continue
 
+                if user_input == '/compress':
+                    if hasattr(self, 'memory_manager') and self.memory_manager:
+                        engine = self.current_engine
+                        before = len(self.shared_conversation_history)
+                        self.shared_conversation_history = self.memory_manager.compress_context(
+                            self.shared_conversation_history, engine=engine)
+                        self._set_shared_conversation_history()
+                        print(f"{Fore.CYAN}已手动压缩上下文: {before} → {len(self.shared_conversation_history)} 条{Style.RESET_ALL}")
+                    else:
+                        print(f"{Fore.YELLOW}记忆系统未加载，无法压缩{Style.RESET_ALL}")
+                    continue
+
+                if user_input.startswith('/memory mode '):
+                    mode = user_input[len('/memory mode '):].strip().lower()
+                    if mode in ('companion', 'work'):
+                        try:
+                            set_system_config('memory_mode', mode)
+                        except Exception:
+                            pass
+                        print(f"{Fore.GREEN}记忆模式已设为: {mode}{Style.RESET_ALL}")
+                    else:
+                        print(f"{Fore.RED}用法: /memory mode <companion|work>{Style.RESET_ALL}")
+                    continue
+
                 if user_input.startswith('/safe'):
                     parts = user_input.split()
                     safety = get_safety()
@@ -1238,6 +1528,42 @@ multi 操作支持一次修改多处：
                     print(f"{icon} 安全模式: {mode_name}")
                     continue
 
+                if user_input.startswith('/diff'):
+                    parts = user_input.split()
+                    if self._get_code_editor_plugin() is None:
+                        print(f"{Fore.YELLOW}code_editor 插件未加载，无法设置 diff 模式{Style.RESET_ALL}")
+                        continue
+                    if len(parts) > 1:
+                        arg = parts[1].lower()
+                        if arg in ('popup', '弹窗', '1'):
+                            self._set_diff_popup_mode(True)
+                        elif arg in ('inline', '内联', '主终端', '2'):
+                            self._set_diff_popup_mode(False)
+                        else:
+                            print(f"{Fore.RED}用法: /diff [popup|inline]{Style.RESET_ALL}")
+                    else:
+                        cur = bool(get_system_config('diff_popup_mode', False))
+                        self._set_diff_popup_mode(not cur)
+                    continue
+
+                if user_input.startswith('/fc'):
+                    from .fc_tools import get_fc_cache, reset_fc_cache
+                    parts = user_input.split()
+                    if len(parts) > 1 and parts[1].lower() in ('reset', 'clear', '清除'):
+                        n = reset_fc_cache()
+                        print(f"{Fore.GREEN}已清除 {n} 条 FC 探测记录，下次请求会重新试探{Style.RESET_ALL}")
+                    else:
+                        cache = get_fc_cache()
+                        if not cache:
+                            print(f"{Fore.GREEN}FC 探测缓存为空：所有模型都按支持 function calling 处理{Style.RESET_ALL}")
+                        else:
+                            print(f"{Fore.CYAN}已探测到不支持 function calling 的模型:{Style.RESET_ALL}")
+                            for key, ok in cache.items():
+                                if ok is False:
+                                    print(f"  {Fore.YELLOW}✗{Style.RESET_ALL} {key}")
+                            print(f"{Fore.WHITE}这些请求会自动跳过 tools 字段；/fc reset 可清除重测{Style.RESET_ALL}")
+                    continue
+
                 if user_input.startswith('/notify'):
                     parts = user_input.split()
                     nm = get_notification_manager()
@@ -1297,6 +1623,10 @@ multi 操作支持一次修改多处：
                         print(f"{Fore.CYAN}{'='*50}{Style.RESET_ALL}")
 
                 self.process_conversation(user_input)
+                # 自动持久化当前会话（opencode 式 /resume 落盘）
+                self._autosave_session()
+                # Plan 模式下：把本轮最后一条 assistant 内容记为当前计划
+                self._sync_current_plan()
             except KeyboardInterrupt:
                 user_id_display = f"[用户ID: {self.user_id}]"
                 print(f"\n{Fore.GREEN}再见! {user_id_display}{Style.RESET_ALL}")
@@ -1377,12 +1707,22 @@ def main():
         import codecs
         sys.stdout = codecs.getwriter('utf-8')(sys.stdout.buffer, 'strict')
 
-    for line in art_lines:
+    def _safe_print(line):
         try:
             print(line, flush=True)
         except UnicodeEncodeError:
             print(line.encode('gbk', errors='ignore').decode('gbk'), flush=True)
-        time.sleep(0.05)
+
+    # 完整开机动画（57 行 + 逐行 0.05s）默认不放，用 --logo 显式召唤
+    if '--logo' in sys.argv or '--banner' in sys.argv:
+        for line in art_lines:
+            _safe_print(line)
+            time.sleep(0.05)
+    else:
+        _safe_print(f"\n{Fore.CYAN}  ▄▀▄  小狸 Pro-CLI v{VERSION}{Style.RESET_ALL}")
+        _safe_print(f"{Fore.CYAN}  ▀▄▀  {Style.RESET_ALL}"
+                    f"{Fore.WHITE}AI 智能编程助手{Style.RESET_ALL}"
+                    f"  {Fore.BLACK}{Style.BRIGHT}(--logo 看完整开机动画){Style.RESET_ALL}")
 
     cli = AICLI()
     cli.run()

xcli_core/cli_display.py

@@ -376,8 +376,51 @@ class DisplayMixin:
 
     # ── 帮助系统 ──
 
+    def show_quick_commands(self):
+        """常用命令速查表（启动时不再刷屏，改由 /help 展示）"""
+        rows = [
+            ('/help [插件名]', '显示帮助信息'),
+            ('/quit', '退出程序'),
+            ('/tui', '切换到 TUI 图形模式'),
+            ('/engine list', '查看可用 AI 引擎'),
+            ('/engine switch <名>', '切换 AI 引擎'),
+            ('/model', '列出已配置的 OpenAI 模型'),
+            ('/model <名称>', '切换当前 OpenAI 模型'),
+            ('/model add', '交互式新增 OpenAI 模型'),
+            ('/model rm <名称>', '删除已配置的模型'),
+            ('/fc [reset]', '查看/清除 FC 能力探测缓存'),
+            ('/safe', '切换安全模式 (普通→人工→无限制)'),
+            ('/diff', '切换 diff 显示模式 (弹窗/主终端)'),
+            ('/notify', '切换任务完成通知 (开/关)'),
+            ('/scheduler 或 /remind', '管理定时任务'),
+            ('/memory', '管理记忆系统 (日记/搜索/聊天记录)'),
+            ('/file.read <文件> [行数]', '直接读取文件内容'),
+            ('@文件路径', '自动读取文件内容并发送给 AI'),
+            ('@图片路径', '自动分析图片并发送描述给 AI'),
+            ('/image engines', '查看图像识别引擎'),
+            ('/chat save|list|open', '保存/查看/加载聊天记录'),
+            ('/resume', '列出历史会话 (自动持久化)'),
+            ('/resume <序号或ID>', '恢复指定会话继续对话'),
+            ('/resume new', '开始一个全新的会话'),
+            ('/plan [任务]', '进入 PLAN 模式：只读调研并生成实施计划'),
+            ('/build', '批准并Plan模式生成的计划，开始执行'),
+            ('/remote', '查看远程连接帮助'),
+            ('/plugin', '管理插件市场 (安装/卸载/搜索)'),
+            ('/protect', '查看进程保护状态'),
+        ]
+        width = max(len(cmd) for cmd, _ in rows)
+        print(f"{Fore.GREEN}常用命令:{Style.RESET_ALL}")
+        for cmd, desc in rows:
+            print(f"  {Fore.WHITE}{cmd.ljust(width)}{Style.RESET_ALL}  {desc}")
+        print()
+        print(f"{Fore.GREEN}启动参数:{Style.RESET_ALL}")
+        print(f"  {Fore.WHITE}{'--verbose / -v'.ljust(width)}{Style.RESET_ALL}  打印引擎/插件/技能的逐条加载明细")
+        print(f"  {Fore.WHITE}{'--logo'.ljust(width)}{Style.RESET_ALL}  播放完整开机动画")
+        print()
+
     def show_help(self):
         """显示帮助信息"""
+        self.show_quick_commands()
         help_file_path = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "HELP.md")
         if os.path.exists(help_file_path):
             try:

xcli_core/cli_history.py

@@ -2,10 +2,11 @@ import os
 import time
 import json
 from colorama import Fore, Style
+from .session import SessionManager
 
 
 class HistoryMixin:
-    """聊天记录管理相关方法"""
+    """聊天记录管理相关方法（含 opencode 式会话持久化 /resume）"""
 
     def save_chat_history(self, name):
         """保存当前聊天记录到文件"""
@@ -58,3 +59,62 @@ class HistoryMixin:
         except Exception as e:
             print(f"{Fore.RED}加载聊天记录失败: {e}{Style.RESET_ALL}")
             return False
+
+    # ── opencode 式会话（自动持久化 + /resume） ──
+
+    def init_session_manager(self):
+        """初始化会话管理器与当前会话状态。在 chat_history_dir 就绪后调用。"""
+        self.session_manager = SessionManager(self.chat_history_dir)
+        self.current_session = None  # 首轮对话时惰性创建
+
+    def _autosave_session(self):
+        """每轮对话后调用：把当前会话落盘（首轮自动建会话）。"""
+        if not getattr(self, 'shared_conversation_history', None):
+            return
+        if not hasattr(self, 'session_manager'):
+            return
+        engine = getattr(self.current_engine, 'name', None) if self.current_engine else None
+        model = getattr(self, 'current_model', None)
+        try:
+            if self.current_session is None:
+                self.current_session = self.session_manager.new_session(engine=engine, model=model)
+            self.session_manager.save(
+                self.current_session, self.shared_conversation_history,
+                engine=engine, model=model,
+            )
+        except Exception:
+            pass  # 自动保存失败不应中断对话
+
+    def handle_resume_command(self, args):
+        """/resume 命令：列出或恢复历史会话。"""
+        args = (args or "").strip()
+        if not hasattr(self, 'session_manager'):
+            self.init_session_manager()
+        if args == 'new':
+            self.current_session = None
+            print(f"{Fore.GREEN}已开始新会话（下一轮对话自动创建）{Style.RESET_ALL}")
+            return
+        sessions = self.session_manager.list_sessions()
+        if not sessions:
+            print(f"{Fore.YELLOW}没有可恢复的会话{Style.RESET_ALL}")
+            return
+        if not args:
+            print(f"{Fore.GREEN}已保存的会话 (按时间倒序):{Style.RESET_ALL}")
+            for i, s in enumerate(sessions, 1):
+                line = f"{Fore.GREEN}{i}. {s.id}{Style.RESET_ALL}  {s.title}"
+                meta = f"   {s.updated_at}  {s.engine or '-'}/{s.model or '-'}  {s.msg_count} 条"
+                print(line + f"{Fore.CYAN}{meta}{Style.RESET_ALL}")
+            print(f"{Fore.CYAN}用法: /resume <序号或ID> 恢复 · /resume new 开新会话{Style.RESET_ALL}")
+            return
+        s = self.session_manager.load(args)
+        if s is None:
+            print(f"{Fore.RED}未找到会话: {args}{Style.RESET_ALL}")
+            return
+        self.shared_conversation_history = list(s.messages)
+        self.current_session = s
+        self._set_shared_conversation_history()
+        print(f"{Fore.GREEN}已恢复会话: {s.id}{Style.RESET_ALL}")
+        print(f"{Fore.CYAN}标题: {s.title} · {s.msg_count} 条记录 · 引擎 {s.engine or '-'}/{s.model or '-'}{Style.RESET_ALL}")
+        cur = getattr(self.current_engine, 'name', None) if self.current_engine else None
+        if s.engine and cur and cur != s.engine:
+            print(f"{Fore.YELLOW}提示: 当前引擎为 {cur}，可用 /engine switch {s.engine} 切回该会话的引擎{Style.RESET_ALL}")

xcli_core/cli_tools.py

@@ -7,6 +7,7 @@ from colorama import Fore, Style
 
 from .constants import UNIFIED_TOOL_MANAGER_AVAILABLE
 from .safety import get_safety
+from .tool_result import normalize_tool_dict, normalize_tool_text
 
 
 class ToolMixin:
@@ -105,18 +106,18 @@ class ToolMixin:
                     result = self.liugin_manager.execute(tool_name, args, **arguments)
                 else:
                     result = self.liugin_manager.execute(tool_name, args)
-                return result
+                # execute() 返回 ToolResult，必须归一成 dict——
+                # 否则上层 result.get('result') 会报
+                # "'ToolResult' object has no attribute 'get'"
+                return normalize_tool_dict(result)
 
-            # 回退：直接调用 handler（旧逻辑）
+            # 回退：直接调用 handler（旧逻辑，handler 可能返回 ToolResult/str）
             for tool in self.liugin_manager.tools:
                 if tool['name'] == tool_name:
                     try:
-                        result = tool['handler'](args)
+                        return normalize_tool_dict(tool['handler'](args))
-                        if result is None:
-                            result = "无结果"
-                        return {"result": result}
                     except Exception as e:
-                        return {"result": f"工具执行错误: {e}"}
+                        return {"result": f"工具执行错误: {e}", "success": False}
 
             return {"result": f"未找到工具: {tool_name}"}
 
@@ -136,7 +137,7 @@ class ToolMixin:
                 return {"index": index, "result": "工具执行已取消"}
             try:
                 result = self.process_tool_call(tool_data)
-                tool_result = result.get('result', '无结果')
+                tool_result = normalize_tool_text(result)
                 with history_lock:
                     self.shared_conversation_history.append({
                         "role": "assistant",
@@ -204,7 +205,7 @@ class ToolMixin:
             return [r['result'] for r in results]
 
     def _execute_single_tool(self, tool_data):
-        """执行单个工具（带延迟和ESC检测）"""
+        """执行单个工具（零等待，执行前非阻塞检测 ESC 取消）"""
         tool_name = tool_data.get('tool', '未知工具')
         tool_args = tool_data.get('args', '')
         if len(tool_args) > 30:
@@ -212,27 +213,16 @@ class ToolMixin:
         else:
             tool_args_display = tool_args
 
-        # TUI 模式下不做延迟和 ESC 检测
+        # 非 TUI 模式：执行前零等待检查 ESC（不再做倒计时延迟，立即执行）
+        showed_hint = False
         if not self.tui_output_callback:
             try:
                 import msvcrt
                 canceled = False
-                spinners = ['⊶', '⊷']
+                # 排空键盘缓冲区：若在此之前按过 ESC 则取消（非阻塞，无等待）
-                spinner_index = 0
+                while msvcrt.kbhit():
-                for i in range(25):
+                    if msvcrt.getch() == b'\x1b':
-                    if msvcrt.kbhit():
+                        canceled = True
-                        key = msvcrt.getch()
-                        if key == b'\x1b':
-                            canceled = True
-                            break
-                    spinner = spinners[spinner_index % 2]
-                    spinner_index += 1
-                    loading_message = f"   正在执行: {spinner} {tool_name} {tool_args_display} (按ESC取消)"
-                    sys.stdout.write(f"\r{loading_message}")
-                    sys.stdout.flush()
-                    time.sleep(0.2)
-                sys.stdout.write("\r" + " " * 60 + "\r")
-                sys.stdout.flush()
                 if canceled:
                     print(f"{Fore.YELLOW}工具执行已取消（用户按ESC键）{Style.RESET_ALL}")
                     cancel_message = f"工具 {tool_name} 执行已被用户取消（按ESC键）。工具参数: {tool_args}"
@@ -244,8 +234,22 @@ class ToolMixin:
             except ImportError:
                 pass
 
+            try:
+                sys.stdout.write(f"\r   正在执行: {tool_name} {tool_args_display}")
+                sys.stdout.flush()
+                showed_hint = True
+            except Exception:
+                pass
+
         tool_results = self.process_tool_call(tool_data)
-        full_result = tool_results.get('result', '无结果')
+        full_result = normalize_tool_text(tool_results)
+
+        if showed_hint:
+            try:
+                sys.stdout.write("\r" + " " * 60 + "\r")
+                sys.stdout.flush()
+            except Exception:
+                pass
         from datetime import datetime
         current_time = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
         self._display_tool_result(tool_name, tool_args, full_result, current_time)

xcli_core/config.py

@@ -1,4 +1,5 @@
 import os
+import re
 import json
 import logging
 from colorama import Fore, Style
@@ -6,6 +7,39 @@ from colorama import Fore, Style
 # 配置文件路径
 CONFIG_FILE = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "config.json")
 
+# <input> 特殊语法：配置模型时可用它交互式询问用户并获取输入
+#   <input>            → 用默认提示询问
+#   <input:提示语>      → 用自定义提示语询问
+_INPUT_PATTERN = re.compile(r"<input(?::([^>]*))?>")
+
+
+def resolve_input_value(value, field_name="", default_prompt=None):
+    """解析配置值中的 <input> 特殊语法，交互式询问用户并返回其输入。
+
+    用法（写在 config.json 的字段值里）:
+        "model": "<input>"                  → 启动时提示"请输入model: "
+        "model": "<input:请输入模型名>"      → 启动时提示自定义文案
+        也可以夹在字符串中间: "prefix<input>suffix"
+
+    不含 <input> 的值原样返回；用户直接回车/中断时返回空串，由调用方兜底。
+    """
+    if not isinstance(value, str) or "<input" not in value:
+        return value
+
+    def _ask(match):
+        label = (match.group(1) or "").strip()
+        prompt = label or default_prompt or (f"请输入{field_name}" if field_name else "请输入")
+        prompt = prompt.rstrip()
+        if prompt and not prompt.endswith((":", "：", "?", "？", "!", "！", " ")):
+            prompt += ": "
+        try:
+            answer = input(f"{Fore.CYAN}{prompt}{Style.RESET_ALL}").strip()
+            return answer
+        except (EOFError, KeyboardInterrupt):
+            return ""
+
+    return _INPUT_PATTERN.sub(_ask, value)
+
 
 def load_config():
     """加载配置文件"""
@@ -42,13 +76,29 @@ def set_system_config(key, value):
     save_config(config)
 
 
-# 配置日志
+# 配置日志：文件留全量 INFO，控制台默认只报 WARNING 以上（--verbose 时放开到 INFO）
+try:
+    from .verbose import is_verbose as _is_verbose
+except ImportError:  # 被当作顶层模块单独导入时
+    try:
+        from verbose import is_verbose as _is_verbose
+    except ImportError:
+        def _is_verbose():
+            return False
+
+_file_handler = logging.FileHandler('xiaoli_cli.log', encoding='utf-8')
+_file_handler.setLevel(logging.INFO)
+_console_handler = logging.StreamHandler()
+_console_handler.setLevel(logging.INFO if _is_verbose() else logging.WARNING)
+
 logging.basicConfig(
     level=logging.INFO,
     format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
-    handlers=[
+    handlers=[_file_handler, _console_handler]
-        logging.FileHandler('xiaoli_cli.log', encoding='utf-8'),
-        logging.StreamHandler()  # 同时输出到控制台
-    ]
 )
+
+# 第三方库的心跳日志（httpx 每发一次请求就打一行）不上屏也不进日志文件
+for _noisy in ('httpx', 'httpcore', 'urllib3', 'requests', 'openai', 'PIL', 'markdown_it'):
+    logging.getLogger(_noisy).setLevel(logging.WARNING)
+
 logger = logging.getLogger(__name__)

xcli_core/constants.py

@@ -6,6 +6,10 @@ VIDEO_FRAME_DELAY = 0.04  # 40毫秒
 LOVE_FILE_PATH = "love.txt"
 DEFAULT_MAX_HISTORY = 999999  # 无限制对话历史
 
+# 版本号（v8.0 基于 v5.4.1 重新立项）
+VERSION = "8.0.3"
+VERSION_NAME = "小狸 Pro-CLI"
+
 import os
 import sys
 import importlib.util

xcli_core/fc_tools.py

@@ -2,7 +2,9 @@
 Function Calling 工具格式转换
 将 MCP 插件定义转为 OpenAI FC 格式，收集所有可用工具
 """
+import os
 import json
+import threading
 from typing import List, Dict, Optional
 
 
@@ -19,8 +21,17 @@ def mcp_to_openai_tools(plugin_manager) -> List[Dict]:
     if not plugin_manager:
         return tools
 
-    # 统一工具管理器
+    # 优先使用统一工具管理器的 MCP 定义（含外部 MCP client 桥接的工具，schema 保真）
-    if hasattr(plugin_manager, 'plugins'):
+    if hasattr(plugin_manager, 'get_mcp_tools'):
+        try:
+            for mcp_def in plugin_manager.get_mcp_tools():
+                fc_def = mcp_to_fc(mcp_def)
+                if fc_def:
+                    tools.append(fc_def)
+        except Exception:
+            pass
+    # 统一工具管理器（兜底，旧实现未暴露 get_mcp_tools 时）
+    elif hasattr(plugin_manager, 'plugins'):
         for plugin in plugin_manager.plugins:
             fc_def = _extract_fc_def(plugin)
             if fc_def:
@@ -36,7 +47,117 @@ def mcp_to_openai_tools(plugin_manager) -> List[Dict]:
     builtin_tools = _get_builtin_tools()
     tools.extend(builtin_tools)
 
-    return tools
+    # ── 去重：同名工具只保留首个 ──
+    # 来源：get_mcp_tools() 的缓存里已含 memory/scheduler/lsp 等，
+    # 下方 _get_builtin_tools() 又追加了一遍，导致同名重复（约 17% 的 FC token 是垃圾）。
+    # 按 function.name 去重，消除双重曝光，对真实工具无影响。
+    seen = set()
+    deduped = []
+    for t in tools:
+        fname = (t.get("function") or {}).get("name", "")
+        if fname in seen:
+            continue
+        seen.add(fname)
+        deduped.append(t)
+    return deduped
+
+
+# ══════════════════════════════════════════════════════════════════
+#  动态工具裁剪（方案 B：按场景分层）
+#
+#  把 40+ 的 FC 工具全量发给模型，每轮要重发数千 token schema，且
+#  模型要在大列表里选对工具，决策面大、易选错、成本高。
+#  这里按「当前用户输入」的信号，只下发本轮相关的工具子集：
+#    - ALWAYS_KEEP：始终保留（含强制 tool_search 规则依赖、记忆、定时）
+#    - 外部桥接工具（含 "mcp__"）默认保留（用户显式配置，误杀代价高）
+#    - 命中关键词信号才追加对应工具（代码/文件/git/shell/浏览器/搜索/子agent/工程化）
+#  纯聊天时模型只背 ~10 个核心工具而非 40+，FC token 可降约 70%+。
+# ══════════════════════════════════════════════════════════════════
+
+# 始终保留的工具（与系统提示里的强制规则/记忆机制强绑定）
+ALWAYS_KEEP = {
+    "tool_search",   # 系统提示的「强制工具查询规则」依赖它
+    "memory",        # 持久化记忆
+    "scheduler",     # 定时任务
+}
+
+# 信号：关键词集合 → 追加「以此前缀开头」的工具
+_SIGNAL_RULES = [
+    # 代码 / 文件操作：Python 等源码信号命中时，连 pylsp 的 lsp__* 一起给
+    (
+        {"写", "创建", "新建", "修改", "编辑", "代码", "脚本", "函数", "bug", "修复",
+         "调试", "重构", "文件", "报错", "语法", "实现", "module", "def ", "import ",
+         "class ", "print(", ".py", ".js", ".ts", ".java", ".go", ".cpp", ".c", ".rs"},
+        ["code_editor", "code_search", "file_manager", "lsp__", "syntax_check", "check"],
+    ),
+    # 版本控制
+    (
+        {"git", "提交", "commit", "分支", "branch", "版本", "diff", "stash", "merge", "rebase"},
+        ["git_tools"],
+    ),
+    # Shell 命令
+    (
+        {"命令", "shell", "终端", "执行", "运行", "cmd", "bash", "powershell", "terminal"},
+        ["cmd_executor"],
+    ),
+    # 浏览器
+    (
+        {"浏览器", "网页", "点击", "browser", "打开网页", "截图", "scroll", "页面"},
+        ["browser_auto"],
+    ),
+    # 联网 / 搜索 / HTTP
+    (
+        {"搜索", "联网", "http", "请求", "url", "curl", "api", "查一下", "上网", "fetch"},
+        ["network_tools", "ai_search"],
+    ),
+    # 子 Agent 并行
+    (
+        {"子agent", "并行", "subagent", "sub_agent", "多智能体"},
+        ["sub_agent"],
+    ),
+    # 工程化自动化
+    (
+        {"测试", "test", "lint", "build", "自动化", "工程化", "ci", "部署", "deploy"},
+        ["auto_engineer"],
+    ),
+]
+
+
+def prune_tools(fc_tools: List[Dict], user_input: str) -> List[Dict]:
+    """
+    按当前用户输入动态裁剪 FC 工具子集。
+
+    返回保留下来的工具列表。无输入或空列表时原样返回（不裁剪）。
+    规则见模块顶部 _SIGNAL_RULES / ALWAYS_KEEP 注释。
+    """
+    if not fc_tools:
+        return fc_tools
+
+    text = (user_input or "").lower()
+
+    keep_prefixes = set()
+    for keywords, prefixes in _SIGNAL_RULES:
+        if any(kw in text for kw in keywords):
+            keep_prefixes.update(prefixes)
+
+    out = []
+    for t in fc_tools:
+        name = (t.get("function") or {}).get("name", "")
+        # 1) 常驻工具
+        if name in ALWAYS_KEEP:
+            out.append(t)
+            continue
+        # 2) 外部桥接工具（用户显式配置）默认保留，避免误杀
+        if "mcp__" in name:
+            out.append(t)
+            continue
+        # 3) 命中当前轮信号 → 追加（前缀匹配，如 lsp__diagnostics 命中 "lsp__"）
+        if any(name.startswith(p) for p in keep_prefixes):
+            out.append(t)
+            continue
+        # 4) 其余（未命中信号的低频工具）本轮裁剪掉
+
+    return out
 
 
 def _extract_fc_def(plugin) -> Optional[Dict]:
@@ -120,11 +241,11 @@ def fc_result_to_message(tool_call_id: str, tool_name: str, result: str) -> Dict
 
     Returns:
         {"role": "tool", "tool_call_id": ..., "content": ...}
+        注意: 官方 OpenAI 的 tool 消息不含 name 字段（严格服务端如讯飞会拒），故不写入。
     """
     return {
         "role": "tool",
         "tool_call_id": tool_call_id,
-        "name": tool_name,
         "content": result[:10000] if result else ""  # 截断过长结果
     }
 
@@ -248,3 +369,81 @@ def _get_builtin_tools() -> List[Dict]:
             }
         }
     ]
+
+
+# ══════════════════════════════════════════════════════════════════
+#  FC 能力探测缓存
+#
+#  部分渠道/模型（如讯飞星火经 one-api 转发的 xophunyuan* 系列）
+#  不支持 function calling，请求体里只要带上非空 tools 字段就会直接
+#  返回 500 RequestParamsError:Invalid Params。
+#  引擎捕获到这种错误后会去掉 tools 降级重试；本缓存把"该模型不支持 FC"
+#  这件事记下来，后续请求直接跳过 tools，避免每次冷启动都白撞一次 500。
+# ══════════════════════════════════════════════════════════════════
+
+_FC_CACHE_FILE = os.path.join(
+    os.path.dirname(os.path.dirname(os.path.abspath(__file__))), ".fc_support.json"
+)
+_fc_cache = None
+_fc_lock = threading.Lock()
+
+
+def _fc_key(base_url, model) -> str:
+    """缓存键：base_url + model（同一渠道下不同模型能力可能不同）"""
+    return f"{(base_url or '').rstrip('/')}::{model or ''}"
+
+
+def _load_fc_cache() -> dict:
+    global _fc_cache
+    if _fc_cache is not None:
+        return _fc_cache
+    try:
+        with open(_FC_CACHE_FILE, 'r', encoding='utf-8') as f:
+            data = json.load(f)
+        _fc_cache = data if isinstance(data, dict) else {}
+    except (FileNotFoundError, json.JSONDecodeError, OSError, ValueError):
+        _fc_cache = {}
+    return _fc_cache
+
+
+def _save_fc_cache(cache: dict):
+    try:
+        with open(_FC_CACHE_FILE, 'w', encoding='utf-8') as f:
+            json.dump(cache, f, ensure_ascii=False, indent=2)
+    except OSError:
+        pass
+
+
+def supports_fc(base_url, model) -> bool:
+    """该 (base_url, model) 是否支持 FC。未探测过时默认 True（乐观，先试一次）"""
+    return _load_fc_cache().get(_fc_key(base_url, model)) is not False
+
+
+def mark_fc_unsupported(base_url, model) -> bool:
+    """标记该 (base_url, model) 不支持 FC 并持久化。首次标记返回 True"""
+    with _fc_lock:
+        cache = _load_fc_cache()
+        key = _fc_key(base_url, model)
+        if cache.get(key) is False:
+            return False
+        cache[key] = False
+        _save_fc_cache(cache)
+        return True
+
+
+def reset_fc_cache(base_url=None, model=None) -> int:
+    """清除 FC 探测缓存。不传参数则全清，返回清除的条目数"""
+    with _fc_lock:
+        cache = _load_fc_cache()
+        if base_url is None and model is None:
+            count = len(cache)
+            cache.clear()
+        else:
+            count = 1 if cache.pop(_fc_key(base_url, model), None) is not None else 0
+        _save_fc_cache(cache)
+        return count
+
+
+def get_fc_cache() -> dict:
+    """返回 FC 探测缓存副本（供 /fc 命令展示）"""
+    return dict(_load_fc_cache())

xcli_core/local_lsp_server.py

@@ -0,0 +1,433 @@
+"""
+自研本地 LSP 服务器 (local_lsp_server)
+======================================
+纯本地、零外部服务依赖的 Python 语言服务器——替代 pylsp。
+
+架构:
+  - 语义引擎: jedi (纯 Python, 已在环境) — 补全/跳转/引用/悬停/签名
+  - 诊断引擎: plugins/py_detect (ast 规则引擎, 中文诊断)
+  - 协议: LSP over stdio (Content-Length 分帧), 与 xiaoli 的 lsp_client.py 完全兼容
+
+特性:
+  - 零网络请求, 不依赖任何外部安装 (无需 pip install pylsp)
+  - 服务器不响应时客户端可超时跳过, 不会无限卡死
+  - 诊断推送自动携带 py_detect 的中文解释 + 修复建议
+
+用法:
+  python -m xcli_core.local_lsp_server
+  或配置进 config.json 的 lspServers: {"command": "python", "args": ["-m", "xcli_core.local_lsp_server"]}
+
+支持的 LSP 方法:
+  initialize / initialized / shutdown / exit
+  textDocument/didOpen|didChange|didClose (通知)
+  textDocument/completion|definition|references|hover (请求)
+  textDocument/publishDiagnostics (主动推送, 来自 py_detect)
+"""
+
+import json
+import os
+import re
+import sys
+from typing import Any, Dict, List, Optional
+
+__version__ = "1.0.0"
+
+# ══════════════════════════════════════════════
+#  Content-Length 分帧 (LSP stdio 传输层)
+# ══════════════════════════════════════════════
+
+
+def read_message(stream) -> Optional[Dict[str, Any]]:
+    """从流中读取一条 Content-Length 分帧消息"""
+    header = b""
+    while b"\r\n\r\n" not in header:
+        chunk = stream.read(1)
+        if not chunk:
+            return None
+        header += chunk
+        if len(header) > 65536:
+            return None
+    m = re.search(rb"Content-Length:\s*(\d+)", header)
+    if not m:
+        return None
+    length = int(m.group(1))
+    body = b""
+    while len(body) < length:
+        chunk = stream.read(length - len(body))
+        if not chunk:
+            return None
+        body += chunk
+    try:
+        return json.loads(body.decode("utf-8"))
+    except (json.JSONDecodeError, UnicodeDecodeError):
+        return None
+
+
+def write_message(stream, obj: Dict[str, Any]) -> None:
+    """向流中写入一条 Content-Length 分帧消息"""
+    data = json.dumps(obj, ensure_ascii=False).encode("utf-8")
+    stream.write(f"Content-Length: {len(data)}\r\n\r\n".encode("ascii"))
+    stream.write(data)
+    stream.flush()
+
+
+# ══════════════════════════════════════════════
+#  路径工具 (file:// URI <-> 文件系统路径)
+# ══════════════════════════════════════════════
+
+
+def uri_to_path(uri: str) -> str:
+    """file:///C:/a/b.py -> C:/a/b.py"""
+    if uri.startswith("file://"):
+        path = uri[7:]
+        if len(path) >= 3 and path[0] == "/" and path[2] == ":":
+            path = path[1:]  # /C:/... -> C:/...
+        return path.replace("/", os.sep) if os.sep == "\\" else path
+    return uri
+
+
+def path_to_uri(path: str) -> str:
+    abspath = os.path.abspath(path)
+    norm = abspath.replace("\\", "/")
+    if not norm.startswith("/"):
+        norm = "/" + norm
+    return "file://" + norm
+
+
+# ══════════════════════════════════════════════
+#  自研本地 LSP 服务器
+# ══════════════════════════════════════════════
+
+
+class LocalLspServer:
+    """本地 Python 语言服务器: jedi 语义 + py_detect 诊断。
+
+    handle(msg) 返回要写出的消息列表（响应/推送），便于单元测试与进程模式复用。
+    """
+
+    def __init__(self):
+        self.documents: Dict[str, str] = {}      # uri -> 最新文本
+        self.root_path: Optional[str] = None
+        self._project = None
+        self._jedi = None
+        self._py_detect = None
+        self._shutdown = False
+
+    # ── 引擎惰性加载 (首次用时才 import, 失败不崩) ──
+    def _get_jedi(self):
+        if self._jedi is None:
+            try:
+                import jedi
+                self._jedi = jedi
+            except Exception:
+                self._jedi = False
+        return self._jedi or None
+
+    def _get_project(self):
+        if self._project is None and self._get_jedi() is not None:
+            try:
+                root = self.root_path
+                if root and os.path.isdir(root):
+                    self._project = self._jedi.Project(root)
+                else:
+                    self._project = False
+            except Exception:
+                self._project = False
+        return self._project if self._project else None
+
+    def _load_py_detect(self):
+        """从 plugins/py_detect.py 加载诊断引擎 (与 code_editor 同源, 零复制)"""
+        if self._py_detect is not None:
+            return self._py_detect if self._py_detect is not False else None
+        try:
+            import importlib.util
+            here = os.path.dirname(os.path.abspath(__file__))
+            plugin_path = os.path.join(here, "..", "plugins", "py_detect.py")
+            if not os.path.exists(plugin_path):
+                self._py_detect = False
+                return None
+            spec = importlib.util.spec_from_file_location("py_detect_server", plugin_path)
+            mod = importlib.util.module_from_spec(spec)
+            spec.loader.exec_module(mod)
+            self._py_detect = mod
+            return mod
+        except Exception:
+            self._py_detect = False
+            return None
+
+    def _script(self, uri: str, text: Optional[str] = None):
+        jedi = self._get_jedi()
+        if jedi is None:
+            return None
+        path = uri_to_path(uri)
+        code = text if text is not None else self.documents.get(uri, "")
+        return jedi.Script(code=code, path=path, project=self._get_project())
+
+    # ── 消息分发: 返回要写出的消息列表 ──
+    def handle(self, msg: Dict[str, Any]) -> List[Dict[str, Any]]:
+        method = msg.get("method")
+        msg_id = msg.get("id")
+        params = msg.get("params") or {}
+
+        # 客户端不应向服务器发响应; 防御性忽略
+        if method is None and msg_id is not None:
+            return []
+
+        if method == "initialize":
+            root_uri = params.get("rootUri") or ""
+            if root_uri:
+                self.root_path = uri_to_path(root_uri)
+            return [self._reply(msg_id, {"capabilities": self._capabilities()})]
+
+        if method == "initialized":
+            return []
+
+        if method == "shutdown":
+            self._shutdown = True
+            return [self._reply(msg_id, None)]
+
+        if method == "exit":
+            self._shutdown = True
+            return []  # 进程层据此退出
+
+        if method == "textDocument/didOpen":
+            td = params.get("textDocument", {})
+            uri = td.get("uri", "")
+            self.documents[uri] = td.get("text", "")
+            return self._diag_push(uri)
+
+        if method == "textDocument/didChange":
+            uri = (params.get("textDocument", {}) or {}).get("uri", "")
+            changes = params.get("contentChanges") or []
+            if changes:
+                self.documents[uri] = changes[-1].get("text", "")
+            return self._diag_push(uri)
+
+        if method == "textDocument/didClose":
+            uri = (params.get("textDocument", {}) or {}).get("uri", "")
+            self.documents.pop(uri, None)
+            return []
+
+        if method == "textDocument/completion":
+            return [self._reply(msg_id, self._completion(params))]
+
+        if method == "textDocument/definition":
+            return [self._reply(msg_id, self._definition(params))]
+
+        if method == "textDocument/references":
+            return [self._reply(msg_id, self._references(params))]
+
+        if method == "textDocument/hover":
+            return [self._reply(msg_id, self._hover(params))]
+
+        if msg_id is not None:
+            return [self._reply(msg_id, None,
+                                error={"code": -32601, "message": f"未知方法: {method}"})]
+        return []
+
+    def _reply(self, msg_id, result, error=None) -> Dict[str, Any]:
+        msg = {"jsonrpc": "2.0", "id": msg_id}
+        if error is not None:
+            msg["error"] = error
+        else:
+            msg["result"] = result
+        return msg
+
+    def _capabilities(self) -> Dict[str, Any]:
+        return {
+            "textDocumentSync": {"openClose": True, "change": 1},  # 1 = full
+            "completionProvider": {"triggerCharacters": [".", "(", ","]},
+            "definitionProvider": True,
+            "referencesProvider": True,
+            "hoverProvider": True,
+        }
+
+    # ── py_detect 诊断 → publishDiagnostics 推送 ──
+    def _diag_push(self, uri: str) -> List[Dict[str, Any]]:
+        pd = self._load_py_detect()
+        text = self.documents.get(uri, "")
+        diags = []
+        if pd is not None:
+            try:
+                issues = pd.analyze(text, uri_to_path(uri) or "<buffer>")
+                sev_map = {"错误": 1, "警告": 2, "提示": 4, "信息": 3}
+                for d in issues:
+                    line0 = max(int(d.get("line", 1)) - 1, 0)
+                    col0 = max(int(d.get("col", 1)) - 1, 0)
+                    sev = sev_map.get(d.get("severity", "警告"), 2)
+                    message = "[{}] {}{}".format(
+                        d.get("code", "?"),
+                        d.get("title", ""),
+                        (" — " + d.get("fix", "")) if d.get("fix") else "",
+                    )
+                    diags.append({
+                        "range": {
+                            "start": {"line": line0, "character": col0},
+                            "end": {"line": line0, "character": col0},
+                        },
+                        "severity": sev,
+                        "source": "py_detect",
+                        "message": message,
+                    })
+            except Exception:
+                diags = []
+        if not diags:
+            return []
+        return [{
+            "jsonrpc": "2.0",
+            "method": "textDocument/publishDiagnostics",
+            "params": {"uri": uri, "diagnostics": diags},
+        }]
+
+    # ── jedi 语义能力 ──
+    def _position(self, params: Dict[str, Any]):
+        pos = params.get("position") or {}
+        return int(pos.get("line", 0) or 0), int(pos.get("character", 0) or 0)
+
+    def _uri(self, params: Dict[str, Any]) -> str:
+        return (params.get("textDocument") or {}).get("uri", "")
+
+    def _completion(self, params: Dict[str, Any]) -> Dict[str, Any]:
+        uri = self._uri(params)
+        line, character = self._position(params)
+        script = self._script(uri)
+        if script is None:
+            return {"items": []}
+        try:
+            completions = script.complete(line, character)
+        except Exception:
+            return {"items": []}
+        items = []
+        seen = set()
+        for c in completions:
+            name = getattr(c, "name", "") or ""
+            if not name or name in seen:
+                continue
+            seen.add(name)
+            ctype = getattr(c, "type", "") or ""
+            detail = (getattr(c, "description", "") or "")[:120]
+            items.append({
+                "label": name,
+                "kind": ctype,
+                "detail": detail,
+                "sortText": f"0{ctype}_{name}",
+            })
+        return {"items": items[:60]}
+
+    def _locations(self, defs) -> List[Dict[str, Any]]:
+        out = []
+        for d in defs:
+            module_path = getattr(d, "module_path", None)
+            if module_path is None:
+                continue  # 内置/无源码定义跳过
+            line = getattr(d, "line", 0) or 0
+            col = getattr(d, "column", 0) or 0
+            out.append({
+                "uri": path_to_uri(str(module_path)),
+                "range": {
+                    "start": {"line": line, "character": col},
+                    "end": {"line": line, "character": col},
+                },
+            })
+        return out
+
+    def _definition(self, params: Dict[str, Any]) -> List[Dict[str, Any]]:
+        uri = self._uri(params)
+        line, character = self._position(params)
+        script = self._script(uri)
+        if script is None:
+            return []
+        try:
+            defs = script.goto(line, character)  # jedi 0.20 API
+            return self._locations(defs)
+        except Exception:
+            return []
+
+    def _references(self, params: Dict[str, Any]) -> List[Dict[str, Any]]:
+        uri = self._uri(params)
+        line, character = self._position(params)
+        script = self._script(uri)
+        if script is None:
+            return []
+        try:
+            refs = script.get_references(line, character)  # jedi 0.20 API
+            return self._locations(refs)
+        except Exception:
+            return []
+
+    def _hover(self, params: Dict[str, Any]) -> Optional[Dict[str, Any]]:
+        uri = self._uri(params)
+        line, character = self._position(params)
+        script = self._script(uri)
+        if script is None:
+            return None
+        parts = []
+        # 类型推断 (jedi 0.20: infer 替代 goto_definitions)
+        try:
+            inferred = script.infer(line, character)
+            seen_types = set()
+            for d in inferred[:5]:
+                dname = getattr(d, "name", "")
+                dtype = getattr(d, "type", "")
+                key = (dname, dtype)
+                if key in seen_types:
+                    continue
+                seen_types.add(key)
+                if dtype:
+                    parts.append(f"类型: {dname}: {dtype}")
+        except Exception:
+            pass
+        # 签名
+        try:
+            for s in script.get_signatures(line, character)[:3]:
+                stxt = getattr(s, "to_string", lambda: "")()
+                if stxt:
+                    parts.append(f"签名: {stxt}")
+        except Exception:
+            pass
+        # 文档
+        try:
+            for h in script.help(line, character)[:3]:
+                doc = getattr(h, "docstring", None)
+                if doc:
+                    d = doc() or ""
+                    if d:
+                        parts.append(d.strip()[:400])
+        except Exception:
+            pass
+        if not parts:
+            return None
+        return {"contents": {"kind": "plaintext", "value": "\n\n".join(parts[:6])}}
+
+
+# ══════════════════════════════════════════════
+#  进程入口
+# ══════════════════════════════════════════════
+
+
+def main() -> None:
+    server = LocalLspServer()
+    stdin = sys.stdin.buffer
+    stdout = sys.stdout.buffer
+    while True:
+        try:
+            msg = read_message(stdin)
+        except Exception:
+            break
+        if msg is None:
+            break
+        try:
+            outputs = server.handle(msg)
+        except Exception:
+            outputs = []
+        for out in outputs:
+            try:
+                write_message(stdout, out)
+            except (BrokenPipeError, OSError):
+                return
+        if server._shutdown and msg.get("method") == "exit":
+            break
+
+
+if __name__ == "__main__":
+    main()

xcli_core/memory.py

@@ -22,7 +22,11 @@ class MemoryManager:
         self.memory_dir = os.path.join(project_dir, "memory")
         self.memory_file = os.path.join(project_dir, "MEMORY.md")
         self.chat_dir = os.path.join(project_dir, "chat_history")
-        self.compress_threshold = 50  # 超过 50 条消息触发压缩
+        self.compress_threshold = 50  # 超过 50 条消息触发压缩（兜底）
+        self.compress_ratio = 0.7    # token 感知压缩阈值：used/max_input_tokens >= 0.7 触发
+        self.keep_recent = 12        # 压缩时保留最近 N 条（编码常跨多轮，略高于原 10）
+        self.max_summary_chars = 1500  # 摘要前单条消息截断上限（原硬编码 200，编码场景丢代码）
+        self.compressed_blobs = []  # 累积的压缩摘要（避免二次摘要失真）
         self._ensure_dirs()
 
     def _ensure_dirs(self):
@@ -299,56 +303,112 @@ class MemoryManager:
     #  上下文压缩
     # ══════════════════════════════════════════════
 
-    def should_compress(self, conversation: list) -> bool:
+    def set_compress_ratio(self, ratio: float):
-        """判断是否需要压缩上下文"""
+        """设置 token 感知压缩阈值（0~1）"""
+        try:
+            self.compress_ratio = float(ratio)
+        except (TypeError, ValueError):
+            pass
+
+    def should_compress(self, conversation: list, prompt_tokens=None, max_input_tokens=None, ratio=None) -> bool:
+        """
+        判断是否需要压缩上下文。
+        优先级:
+          1. token 感知: prompt_tokens/max_input_tokens >= ratio 触发（引擎需返回 max_input_tokens）
+          2. 兜底: 消息数超过 compress_threshold
+        """
+        if ratio is None:
+            ratio = self.compress_ratio
+        if prompt_tokens and max_input_tokens:
+            try:
+                if int(max_input_tokens) > 0 and (int(prompt_tokens) / int(max_input_tokens)) >= ratio:
+                    return True
+            except (TypeError, ValueError):
+                pass
         return len(conversation) > self.compress_threshold
 
-    def compress_context(self, conversation: list, engine=None) -> list:
+    def _is_compressed_marker(self, m: dict) -> bool:
+        """判断一条消息是否是我们自己产出的压缩摘要（避免二次摘要失真）"""
+        return bool(m.get("role") == "system"
+                    and str(m.get("content", "")).startswith("[上下文压缩]"))
+
+    def _prepare_for_summary(self, msg: dict) -> str:
+        """把单条消息转成摘要输入：保真优先——保留工具调用 JSON，长内容截断但尽量留代码块。"""
+        role = msg.get("role", "unknown")
+        content = msg.get("content", "")
+        if not isinstance(content, str):
+            return f"[{role}]: (非文本/结构化内容，长度 {len(str(content))})\n"
+
+        is_tool = ('"action": "use_tool"' in content) or ('"tool":' in content)
+        cap = self.max_summary_chars
+
+        if len(content) <= cap:
+            body = content
+        else:
+            # 尽量保留代码块（```...```）以保真编码上下文
+            fences = re.findall(r"```.*?```", content, re.DOTALL)
+            if fences:
+                kept = "\n".join(fences[:2])
+                if len(kept) > cap:
+                    kept = kept[:cap] + "\n...(截断)"
+                body = f"(内容过长已截断，保留代码块)\n{kept}"
+            else:
+                body = content[:cap] + "\n...(截断)"
+
+        marker = " [工具调用]" if is_tool else ""
+        return f"[{role}]{marker}: {body}\n"
+
+    def compress_context(self, conversation: list, engine=None, keep_recent=None) -> list:
         """
-        压缩对话历史
+        压缩对话历史。
-        策略: 保留最近 10 条 + 用 AI 总结前面的内容
+        策略: 保留最近 keep_recent 条 + 用当前引擎 AI 总结前面的内容。
+        关键改进:
+          - keep_recent 可配置（默认 12，编码常跨多轮）
+          - 已压缩摘要(标记消息)不再二次摘要，避免失真累积
+          - 摘要按角色分条，保留工具调用/代码修改结构
         """
-        if len(conversation) <= 10:
+        if keep_recent is None:
+            keep_recent = self.keep_recent
+        if len(conversation) <= keep_recent:
             return conversation
 
-        keep_recent = 10
-        to_compress = conversation[:-keep_recent]
         recent = conversation[-keep_recent:]
+        prefix = conversation[:-keep_recent]
+        # 只摘要尚未压缩过的"真消息"，跳过历史摘要本身 → 不二次失真
+        new_to_compress = [m for m in prefix if not self._is_compressed_marker(m)]
+        if not new_to_compress:
+            # 前缀全是历史摘要，没有新内容可压，保持现状即可（不丢摘要）
+            return conversation
 
-        # 尝试用 AI 生成摘要
+        summary = self._ai_summarize(new_to_compress, engine)
-        summary = self._ai_summarize(to_compress, engine)
+        block = f"[上下文压缩] 之前 {len(new_to_compress)} 条对话摘要:\n{summary}"
+        self.compressed_blobs.append(block)
 
-        # 构建压缩后的历史
+        # 累积所有摘要合并为单条 system（而非每次重压旧摘要）
-        compressed = [{
+        merged = "\n\n".join(self.compressed_blobs)
-            "role": "system",
+        compressed = [{"role": "system", "content": merged}]
-            "content": f"[上下文压缩] 以下是之前 {len(to_compress)} 条对话的摘要:\n{summary}"
-        }]
         compressed.extend(recent)
         return compressed
 
     def _ai_summarize(self, messages: list, engine=None) -> str:
-        """用 AI 生成对话摘要"""
+        """用当前引擎生成对话摘要（保真优先）"""
         if not engine:
             return self._simple_summarize(messages)
 
-        # 构建摘要请求
+        conversation_text = "".join(self._prepare_for_summary(m) for m in messages)
-        conversation_text = ""
-        for msg in messages:
-            role = msg.get("role", "unknown")
-            content = msg.get("content", "")
-            if isinstance(content, str):
-                conversation_text += f"[{role}]: {content[:200]}\n"
 
         prompt = (
-            "请用简洁的中文总结以下对话的关键信息（工具调用、代码修改、重要决策）。"
+            "请用简洁中文总结以下对话的关键信息，要求:\n"
-            "控制在 200 字以内。\n\n"
+            "1. 按角色(用户/助手)分条，保留每条的要点；\n"
+            "2. 重点保留: 工具调用名称与参数、代码修改点(读/改/创建的文件与位置)、重要决策；\n"
+            "3. 控制总体在 400 字以内，不要编造未提及的信息。\n\n"
             f"{conversation_text}"
         )
 
         try:
             summary = engine.generate_response(
                 prompt,
-                system_prompt="你是摘要生成器。只输出摘要，不要多余内容。"
+                system_prompt="你是摘要生成器。只输出结构化摘要，不要多余内容。"
             )
             if summary and len(summary) > 10:
                 return summary
@@ -358,23 +418,25 @@ class MemoryManager:
         return self._simple_summarize(messages)
 
     def _simple_summarize(self, messages: list) -> str:
-        """简单摘要（不依赖 AI）"""
+        """简单摘要（不依赖 AI，保真优先）"""
         topics = []
         tool_calls = []
+        edits = []
         for msg in messages:
             content = msg.get("content", "")
             role = msg.get("role", "")
-            if role == "user" and isinstance(content, str):
+            if not isinstance(content, str):
+                continue
+            if role == "user":
                 # 提取用户请求
                 if len(content) < 100:
                     topics.append(content)
                 else:
                     topics.append(content[:50] + "...")
-            if role == "assistant" and isinstance(content, str):
+            if role == "assistant":
                 # 提取工具调用
                 if '"action": "use_tool"' in content or '"tool":' in content:
                     try:
-                        # 尝试解析工具名
                         for part in content.split('"tool":'):
                             if len(part) > 2:
                                 tool_name = part.split('"')[1] if '"' in part else ""
@@ -382,12 +444,19 @@ class MemoryManager:
                                     tool_calls.append(tool_name)
                     except Exception:
                         pass
+                # 提取代码编辑点（edit 命令）
+                for m in re.finditer(r'edit\s+(\S+)', content):
+                    fn = m.group(1)
+                    if fn not in edits:
+                        edits.append(fn)
 
         summary_parts = []
         if topics:
             summary_parts.append(f"讨论了 {len(topics)} 个话题: {'; '.join(topics[:5])}")
         if tool_calls:
             summary_parts.append(f"使用了工具: {', '.join(tool_calls[:8])}")
+        if edits:
+            summary_parts.append(f"编辑了文件: {', '.join(edits[:8])}")
         summary_parts.append(f"共 {len(messages)} 条对话")
 
         return "。".join(summary_parts)

xcli_core/model_registry.py

@@ -0,0 +1,145 @@
+"""OpenAI 格式模型注册表。
+
+把「多个 OpenAI 格式模型」从 config.json 的单槽位升级为可管理的列表，
+支持在线增删、切换、持久化，免去每次手改配置文件。
+
+配置结构（api.engines.openai）:
+    {
+      "models": [
+        {"name": "讯飞xop", "base_url": "...", "api_key": "...", "model": "xophunyuan7bmt"},
+        {"name": "gpt4o",   "base_url": "https://api.openai.com/v1", "api_key": "...", "model": "gpt-4o"}
+      ],
+      "current": "讯飞xop"
+    }
+
+向后兼容：若没有 models 列表，把扁平的 api_key/base_url/model 当作单条隐式模型，
+其 name 取 model 值（空则 "default"）。
+"""
+
+from xcli_core.config import load_config, save_config
+
+# 测试可临时覆盖配置路径
+_CONFIG_PATH = None
+
+
+def _set_config_path(path):
+    """仅供测试：指向临时 config.json"""
+    global _CONFIG_PATH
+    _CONFIG_PATH = path
+
+
+def _load():
+    if _CONFIG_PATH:
+        try:
+            with open(_CONFIG_PATH, 'r', encoding='utf-8') as f:
+                return __import__('json').load(f)
+        except (FileNotFoundError, __import__('json').JSONDecodeError):
+            return {"api": {"engines": {}}, "system": {}}
+    return load_config()
+
+
+def _persist(cfg):
+    if _CONFIG_PATH:
+        with open(_CONFIG_PATH, 'w', encoding='utf-8') as f:
+            __import__('json').dump(cfg, f, ensure_ascii=False, indent=2)
+        return
+    save_config(cfg)
+
+
+def _openai_block(cfg):
+    return cfg.setdefault("api", {}).setdefault("engines", {}).setdefault("openai", {})
+
+
+def list_models(cfg=None):
+    """返回 (models:list[dict], current:str|None)。
+
+    无 models 列表时返回单条隐式模型（由扁平 api_key/base_url/model 构成）。
+    """
+    cfg = cfg or _load()
+    block = _openai_block(cfg)
+    models = block.get("models")
+    if not models:
+        m = block.get("model") or "default"
+        implicit = {
+            "name": m,
+            "base_url": block.get("base_url", ""),
+            "api_key": block.get("api_key", ""),
+            "model": block.get("model", ""),
+        }
+        return [implicit], (block.get("current") or m)
+    current = block.get("current") or (models[0]["name"] if models else None)
+    return models, current
+
+
+def get_model(name=None, cfg=None):
+    """取某模型配置；name=None 取 current。找不到返回 None。"""
+    models, current = list_models(cfg)
+    target = name or current
+    if not target:
+        return None
+    for m in models:
+        if m.get("name") == target:
+            return dict(m)
+    return None
+
+
+def set_current(name):
+    """把 current 指针指向 name 并落盘。name 必须存在，否则返回 False。"""
+    cfg = _load()
+    block = _openai_block(cfg)
+    models = block.get("models") or []
+    if not any(m.get("name") == name for m in models):
+        return False
+    block["current"] = name
+    _persist(cfg)
+    return True
+
+
+def add_model(entry):
+    """新增或更新模型，落盘并返回其 name。
+
+    entry: {"name", "base_url", "api_key", "model"}。同名则更新。
+    首次添加会从扁平配置迁移出第一条隐式模型，且保持 current 不跳变。
+    """
+    name = (entry.get("name") or "").strip()
+    if not name:
+        raise ValueError("模型 name 不能为空")
+    cfg = _load()
+    block = _openai_block(cfg)
+    models = block.get("models")
+    if not models:
+        old_name = block.get("model") or "default"
+        models = [{
+            "name": old_name,
+            "base_url": block.get("base_url", ""),
+            "api_key": block.get("api_key", ""),
+            "model": block.get("model", ""),
+        }]
+        block["models"] = models
+        if not block.get("current"):
+            block["current"] = old_name
+    for m in models:
+        if m.get("name") == name:
+            m.update(entry)
+            _persist(cfg)
+            return name
+    models.append(dict(entry))
+    _persist(cfg)
+    return name
+
+
+def remove_model(name):
+    """删除模型，返回是否成功（找不到返回 False）。
+    若删除的是 current，则 current 回退到列表第一个。
+    """
+    cfg = _load()
+    block = _openai_block(cfg)
+    models = block.get("models") or []
+    new = [m for m in models if m.get("name") != name]
+    if len(new) == len(models):
+        return False
+    block["models"] = new
+    if block.get("current") == name:
+        block["current"] = new[0]["name"] if new else None
+    _persist(cfg)
+    return True

xcli_core/provider_router.py

@@ -0,0 +1,269 @@
+"""
+Provider 抽象层（opencode 式原生多 Provider 路由）
+
+小狸核心引擎是 OpenAI 兼容的 OpenaiAI（已支持 base_url，讯飞/xunfei 即借此接入）。
+本模块在此基础上提供：
+  - PROVIDER_PRESETS：主流 Provider 的默认 base_url / 模型前缀 / 协议族
+  - resolve_provider()：从 config["providers"] 解析出 {api_key, base_url, model, headers}
+  - convert_messages() / convert_tools()：OpenAI ↔ Anthropic ↔ Google 线缆格式互转（纯函数）
+  - list_providers()：给 /providers 命令用的汇总
+
+设计原则：纯函数 + 配置驱动，不触网、不强求真实 key，不改动现有 engine 行为。
+真实调用仍由 OpenaiAI 在用户填入 base_url/api_key 后完成（OpenAI 兼容族直接可用；
+非兼容族可借助 convert_* 在 engine 层二次适配，本模块只负责「翻译」与「注册」）。
+"""
+from __future__ import annotations
+
+import os
+from typing import Any, Dict, List, Optional
+
+# ───────────────────────── Provider 预设 ─────────────────────────
+# protocol: openai 兼容 | anthropic | google
+PROVIDER_PRESETS: Dict[str, Dict[str, Any]] = {
+    "openai": {
+        "label": "OpenAI",
+        "base_url": "https://api.openai.com/v1",
+        "protocol": "openai",
+        "models": ["gpt-4o", "gpt-4o-mini", "o1", "o3-mini"],
+        "env_key": "OPENAI_API_KEY",
+    },
+    "anthropic": {
+        "label": "Anthropic Claude",
+        "base_url": "https://api.anthropic.com/v1",
+        "protocol": "anthropic",
+        "models": ["claude-opus-4", "claude-sonnet-4", "claude-3-5-sonnet"],
+        "env_key": "ANTHROPIC_API_BASE",
+    },
+    "google": {
+        "label": "Google Gemini",
+        "base_url": "https://generativelanguage.googleapis.com/v1beta/openai",
+        "protocol": "google",
+        "models": ["gemini-2.0-pro", "gemini-2.0-flash", "gemini-1.5-pro"],
+        "env_key": "GEMINI_API_KEY",
+    },
+    "deepseek": {
+        "label": "DeepSeek",
+        "base_url": "https://api.deepseek.com/v1",
+        "protocol": "openai",
+        "models": ["deepseek-chat", "deepseek-reasoner"],
+        "env_key": "DEEPSEEK_API_KEY",
+    },
+    "moonshot": {
+        "label": "Moonshot (Kimi)",
+        "base_url": "https://api.moonshot.cn/v1",
+        "protocol": "openai",
+        "models": ["moonshot-v1-8k", "moonshot-v1-32k"],
+        "env_key": "MOONSHOT_API_KEY",
+    },
+    "qwen": {
+        "label": "阿里通义千问",
+        "base_url": "https://dashscope.aliyuncs.com/compatible-mode/v1",
+        "protocol": "openai",
+        "models": ["qwen-max", "qwen-plus", "qwen-turbo"],
+        "env_key": "DASHSCOPE_API_KEY",
+    },
+    "xunfei": {
+        "label": "讯飞开放平台",
+        "base_url": "https://spark-api-open.xf-yun.com/v1",
+        "protocol": "openai",
+        "models": ["generalv3.5", "generalv4"],
+        "env_key": "XUNFEI_API_KEY",
+    },
+    "ollama": {
+        "label": "Ollama (本地)",
+        "base_url": "http://localhost:11434/v1",
+        "protocol": "openai",
+        "models": ["qwen2.5", "llama3.1", "deepseek-r1"],
+        "env_key": "",
+    },
+}
+
+# 协议族 → 该族内「消息格式」标识（转换时使用）
+_PROTOCOL_FORMAT = {
+    "openai": "openai",
+    "anthropic": "anthropic",
+    "google": "google",
+}
+
+
+def get_provider_preset(name: str) -> Optional[Dict[str, Any]]:
+    return PROVIDER_PRESETS.get(name)
+
+
+def list_providers() -> List[str]:
+    return list(PROVIDER_PRESETS.keys())
+
+
+def resolve_provider(name: str, providers_cfg: Optional[Dict[str, Any]] = None,
+                     env: Optional[Dict[str, str]] = None) -> Dict[str, Any]:
+    """从 config["providers"][name] 解析出连接参数。
+
+    优先级：providers_cfg 显式字段 > 环境变量 > 预设默认值。
+    返回：{name, label, base_url, api_key, model, protocol, headers}
+    缺失 api_key 时 api_key 为空字符串（调用层据此判断是否配置完成）。
+    """
+    preset = PROVIDER_PRESETS.get(name)
+    if preset is None:
+        return {}
+    env = env if env is not None else dict(os.environ)
+    cfg = (providers_cfg or {}).get(name, {}) if providers_cfg else {}
+
+    api_key = cfg.get("api_key") or (env.get(preset["env_key"]) if preset["env_key"] else "")
+    base_url = cfg.get("base_url") or preset["base_url"]
+    model = cfg.get("model") or (preset["models"][0] if preset["models"] else "")
+    protocol = preset["protocol"]
+
+    headers: Dict[str, str] = {}
+    # Anthropic 需要 beta header与协议版本（即便走 OpenAI 代理也保留以便真实适配）
+    if protocol == "anthropic":
+        headers["anthropic-version"] = "2023-06-01"
+    # Google 的 Gemini OpenAI 兼容端点需要把 key 放 query/header，这里统一放 header
+    if protocol == "google" and api_key:
+        headers["x-goog-api-key"] = api_key
+
+    return {
+        "name": name,
+        "label": preset["label"],
+        "base_url": base_url,
+        "api_key": api_key or "",
+        "model": model,
+        "protocol": protocol,
+        "headers": headers,
+    }
+
+
+def _wire_format(protocol: str) -> str:
+    return _PROTOCOL_FORMAT.get(protocol, "openai")
+
+
+# ───────────────────────── 消息格式互转 ─────────────────────────
+def convert_messages(target: str, messages: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
+    """把小狸内部 OpenAI 风格 messages 转换为 target provider 的线缆格式。
+
+    target ∈ {"openai", "anthropic", "google"}
+    内部统一表示：[{"role": "system"|"user"|"assistant"|"tool", "content": ..., "tool_calls"?, "name"?}]
+    - anthropic：system 抽成顶层；user/assistant 交替，content 包装为 blocks
+    - google：system 抽成 systemInstruction；content 包装为 parts
+    纯转换，不调用网络。
+    """
+    fmt = _wire_format(target) if target in _PROTOCOL_FORMAT.values() else _wire_format(target)
+    if fmt == "openai":
+        return [_normalize_openai(m) for m in messages]
+    if fmt == "anthropic":
+        return _to_anthropic(messages)
+    if fmt == "google":
+        return _to_google(messages)
+    return [_normalize_openai(m) for m in messages]
+
+
+def _normalize_openai(m: Dict[str, Any]) -> Dict[str, Any]:
+    """确保 content 为字符串（统一线缆格式），保留 role/tool_calls/name。"""
+    out = {"role": m.get("role", "user")}
+    content = m.get("content")
+    if content is None:
+        content = ""
+    if isinstance(content, list):
+        # 多模态 blocks → 拼接文本
+        text = "".join(p.get("text", "") for p in content if isinstance(p, dict))
+        content = text
+    out["content"] = content
+    if "tool_calls" in m:
+        out["tool_calls"] = m["tool_calls"]
+    if "name" in m:
+        out["name"] = m["name"]
+    return out
+
+
+def _to_anthropic(messages: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
+    """Anthropic Messages API 格式：system 单独抽离，消息为 user/assistant 交替。"""
+    out: List[Dict[str, Any]] = []
+    for m in messages:
+        role = m.get("role")
+        if role == "system":
+            continue  # 顶层 system 由调用方单独处理
+        content = m.get("content", "")
+        blocks = []
+        if isinstance(content, list):
+            for p in content:
+                if isinstance(p, dict) and p.get("type") == "text":
+                    blocks.append({"type": "text", "text": p.get("text", "")})
+                elif isinstance(p, dict) and p.get("type") == "image":
+                    blocks.append(p)
+        else:
+            blocks.append({"type": "text", "text": str(content)})
+        out.append({"role": "user" if role in ("user", "tool") else "assistant",
+                    "content": blocks})
+    return out
+
+
+def _to_google(messages: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
+    """Google Gemini 格式：system 抽为 systemInstruction，消息为 content.parts。"""
+    out: List[Dict[str, Any]] = []
+    for m in messages:
+        role = m.get("role")
+        if role == "system":
+            continue
+        content = m.get("content", "")
+        parts = []
+        if isinstance(content, list):
+            for p in content:
+                if isinstance(p, dict) and p.get("type") == "text":
+                    parts.append({"text": p.get("text", "")})
+                elif isinstance(p, dict) and p.get("type") == "image":
+                    parts.append(p)
+        else:
+            parts.append({"text": str(content)})
+        out.append({"role": "user" if role in ("user", "tool") else "model",
+                    "parts": parts})
+    return out
+
+
+def convert_tools(target: str, tools: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
+    """把 OpenAI FC tools([{type, function:{name,description,parameters}}]) 转目标格式。
+
+    - openai：原样
+    - anthropic：tools=[{name, description, input_schema}]
+    - google：tools=[{functionDeclarations:[{name,description,parameters}]}]
+    """
+    fmt = _wire_format(target)
+    if fmt == "openai":
+        return tools
+    if fmt == "anthropic":
+        out = []
+        for t in tools:
+            fn = t.get("function", {})
+            out.append({
+                "name": fn.get("name", ""),
+                "description": fn.get("description", ""),
+                "input_schema": fn.get("parameters", {"type": "object", "properties": {}}),
+            })
+        return out
+    if fmt == "google":
+        decls = []
+        for t in tools:
+            fn = t.get("function", {})
+            decls.append({
+                "name": fn.get("name", ""),
+                "description": fn.get("description", ""),
+                "parameters": fn.get("parameters", {"type": "object", "properties": {}}),
+            })
+        return [{"functionDeclarations": decls}] if decls else []
+    return tools
+
+
+def provider_summary(name: str, providers_cfg: Optional[Dict[str, Any]] = None,
+                     env: Optional[Dict[str, str]] = None) -> Dict[str, Any]:
+    """给 /providers 命令用的单条汇总（是否配置、协议、模型数）。"""
+    preset = PROVIDER_PRESETS.get(name)
+    if not preset:
+        return {}
+    resolved = resolve_provider(name, providers_cfg, env)
+    configured = bool(resolved.get("api_key"))
+    return {
+        "name": name,
+        "label": preset["label"],
+        "protocol": preset["protocol"],
+        "configured": configured,
+        "model_count": len(preset["models"]),
+        "base_url": resolved.get("base_url", ""),
+    }

xcli_core/rich_tui.py

@@ -0,0 +1,134 @@
+"""
+v8.0 Rich 面板式 TUI + 可截取快照出口
+======================================
+- 通过 cli.tui_output_callback = RichTUI.push 接收逐行输出，渲染为 rich 面板
+- Console(record=True) 快照出口: export_text / export_svg / export_html
+  → AI 开发 TUI 时能把渲染结果读回自验，补上“看不见渲染”的反馈闭环
+
+设计取舍（与 7.0 重构前 textual TUI 的区别）:
+- 不用 textual 的异步全屏 App，避免与同步 input() 主循环冲突；
+- 采用 rich 顺序面板流（一问一答往下渲染），天然支持 record=True 快照；
+- 这正是“让 AI 能看见自己写的 TUI”的关键能力。
+"""
+import os
+from rich.console import Console
+from rich.panel import Panel
+from rich.text import Text
+from rich.box import ROUNDED
+
+SNAP_DIR = os.path.join(
+    os.path.dirname(os.path.dirname(os.path.abspath(__file__))), ".tui_snapshots"
+)
+
+
+class RichTUI:
+    """rich 面板式 TUI：聊天面板 + 状态栏 + 快照出口。"""
+
+    def __init__(self, cli):
+        self.cli = cli
+        # record=True 是关键：渲染过程被记录，可随时导出文本/SVG/HTML 自验
+        self.console = Console(record=True, width=100)
+        self.version = "8.0.3"
+        self.engine = ""
+        self.model = ""
+        self.mode = "普通"
+        self.used_tokens = None
+        self.context_window = None
+        self.msg_count = 0
+        self._log = []
+
+    # ── 状态设置（由 cli 在切换引擎/压缩后刷新）──
+    def set_status(self, engine=None, model=None, mode=None,
+                   used_tokens=None, context_window=None, msg_count=None):
+        if engine is not None:
+            self.engine = engine
+        if model is not None:
+            self.model = model
+        if mode is not None:
+            self.mode = mode
+        if used_tokens is not None:
+            self.used_tokens = used_tokens
+        if context_window is not None:
+            self.context_window = context_window
+        if msg_count is not None:
+            self.msg_count = msg_count
+
+    def banner(self):
+        self.console.print(Panel(
+            Text.from_markup(f"[bold cyan]小狸 Pro-CLI[/]  [yellow]v{self.version}[/]  ·  Rich TUI"),
+            subtitle="[dim]自然语言对话 · /exit 返回 CLI · /snapshot 导出快照[/]",
+            box=ROUNDED, border_style="cyan"))
+        self.console.print()
+
+    def _bar(self):
+        if self.used_tokens and self.context_window:
+            ratio = min(1.0, self.used_tokens / self.context_window)
+            filled = int(ratio * 10)
+            bar = "▓" * filled + "░" * (10 - filled)
+            tok = f"{self.used_tokens}/{self.context_window} {bar}"
+        elif self.used_tokens:
+            tok = f"{self.used_tokens} tokens"
+        else:
+            tok = "-"
+        return (f"[dim]引擎 {self.engine or '-'} · 模型 {self.model or '-'} · "
+                f"模式 {self.mode} · {tok} · 消息 {self.msg_count}[/]")
+
+    # ── 回调入口：DisplayMixin._output 把每行输出投到这里 ──
+    def push(self, message):
+        m = (message or "").strip("\n")
+        if not m:
+            return
+        self._log.append(m)
+        if len(self._log) > 300:
+            self._log = self._log[-300:]
+        # 原始 message 已由 DisplayMixin 产出合法 rich markup，直接打印；
+        # 个别工具结果含裸方括号可能破坏 markup，失败则降级为纯文本。
+        try:
+            self.console.print(m)
+        except Exception:
+            self.console.print(m, markup=False)
+
+    # ── 快照出口（AI 自验渲染的关键能力）──
+    def snapshot_text(self):
+        return self.console.export_text(clear=False)
+
+    def snapshot_svg(self, title="小狸 Pro-CLI v8.0 Rich TUI"):
+        return self.console.export_svg(title=title, clear=False)
+
+    def snapshot_html(self, title="小狸 Pro-CLI v8.0 Rich TUI"):
+        # rich 新版 export_html 不接受 title 参数，标题作为注释写入
+        html = self.console.export_html(clear=False)
+        return f"<!-- {title} -->\n{html}"
+
+    def save_snapshot(self, stem):
+        os.makedirs(os.path.dirname(stem), exist_ok=True)
+        with open(stem + ".txt", "w", encoding="utf-8") as f:
+            f.write(self.snapshot_text())
+        with open(stem + ".svg", "w", encoding="utf-8") as f:
+            f.write(self.snapshot_svg())
+        with open(stem + ".html", "w", encoding="utf-8") as f:
+            f.write(self.snapshot_html())
+        return stem + ".txt", stem + ".svg", stem + ".html"
+
+    # ── 主循环 ──
+    def run(self):
+        self.banner()
+        try:
+            import readline  # 可选：输入历史补全
+        except Exception:
+            pass
+        while True:
+            try:
+                ui = input("> ").strip()
+            except (EOFError, KeyboardInterrupt):
+                break
+            if ui in ("/exit", "/quit"):
+                break
+            if ui == "/snapshot":
+                paths = self.save_snapshot(os.path.join(SNAP_DIR, "live"))
+                print(f"[快照已保存] {paths[2]}")
+                continue
+            if ui:
+                self.cli.process_conversation(ui)
+                self.msg_count = len(self.cli.shared_conversation_history)
+                self.console.print(Text.from_markup(self._bar()))

xcli_core/safety.py

@@ -18,6 +18,65 @@ MODE_NAMES = {
     MODE_MANUAL: "人工确认",
 }
 
+# ── Plan 模式只读白名单（黑名单法）──
+# 值为 "*" 表示该工具整类视为写/有副作用，一律拦截；
+# 值为列表时，仅当首个操作词命中列表中的写操作关键字才拦截；
+# 值为 [] 表示该工具整类只读，一律放行（调研必需）；
+# 未登记的未知工具默认拦截（保守）。
+PLAN_WRITE_RULES = {
+    # 整类拦截
+    "cmd_executor": "*",
+    "browser_auto": "*",
+    "network_tools": "*",
+    "scheduler": "*",
+    # 仅部分操作拦截
+    "code_editor": ["write", "create", "append", "edit", "insert", "delete_lines",
+                    "multi", "replace", "rename", "move"],
+    "git_tools": ["commit", "push", "checkout", "reset", "merge", "revert", "clean",
+                  "stash", "tag", "fetch", "pull", "branch -", "branch delete",
+                  "branch -d", "branch -D", "remote add", "remote set", "amend"],
+    "file_manager": ["write", "delete", "move", "copy", "rename", "mkdir", "rm",
+                     "create", "save", "touch"],
+    "sub_agent": ["run", "dispatch", "execute", "spawn", "create", "new", "task", "assign"],
+    "auto_engineer": ["run", "test", "build", "fix", "format", "apply", "generate",
+                      "scaffold", "create", "refactor", "exec"],
+    "memory": ["write", "save", "add", "set", "delete", "edit", "update", "forget"],
+    # 整类放行（只读探索必需）
+    "tool_search": [],
+    "ai_search": [],
+    "code_search": [],
+}
+
+
+def plan_is_write_operation(tool_name: str, tool_args) -> bool:
+    """Plan 模式下判断一次工具调用是否属于「写操作」（应被拦截）。
+
+    返回 True 表示会修改状态/文件/系统，应拦截；False 表示只读，可放行。
+    """
+    rules = PLAN_WRITE_RULES.get(tool_name)
+    if rules is None:
+        # 未登记的未知工具：保守拦截
+        return True
+    if rules == "*":
+        return True
+    if not rules:
+        return False
+    args_l = str(tool_args).strip().lower()
+    parts = args_l.split()
+    op = parts[0] if parts else ""
+    for kw in rules:
+        kwp = kw.split()
+        if len(kwp) == 1:
+            # 单字关键字：精确匹配操作词，或作为其前缀/后缀（避免 "editor" 误中 "edit"）
+            if op == kw or op.startswith(kw) or op.endswith(kw):
+                return True
+        else:
+            # 多词规则，如 "remote add" / "branch -d"
+            if args_l.startswith(kw) or " ".join(parts[:len(kwp)]) == kw:
+                return True
+    return False
+
+
 class SafetyLayer:
     """统一安全检查层"""
 
@@ -26,6 +85,7 @@ class SafetyLayer:
         self.cli = None
         self._analysis_cache = {}  # 缓存分析结果
         self._lock = threading.Lock()
+        self._permissions_override = None  # set_permissions 注入的显式权限
 
     def set_cli(self, cli):
         self.cli = cli
@@ -44,6 +104,62 @@ class SafetyLayer:
         self.mode = (self.mode + 1) % 3
         return self.get_mode_name()
 
+    # ── 声明式权限（opencode 式 allow/ask/deny）──
+    # 规则语法：
+    #   "*"               → 匹配任意工具/操作
+    #   "tool"            → 精确匹配工具名
+    #   "tool:*"          → 该工具任意操作
+    #   "tool:op"         → 工具名 + 参数首个词命中 op 前缀
+    # 优先级：deny（阻断）> ask（强制确认）> allow（自动放行）> 默认（按模式）
+
+    def set_permissions(self, allow=None, ask=None, deny=None):
+        """显式注入权限规则（测试或启动时调用），优先级最高。"""
+        self._permissions_override = {
+            "allow": list(allow or []),
+            "ask": list(ask or []),
+            "deny": list(deny or []),
+        }
+
+    def _get_permissions(self) -> dict:
+        if self._permissions_override is not None:
+            return self._permissions_override
+        return self._load_permissions()
+
+    def _load_permissions(self) -> dict:
+        try:
+            from .config import load_config
+        except ImportError:
+            try:
+                from config import load_config
+            except ImportError:
+                return {"allow": [], "ask": [], "deny": []}
+        cfg = load_config() or {}
+        perm = cfg.get("permissions") or {}
+        return {
+            "allow": perm.get("allow", []) or [],
+            "ask": perm.get("ask", []) or [],
+            "deny": perm.get("deny", []) or [],
+        }
+
+    def _permission_match(self, rule: str, tool_name: str, tool_args) -> bool:
+        rule = (rule or "").strip()
+        if not rule:
+            return False
+        if rule == "*":
+            return True
+        if ":" in rule:
+            rtool, raction = rule.split(":", 1)
+            rtool = rtool.strip()
+            raction = raction.strip()
+            if rtool != tool_name:
+                return False
+            if raction in ("*", ""):
+                return True
+            op = (str(tool_args).strip().split()[0].lower()
+                  if str(tool_args).strip() else "")
+            return op == raction or op.startswith(raction)
+        return rule == tool_name
+
     # ── 核心：拦截检查 ──
 
     def check(self, tool_name: str, tool_args: str) -> Tuple[bool, str]:
@@ -51,6 +167,34 @@ class SafetyLayer:
         检查工具调用是否安全
         返回: (允许执行, 消息)
         """
+        # ── Plan 模式只读约束（最高优先级，先于所有安全模式判断）──
+        # plan 模式下只允许只读调研工具，写操作一律暂挂，待 /build 批准后执行。
+        cli = self.cli
+        if cli is not None and getattr(cli, 'plan_mode', False):
+            if plan_is_write_operation(tool_name, tool_args):
+                return False, "PLAN 模式：写操作已暂挂（只读调研中）。请在 /build 批准后执行。"
+            return True, ""
+
+        # ── 声明式权限（config.permissions: deny > ask > allow）──
+        perm = self._get_permissions()
+        # deny 最高优先级：直接阻断（即便无限制模式也生效）
+        for rule in perm.get("deny", []):
+            if self._permission_match(rule, tool_name, tool_args):
+                return False, f"权限拒绝（deny 规则 '{rule}' 命中）: {tool_name}"
+        # ask：强制人工确认（覆盖自动放行，即便无限制模式也确认）
+        for rule in perm.get("ask", []):
+            if self._permission_match(rule, tool_name, tool_args):
+                return self._ask_user_confirm(
+                    tool_name, tool_args,
+                    risk_level="中",
+                    impact="命中 ask 规则，需你确认后执行",
+                    reason=f"权限规则 '{rule}'"
+                )
+        # allow：自动放行（跳过 AI 风险分析）
+        for rule in perm.get("allow", []):
+            if self._permission_match(rule, tool_name, tool_args):
+                return True, ""
+
         # 无限制模式：直接放行
         if self.mode == MODE_UNRESTRICTED:
             return True, ""

xcli_core/session.py

@@ -0,0 +1,201 @@
+"""会话（Session）持久化与恢复。
+
+把 openai/ollama/manual 等引擎的对话历史（self.shared_conversation_history）
+按「会话」维度持久化到 chat_history/<id>.json，支持：
+  - 自动落盘（每轮对话后调用 SessionManager.save）
+  - 列出历史会话（/resume、/sessions）
+  - 一键恢复（把 messages 重新载入 shared_conversation_history）
+
+存储格式（新）：
+  {
+    "id": "20260805-171045-a1b2c3",
+    "title": "用户首条消息前若干个字",
+    "created_at": "2026-08-05 17:10:45",
+    "updated_at": "2026-08-05 17:15:22",
+    "engine": "openai",
+    "model": "xophunyuan7bmt",
+    "messages": [ ... ]
+  }
+
+兼容旧 /chat save 写入的文件（无 id 字段、用 "conversation" 键）：
+  读取时回退——id 取文件名，messages 取 conversation 键。
+"""
+import os
+import re
+import time
+import json
+import uuid
+
+
+def _make_session_id() -> str:
+    """生成形如 20260805-171045-a1b2c3 的会话 ID。"""
+    ts = time.strftime("%Y%m%d-%H%M%S")
+    suffix = uuid.uuid4().hex[:6]
+    return f"{ts}-{suffix}"
+
+
+def _title_from_messages(messages, limit: int = 40) -> str:
+    """从首条 user 消息生成会话标题。"""
+    for m in messages or []:
+        if isinstance(m, dict) and m.get("role") == "user":
+            content = m.get("content", "")
+            if isinstance(content, list):  # 多模态 content 可能是列表
+                parts = [c.get("text", "") for c in content if isinstance(c, dict)]
+                content = " ".join(parts)
+            content = (content or "").strip().replace("\n", " ")
+            return content[:limit] if content else "(空消息)"
+    return "(未命名会话)"
+
+
+class Session:
+    """单个会话的轻量容器。"""
+
+    def __init__(self, sid, title, created_at, updated_at, engine, model, messages):
+        self.id = sid
+        self.title = title
+        self.created_at = created_at
+        self.updated_at = updated_at
+        self.engine = engine
+        self.model = model
+        self.messages = messages or []
+
+    @property
+    def msg_count(self) -> int:
+        return len(self.messages)
+
+    def to_dict(self) -> dict:
+        return {
+            "id": self.id,
+            "title": self.title,
+            "created_at": self.created_at,
+            "updated_at": self.updated_at,
+            "engine": self.engine,
+            "model": self.model,
+            "messages": self.messages,
+        }
+
+    @classmethod
+    def from_dict(cls, data: dict, fallback_id: str = ""):
+        # 兼容旧 /chat save 格式：无 id、用 "conversation" 键
+        sid = data.get("id") or fallback_id
+        messages = data.get("messages")
+        if messages is None and "conversation" in data:
+            messages = data["conversation"]
+        return cls(
+            sid=sid,
+            title=data.get("title") or _title_from_messages(messages),
+            created_at=data.get("created_at") or data.get("timestamp") or "",
+            updated_at=data.get("updated_at") or data.get("timestamp") or "",
+            engine=data.get("engine") or "",
+            model=data.get("model") or "",
+            messages=messages or [],
+        )
+
+
+class SessionManager:
+    """会话的读写与索引管理。"""
+
+    def __init__(self, history_dir: str):
+        self.history_dir = history_dir
+        if history_dir and not os.path.exists(history_dir):
+            try:
+                os.makedirs(history_dir)
+            except OSError:
+                pass
+
+    # ── 新建 ──
+    def new_session(self, engine: str = "", model: str = ""):
+        now = time.strftime("%Y-%m-%d %H:%M:%S")
+        return Session(
+            sid=_make_session_id(),
+            title="(新会话)",
+            created_at=now,
+            updated_at=now,
+            engine=engine or "",
+            model=model or "",
+            messages=[],
+        )
+
+    # ── 保存 ──
+    def save(self, session: Session, messages, engine: str = "", model: str = ""):
+        """把当前对话内容写回到 session 对应的文件。"""
+        if session is None:
+            return False
+        session.messages = list(messages or [])
+        if not session.engine and engine:
+            session.engine = engine
+        if not session.model and model:
+            session.model = model
+        # 首条用户消息生成标题（仅当仍是占位标题时）
+        if session.title in ("(新会话)", "(未命名会话)", ""):
+            session.title = _title_from_messages(session.messages)
+        session.updated_at = time.strftime("%Y-%m-%d %H:%M:%S")
+        if not session.created_at:
+            session.created_at = session.updated_at
+        file_path = os.path.join(self.history_dir, f"{session.id}.json")
+        try:
+            with open(file_path, "w", encoding="utf-8") as f:
+                json.dump(session.to_dict(), f, ensure_ascii=False, indent=2, default=str)
+            return True
+        except OSError:
+            return False
+
+    # ── 列出 ──
+    def list_sessions(self) -> list:
+        """返回按更新时间倒序的 Session 列表。"""
+        sessions = []
+        try:
+            files = [f for f in os.listdir(self.history_dir) if f.endswith(".json")]
+        except OSError:
+            return sessions
+        for f in files:
+            path = os.path.join(self.history_dir, f)
+            try:
+                with open(path, "r", encoding="utf-8") as fh:
+                    data = json.load(fh)
+                sid = data.get("id") or f[:-5]
+                s = Session.from_dict(data, fallback_id=sid)
+                # 跳过明显空会话（无消息且非手动命名）
+                sessions.append(s)
+            except (OSError, json.JSONDecodeError, ValueError):
+                continue
+        sessions.sort(key=lambda s: s.updated_at, reverse=True)
+        return sessions
+
+    # ── 按 id 或序号解析 ──
+    def resolve(self, token: str, sessions: list = None) -> Session:
+        """token 可以是会话 id（含文件名去后缀）或列表序号(从1)。"""
+        if sessions is None:
+            sessions = self.list_sessions()
+        token = (token or "").strip()
+        # 序号
+        if token.isdigit():
+            idx = int(token) - 1
+            if 0 <= idx < len(sessions):
+                return sessions[idx]
+        # id（精确或前缀）
+        for s in sessions:
+            if s.id == token or s.id.startswith(token):
+                return s
+        # 文件名（去掉 .json）
+        if token.endswith(".json"):
+            token = token[:-5]
+        for s in sessions:
+            if s.id == token:
+                return s
+        return None
+
+    def load(self, token: str) -> Session:
+        """读取指定会话的完整内容（含 messages）。"""
+        sessions = self.list_sessions()
+        s = self.resolve(token, sessions)
+        if s is None:
+            return None
+        # 重新从文件读取以确保 messages 完整
+        path = os.path.join(self.history_dir, f"{s.id}.json")
+        try:
+            with open(path, "r", encoding="utf-8") as fh:
+                data = json.load(fh)
+            return Session.from_dict(data, fallback_id=s.id)
+        except (OSError, json.JSONDecodeError, ValueError):
+            return s

xcli_core/tool_result.py

@@ -117,3 +117,57 @@ class ToolResult:
     def to_string(self) -> str:
         """转为字符串（兼容旧插件 handle() 返回值）"""
         return str(self)
+
+
+# ── 归一化辅助（跨层兼容：ToolResult / dict / str / None）──
+
+def normalize_tool_text(result: Any, default: str = "无结果") -> str:
+    """把任意工具返回值统一成字符串。
+
+    工具链里三种返回形态都可能出现：
+      - ToolResult（新插件 / UnifiedToolManager.execute）
+      - {"result": ...}（旧插件 / to_dict）
+      - 纯字符串（最老的 handler）
+    统一收敛成字符串，避免上层对 ToolResult 调用 .get() 报
+    "'ToolResult' object has no attribute 'get'"。
+    """
+    if result is None:
+        return default
+    if isinstance(result, ToolResult):
+        text = str(result)
+    elif isinstance(result, dict):
+        text = result.get("result", result.get("content", ""))
+        if isinstance(text, (dict, list)):
+            import json as _json
+            try:
+                text = _json.dumps(text, ensure_ascii=False)
+            except Exception:
+                text = str(text)
+        elif not isinstance(text, str):
+            text = str(text) if text is not None else ""
+    else:
+        text = str(result)
+    text = text.strip() if isinstance(text, str) else str(text)
+    return text if text else default
+
+
+def normalize_tool_dict(result: Any, default: str = "无结果") -> dict:
+    """把任意工具返回值统一成 {"result": str, ...} 结构。
+
+    额外附带 success / error_code / tool_result，方便需要细节的调用方，
+    同时保证 .get('result') 一定可用。
+    """
+    text = normalize_tool_text(result, default)
+    if isinstance(result, ToolResult):
+        return {
+            "result": text,
+            "success": result.success,
+            "error_code": result.error_code,
+            "tool_result": result,
+        }
+    if isinstance(result, dict):
+        out = dict(result)
+        out["result"] = text
+        out.setdefault("success", True)
+        return out
+    return {"result": text, "success": True}

xcli_core/tui.py

@@ -1,6 +1,6 @@
 """TUI 界面模块 - Textual TUI 实现 (动画增强版 + 多行输入)"""
 
-from .constants import TEXTUAL_AVAILABLE
+from .constants import TEXTUAL_AVAILABLE, VERSION
 
 if TEXTUAL_AVAILABLE:
     import re
@@ -11,6 +11,8 @@ if TEXTUAL_AVAILABLE:
     from textual.reactive import var
     from textual.binding import Binding
     from textual import on
+    from .vim_keys import VimInputState
+    import os
     from rich.syntax import Syntax
     from rich.panel import Panel
     from rich.box import ROUNDED
@@ -197,6 +199,28 @@ if TEXTUAL_AVAILABLE:
         transition: color 300ms in_out_cubic;
     }}
 
+    /* ── 文件树 / 会话列表 ── */
+    .file-tree-item {{
+        padding: 0 0 0 1;
+        color: {_Theme.TEXT_MUTED};
+    }}
+    .file-tree-dir {{
+        color: {_Theme.ACCENT};
+        text-style: bold;
+    }}
+    .session-item {{
+        padding: 0 0 0 1;
+        color: {_Theme.TEXT_MUTED};
+    }}
+    .session-item-active {{
+        color: {_Theme.SUCCESS};
+        text-style: bold;
+    }}
+    .plan-badge {{
+        color: {_Theme.WARNING};
+        text-style: bold;
+    }}
+
     /* ── 消息样式 ── */
     .msg-user {{
         color: {_Theme.USER};
@@ -297,6 +321,97 @@ if TEXTUAL_AVAILABLE:
         background: {_Theme.BG};
     }}
 
+    /* ── 自定义顶栏 (opencode 风格) ── */
+    #tui-header {{
+        height: 1;
+        width: 100%;
+        dock: top;
+        background: {_Theme.BG_LIGHT};
+        color: {_Theme.TEXT_MUTED};
+        padding: 0 1;
+        border-bottom: solid {_Theme.BORDER};
+    }}
+    #tui-header .hdr-left {{ color: {_Theme.ACCENT}; text-style: bold; }}
+    #tui-header .hdr-mid {{ color: {_Theme.TEXT_MUTED}; }}
+    #tui-header .hdr-right {{ color: {_Theme.SUCCESS}; }}
+
+    /* ── 侧栏标签页 ── */
+    #sidebar-tabs {{
+        height: 1;
+        width: 100%;
+        dock: top;
+        background: {_Theme.BG};
+        padding: 0;
+    }}
+    .sidebar-tab {{
+        display: block;
+        height: 1;
+        padding: 0 1;
+        color: {_Theme.TEXT_DIM};
+        text-style: bold;
+    }}
+    .sidebar-tab:hover {{
+        color: {_Theme.TEXT_MUTED};
+    }}
+    .sidebar-tab-active {{
+        color: {_Theme.ACCENT};
+        border-bottom: solid {_Theme.ACCENT};
+    }}
+
+    /* ── 侧栏内容面板 ── */
+    #sidebar-content {{
+        height: 1fr;
+        padding: 0 1;
+    }}
+    .panel-section {{
+        padding: 1 0 0 0;
+        border-bottom: solid {_Theme.BORDER};
+    }}
+    .panel-title {{
+        color: {_Theme.TEXT_DIM};
+        padding: 0 0 1 0;
+    }}
+
+    /* ── 消息角色色条 (opencode 风格) ── */
+    .msg-user {{
+        color: {_Theme.USER};
+        padding: 0 0 0 2;
+        border-left: solid {_Theme.USER};
+    }}
+    .msg-ai {{
+        color: {_Theme.AI};
+        padding: 0 0 0 2;
+        border-left: solid {_Theme.GLOW};
+    }}
+    .msg-tool-ok {{
+        color: {_Theme.TOOL};
+        padding: 0 0 0 2;
+        border-left: solid {_Theme.TOOL};
+    }}
+    .msg-tool-err {{
+        color: {_Theme.TOOL_ERR};
+        padding: 0 0 0 2;
+        border-left: solid {_Theme.TOOL_ERR};
+    }}
+    .msg-system {{
+        color: {_Theme.ACCENT};
+        padding: 0 0 0 2;
+        border-left: solid {_Theme.ACCENT};
+    }}
+    .msg-error {{
+        color: {_Theme.ERROR};
+        padding: 0 0 0 2;
+        border-left: solid {_Theme.ERROR};
+    }}
+
+    /* ── 精简欢迎框 ── */
+    .welcome-compact {{
+        color: {_Theme.WELCOME_ACCENT};
+        padding: 1 2;
+        border-bottom: solid {_Theme.BORDER};
+        text-style: bold;
+    }}
+
     """
 
     class _TUIBridge:
@@ -317,6 +432,24 @@ if TEXTUAL_AVAILABLE:
                 return True
             return False
 
+        def sessions(self) -> list:
+            mgr = getattr(self.cli, 'session_manager', None)
+            if mgr and hasattr(mgr, 'list_sessions'):
+                try:
+                    return mgr.list_sessions()
+                except Exception:
+                    return []
+            return []
+
+        def plan_mode(self) -> bool:
+            return bool(getattr(self.cli, 'plan_mode', False))
+
+        def current_plan(self) -> str:
+            return getattr(self.cli, 'current_plan', '') or ""
+
+        def cwd(self) -> str:
+            return os.getcwd()
+
 
     # ═══════════════════════════════════════════════════
     #  动画辅助组件
@@ -379,7 +512,12 @@ if TEXTUAL_AVAILABLE:
             self.styles.opacity = 0
 
         def on_mount(self):
-            self.animate("opacity", value=1.0, duration=0.3, easing="out_cubic")
+            # textual >= 8: Widget.opacity 是只读属性，直接 animate 会抛
+            # "property 'opacity' has no setter"。改为在 styles 上动画，失败则直接到位。
+            try:
+                self.styles.animate("opacity", value=1.0, duration=0.3, easing="out_cubic")
+            except Exception:
+                self.styles.opacity = 1.0
 
 
     class TypingIndicator(Static):
@@ -423,6 +561,125 @@ if TEXTUAL_AVAILABLE:
                 return
             # 其他键正常处理
 
+
+    # ═══════════════════════════════════════════════════
+    #  Vim 风格模态输入框（NORMAL / INSERT）
+    # ═══════════════════════════════════════════════════
+
+    class VimSendTextArea(SendTextArea):
+        """
+        带 Vim 模态编辑的输入框。
+
+        模式：
+          INSERT  —— 默认，正常打字；Esc 进入 NORMAL
+          NORMAL  —— h/j/k/l 移动，i/a/o 进入插入，dd 删行，x 删字，
+                    w/b 跳词，0/$ 行首行尾，gg/G 文首文尾，u 撤销 …
+
+        坐标：VimInputState 用字符偏移 offset，这里负责与 TextArea 的
+        (row, col) 光标互相转换。所有 vim 逻辑都在 vim_keys.py（纯函数，
+        可单测）；本类只做「按键映射 + 把结果写回 TextArea」。
+        """
+        def __init__(self, *args, **kwargs):
+            super().__init__(*args, **kwargs)
+            self.vim_enabled = True
+            self.vim = VimInputState(mode="INSERT")
+
+        # ── offset <-> (row, col) ──
+        def _cur_off(self) -> int:
+            r, c = self.cursor_location
+            text = self.text
+            lines = text.split("\n")
+            return sum(len(l) + 1 for l in lines[:r]) + c
+
+        def _set_off(self, off: int):
+            text = self.text
+            off = max(0, min(off, len(text)))
+            r = text.count("\n", 0, off)
+            line_start = text.rfind("\n", 0, off)
+            c = off - (line_start + 1)
+            self.cursor_location = (r, c)
+
+        # ── key 名字 -> vim token ──
+        @staticmethod
+        def _vim_token(key: str):
+            m = {
+                "escape": "escape", "h": "h", "l": "l", "j": "j", "k": "k",
+                "w": "w", "b": "b", "e": "e", "0": "0", "$": "$", "^": "^",
+                "g": "g", "G": "G", "d": "d", "x": "x", "u": "u",
+                "i": "i", "a": "a", "o": "o", "O": "O", "I": "I", "A": "A",
+                "left": "left", "right": "right", "up": "up", "down": "down",
+                "shift+i": "I", "shift+a": "A", "shift+o": "O",
+                "shift+g": "G", "shift+6": "^", "shift+4": "$",
+            }
+            return m.get(key)
+
+        def _notify_mode(self):
+            app = self.app
+            if hasattr(app, "_update_vim_mode_display"):
+                try:
+                    app._update_vim_mode_display()
+                except Exception:
+                    pass
+
+        def _apply(self, res):
+            try:
+                if res.text != self.text:
+                    self.text = res.text
+                self._set_off(res.offset)
+                self.vim.mode = res.mode
+                self.vim.pending = res.pending
+                if res.action == "undo":
+                    try:
+                        self.undo()
+                    except Exception:
+                        pass
+                self._notify_mode()
+            except Exception:
+                # 任何意外都不应让输入框崩；退回默认行为
+                pass
+
+        async def _on_key(self, event):
+            # 回车：INSERT（或 vim 关闭）发送；NORMAL 下移一行
+            if event.key in ("enter", "\r", "\n"):
+                if self.vim_enabled and self.vim.mode == "NORMAL":
+                    event.prevent_default()
+                    res = self.vim.feed(self.text, self._cur_off(), "j")
+                    self._apply(res)
+                else:
+                    event.prevent_default()
+                    try:
+                        self.app.action_send_message()
+                    except Exception:
+                        pass
+                return
+
+            if not self.vim_enabled:
+                return  # 退化为普通 TextArea
+
+            if self.vim.mode == "INSERT":
+                if event.key == "escape":
+                    event.prevent_default()
+                    res = self.vim.feed(self.text, self._cur_off(), "escape")
+                    self._apply(res)
+                # 其余键交给 TextArea 默认处理（打字）
+                return
+
+            # ── NORMAL 模式 ──
+            if event.key == "shift+enter":
+                token = "o"
+            else:
+                token = self._vim_token(event.key)
+            if token is None:
+                # ctrl+/shift+ 组合键放回去，让 App 级绑定处理（聚焦切换等）
+                if event.key.startswith("ctrl+") or event.key.startswith("shift+"):
+                    return
+                # 普通单键未映射：吞掉，避免 TextArea 在 NORMAL 下误编辑
+                event.prevent_default()
+                return
+            event.prevent_default()
+            res = self.vim.feed(self.text, self._cur_off(), token)
+            self._apply(res)
+
     # ═══════════════════════════════════════════════════
     #  TUI 主应用
     # ═══════════════════════════════════════════════════
@@ -440,11 +697,27 @@ if TEXTUAL_AVAILABLE:
             Binding("ctrl+enter", "send_message", "发送", show=False),
             Binding("shift+enter", "newline", "换行", show=False),
             Binding("f1", "toggle_sidebar", "侧栏", show=True),
+            Binding("ctrl+p", "toggle_plan", "PLAN", show=True),
+            Binding("1", "tab_engine", "引擎", show=False),
+            Binding("2", "tab_files", "文件", show=False),
+            Binding("3", "tab_sessions", "会话", show=False),
+            Binding("4", "tab_status", "状态", show=False),
             Binding("escape", "cancel", "取消", show=False),
+            # ── Vim 风格导航（输入框未聚焦时生效）──
+            Binding("j", "chat_down", "↓聊天", show=False),
+            Binding("k", "chat_up", "↑聊天", show=False),
+            Binding("i", "focus_insert", "插入", show=False),
+            Binding("ctrl+f", "chat_page_down", "翻页↓", show=True),
+            Binding("ctrl+b", "chat_page_up", "翻页↑", show=True),
+            Binding("ctrl+d", "chat_half_down", "半页↓", show=True),
+            Binding("ctrl+u", "chat_half_up", "半页↑", show=True),
+            Binding("ctrl+space", "toggle_focus", "聚焦切换", show=True),
         ]
 
         sidebar_visible = var(True)
         is_generating = var(False)
+        plan_mode = var(False)
+        _sidebar_tab: str = "engine"  # 当前侧栏标签: engine/files/sessions/status
 
         def __init__(self, cli):
             super().__init__()
@@ -454,41 +727,80 @@ if TEXTUAL_AVAILABLE:
             self._typing_widget = None
 
         def compose(self) -> ComposeResult:
-            yield Header(show_clock=True)
+            # ── 自定义顶栏 (opencode 风格: 左项目名 | 中模型 | 右状态) ──
+            yield Static("", id="tui-header")
             with Horizontal(id="app-container"):
                 with Vertical(id="main"):
                     yield VerticalScroll(id="chat-scroll")
                     with Vertical(id="input-area"):
                         with Container(id="input-wrapper"):
-                            yield SendTextArea(
+                            yield VimSendTextArea(
-                                placeholder="输入消息... (回车发送, Shift+回车换行)",
+                                placeholder="输入消息... (回车发送, Shift+回车换行, Esc 进 Vim 普通模式)",
                                 id="user-input",
                                 soft_wrap=True,
                                 tab_behavior="indent",
                             )
                         yield Static(
-                            "  回车发送 | Shift+回车换行 | Ctrl+L 清屏 | F1 侧栏",
+                            "  回车发送 | Shift+回车换行 | Ctrl+L 清屏 | F1 侧栏 | Ctrl+P PLAN",
                             id="input-hint"
                         )
+                # ── 侧栏：标签页 + 内容区 ──
                 with Vertical(id="sidebar"):
-                    yield Static("  引擎", classes="sidebar-header")
+                    # 标签行
-                    yield Vertical(id="engine-list", classes="sidebar-section")
+                    with Horizontal(id="sidebar-tabs"):
-                    yield Rule(line_style="heavy")
+                        yield Static(" 引擎 ", classes="sidebar-tab sidebar-tab-active", id="tab-engine")
-                    yield Static(" 工具", classes="sidebar-header")
+                        yield Static(" 文件 ", classes="sidebar-tab", id="tab-files")
-                    yield Vertical(id="tool-list", classes="sidebar-section")
+                        yield Static(" 会话 ", classes="sidebar-tab", id="tab-sessions")
-                    yield Rule(line_style="heavy")
+                        yield Static(" 状态 ", classes="sidebar-tab", id="tab-status")
-                    yield Static(" 状态", classes="sidebar-header")
+                    # 内容区（根据当前标签切换显示）
-                    yield Vertical(id="status-info", classes="sidebar-section")
+                    with VerticalScroll(id="sidebar-content"):
+                        yield Vertical(id="panel-engine", classes="panel-section")
+                        yield Vertical(id="panel-files", classes="panel-section")
+                        yield Vertical(id="panel-sessions", classes="panel-section")
+                        yield Vertical(id="panel-status", classes="panel-section")
             yield Static(" 就绪 | Ctrl+C 退出", id="status-bar")
 
         def on_mount(self):
-            self._render_welcome_animated()
+            # 初始化顶栏
-            self._update_sidebar()
+            self._update_header()
+            # 初始化侧栏: 默认显示引擎面板,隐藏其余
+            self._switch_sidebar_tab("engine")
+            # 精简欢迎(一行)
+            self._render_welcome_compact()
+            # 初始化 Vim 键位开关（默认开，可被 config.vim_mode 关闭）
+            self._init_vim_mode()
             self.query_one("#user-input").focus()
-            # 延迟显示输入提示（淡入效果）
+            def _show_hint():
-            self.set_timer(1.5, lambda: self.query_one("#input-hint").add_class("visible"))
+                try:
+                    self.query_one("#input-hint").add_class("visible")
+                except Exception:
+                    pass
+            self.set_timer(1.5, _show_hint)
 
-        # ── 动画欢迎界面 ──
+        def _init_vim_mode(self):
+            try:
+                ta = self.query_one("#user-input")
+            except Exception:
+                return
+            cfg = getattr(self.cli, "config", None)
+            enabled = True
+            if isinstance(cfg, dict) and "vim_mode" in cfg:
+                enabled = bool(cfg["vim_mode"])
+            ta.vim_enabled = enabled
+            self._update_vim_mode_display()
+
+        # ── 欢迎界面 ──
+
+        def _render_welcome_compact(self):
+            """opencode 风格: 精简一行欢迎"""
+            scroll = self.query_one("#chat-scroll")
+            engine = self.bridge.current_engine()
+            welcome = Static(
+                f"  ▸ 小狸 Pro-CLI v8.0  │  {engine}  │  "
+                f"{len(self.bridge.tools())} 工具  │  /help 查看命令",
+                classes="welcome-compact"
+            )
+            scroll.mount(welcome)
 
         def _render_welcome_animated(self):
             """逐行动画显示欢迎界面"""
@@ -498,7 +810,7 @@ if TEXTUAL_AVAILABLE:
                 "",
                 "  ╔══════════════════════════════════════════════╗",
                 "  ║                                              ║",
-                "  ║    小狸 Pro-CLI v5.4.1                        ║",
+                "  ║    小狸 Pro-CLI v8.0.3                        ║",
                 "  ║    智能编程助手 · 动画增强版                  ║",
                 "  ║                                              ║",
                 "  ╚══════════════════════════════════════════════╝",
@@ -542,7 +854,7 @@ if TEXTUAL_AVAILABLE:
                 ("", "msg-dim"),
                 ("  ╔══════════════════════════════════════════════╗", "msg-welcome"),
                 ("  ║                                              ║", "msg-welcome"),
-                ("  ║    小狸 Pro-CLI v5.4.1                        ║", "msg-welcome"),
+                ("  ║    小狸 Pro-CLI v8.0.3                        ║", "msg-welcome"),
                 ("  ║    智能编程助手 · 动画增强版                  ║", "msg-welcome"),
                 ("  ║                                              ║", "msg-welcome"),
                 ("  ╚══════════════════════════════════════════════╝", "msg-welcome"),
@@ -559,35 +871,197 @@ if TEXTUAL_AVAILABLE:
         # ── 侧栏更新 ──
 
         def _update_sidebar(self):
-            engine_container = self.query_one("#engine-list")
+            """按当前标签页更新侧栏内容"""
-            engine_container.remove_children()
+            tab = self._sidebar_tab
+
+            # ── 引擎面板 ──
+            engine_panel = self.query_one("#panel-engine")
+            engine_panel.remove_children()
             current = self.bridge.current_engine()
             for name in self.bridge.engines():
                 if name == current:
-                    engine_container.mount(Static(f"  ▸ {name}", classes="engine-item-active"))
+                    engine_panel.mount(Static(f"  ▸ {name}", classes="engine-item-active"))
                 else:
-                    engine_container.mount(Static(f"    {name}", classes="engine-item"))
+                    engine_panel.mount(Static(f"    {name}", classes="engine-item"))
-
+            # 工具列表也放引擎面板下方
-            tool_container = self.query_one("#tool-list")
+            engine_panel.mount(Static("  工具", classes="panel-title"))
-            tool_container.remove_children()
+            for tool in self.bridge.tools()[:12]:
-            for tool in self.bridge.tools():
                 name = tool.get('name', '?')
-                tool_container.mount(Static(f"  • {name}", classes="tool-item"))
+                engine_panel.mount(Static(f"  • {name}", classes="tool-item"))
 
-            status_container = self.query_one("#status-info")
+            # ── 文件树面板 ──
-            status_container.remove_children()
+            self._update_file_tree()
-            status_container.mount(Static(f"  对话: {len(self.bridge.history())} 条", classes="tool-item"))
-            status_container.mount(Static(f"  工具: {len(self.bridge.tools())} 个", classes="tool-item"))
-            status_container.mount(Static(f"  引擎: {len(self.bridge.engines())} 个", classes="tool-item"))
 
-            # 工具数
+            # ── 会话面板 ──
-            if hasattr(self.cli, 'liugin_manager'):
+            self._update_session_list()
-                status_container.mount(Static(f"  工具: {len(self.cli.liugin_manager.tools)} 个", classes="tool-item"))
+
+            # ── 状态面板 ──
+            status_panel = self.query_one("#panel-status")
+            status_panel.remove_children()
+            status_panel.mount(Static(f"  对话: {len(self.bridge.history())} 条", classes="tool-item"))
+            status_panel.mount(Static(f"  工具: {len(self.bridge.tools())} 个", classes="tool-item"))
+            status_panel.mount(Static(f"  引擎: {len(self.bridge.engines())} 个", classes="tool-item"))
+            if self.bridge.plan_mode():
+                status_panel.mount(Static("  模式: 📋 PLAN", classes="plan-badge"))
+            else:
+                status_panel.mount(Static("  模式: 普通", classes="tool-item"))
+
+            # 更新标签高亮
+            self._sync_tab_highlight()
+
+        # ── 文件树 ──
+
+        @staticmethod
+        def _build_file_tree(root, max_depth=2, max_items=40):
+            """生成 (name, is_dir) 列表，用于侧栏文件树展示。"""
+            items = []
+            try:
+                entries = sorted(os.listdir(root))
+            except OSError:
+                return items
+            dirs = [e for e in entries if os.path.isdir(os.path.join(root, e))]
+            files = [e for e in entries if os.path.isfile(os.path.join(root, e))]
+            skip = {'.git', '__pycache__', '.tui_snapshots', '.pytest_cache',
+                    'node_modules', '.workbuddy', 'chat_history', 'memory',
+                    '.idea', '.vscode', '.venv', 'venv'}
+            shown = 0
+            for name in dirs + files:
+                if name in skip or name.startswith('.'):
+                    continue
+                if shown >= max_items:
+                    items.append(("…", False))
+                    break
+                is_dir = os.path.isdir(os.path.join(root, name))
+                items.append((name, is_dir))
+                shown += 1
+            return items
+
+        def _update_file_tree(self):
+            container = self.query_one("#panel-files")
+            container.remove_children()
+            root = self.bridge.cwd()
+            container.mount(Static(f"  📂 {os.path.basename(root)}", classes="file-tree-dir"))
+            for name, is_dir in self._build_file_tree(root):
+                icon = "📁" if is_dir else "📄"
+                container.mount(Static(f"  {icon} {name}", classes="file-tree-item"))
+
+        # ── 会话列表 ──
+
+        @staticmethod
+        def _format_rel_time(ts: str) -> str:
+            if not ts:
+                return ""
+            try:
+                from datetime import datetime
+                dt = datetime.fromisoformat(ts.replace('Z', '+00:00'))
+                diff = (datetime.now() - dt).total_seconds()
+                if diff < 60:
+                    return "刚刚"
+                if diff < 3600:
+                    return f"{int(diff // 60)}分钟前"
+                if diff < 86400:
+                    return f"{int(diff // 3600)}小时前"
+                return f"{int(diff // 86400)}天前"
+            except Exception:
+                return ""
+
+        def _update_session_list(self):
+            container = self.query_one("#panel-sessions")
+            container.remove_children()
+            sessions = self.bridge.sessions()[:8]
+            if not sessions:
+                container.mount(Static("  (无历史会话)", classes="session-item"))
+                return
+            for i, s in enumerate(sessions, 1):
+                title = (s.title or s.id)[:22]
+                t = self._format_rel_time(s.updated_at)
+                container.mount(Static(f"  {i}. {title}  {t}", classes="session-item"))
+
+        # ── 标签页切换 ──
+
+        def _sync_tab_highlight(self):
+            """同步标签高亮状态"""
+            tab_map = {"engine": "tab-engine", "files": "tab-files",
+                       "sessions": "tab-sessions", "status": "tab-status"}
+            for tid in tab_map.values():
+                try:
+                    w = self.query_one(f"#{tid}")
+                    if tid == tab_map.get(self._sidebar_tab, ""):
+                        w.add_class("sidebar-tab-active")
+                    else:
+                        w.remove_class("sidebar-tab-active")
+                except Exception:
+                    pass
+
+        def _switch_sidebar_tab(self, tab: str):
+            """切换侧栏标签页"""
+            self._sidebar_tab = tab
+            # 显示/隐藏面板
+            panel_map = {
+                "engine": "panel-engine", "files": "panel-files",
+                "sessions": "panel-sessions", "status": "panel-status",
+            }
+            for pt, pid in panel_map.items():
+                try:
+                    p = self.query_one(f"#{pid}")
+                    if pt == tab:
+                        p.remove_class("hidden")
+                    else:
+                        p.add_class("hidden")
+                except Exception:
+                    pass
+            self._sync_tab_highlight()
+            self._update_sidebar()
+
+        def action_tab_engine(self):
+            self._switch_sidebar_tab("engine")
+
+        def action_tab_files(self):
+            self._switch_sidebar_tab("files")
+
+        def action_tab_sessions(self):
+            self._switch_sidebar_tab("sessions")
+
+        def action_tab_status(self):
+            self._switch_sidebar_tab("status")
+
+        def _update_header(self):
+            """更新自定义顶栏内容"""
+            hdr = self.query_one("#tui-header", expect_type=Static)
+            engine = self.bridge.current_engine()
+            plan_tag = " [PLAN]" if self.bridge.plan_mode() else ""
+            hist_len = len(self.bridge.history())
+            tok = self._token_str().strip()
+            hdr.update(
+                f" 小狸 v8.0{plan_tag} │ {engine} │ {hist_len}条对话{tok}"
+            )
 
         def _update_status(self, text: str):
             bar = self.query_one("#status-bar")
             engine = self.bridge.current_engine()
-            bar.update(f" {text} | {engine} | {len(self.bridge.history())} 条对话")
+            prefix = "📋 PLAN | " if self.plan_mode else ""
+            bar.update(f" {prefix}{text} | {engine} | {len(self.bridge.history())} 条对话{self._token_str()}")
+
+        def _token_str(self):
+            """token 用量进度（v8.0 token 感知压缩联动，引擎支持时显示）"""
+            e = getattr(self.cli, 'current_engine', None)
+            if not e:
+                return ""
+            used = getattr(e, 'last_prompt_tokens', None)
+            win = None
+            if hasattr(e, 'context_window'):
+                try:
+                    win = e.context_window()
+                except Exception:
+                    win = None
+            if used and win:
+                ratio = min(1.0, used / win)
+                filled = int(ratio * 10)
+                bar_s = "▓" * filled + "░" * (10 - filled)
+                return f" | {used}/{win} {bar_s}"
+            if used:
+                return f" | {used} tok"
+            return ""
 
         # ── 消息追加 ──
 
@@ -743,27 +1217,36 @@ if TEXTUAL_AVAILABLE:
                 'cls': lambda: self.action_clear(),
                 'model': lambda: self._switch_model(args),
                 'engine': lambda: self._switch_model(args),
-                'about': lambda: self._system(" 小狸 Pro-CLI v5.4.1 - 智能编程助手"),
+                'about': lambda: self._system(f" 小狸 Pro-CLI v{VERSION} - 智能编程助手"),
                 'status': lambda: self._show_status(),
                 'tools': lambda: self._show_tools(),
                 'engines': lambda: self._show_engines(),
                 'tui': lambda: self._system("已在 TUI 模式中"),
                 'manual': lambda: self._handle_manual(args),
+                'plan': lambda: self._tui_plan(args),
+                'build': lambda: self._tui_build(),
+                'sessions': lambda: self._tui_sessions(),
+                'session': lambda: self._tui_session(args),
+                'resume': lambda: self._tui_resume(args),
+                'snapshot': lambda: self._save_screenshot(),
+                'screenshot': lambda: self._save_screenshot(),
+                'vim': lambda: self._toggle_vim(),
             }
 
             handler = cmds.get(name)
             if handler:
                 handler()
-            else:
+                return
-                if name in self.cli.liugin_commands:
+            if name in self.cli.liugin_commands:
-                    try:
+                try:
-                        result = self.cli.liugin_commands[name](args)
+                    result = self.cli.liugin_commands[name](args)
-                        if result:
+                    if result:
-                            self._system(result[:500])
+                        self._system(result[:500])
-                    except Exception as e:
+                except Exception as e:
-                        self._error(f"插件命令失败: {e}")
+                    self._error(f"插件命令失败: {e}")
-                else:
+                return
-                    self._error(f"未知命令: /{name}，输入 /help 查看帮助")
+            # 未知命令转发给 cli 完整路由（/agent /providers /chat 等）
+            self._forward_to_cli(cmd)
 
         def _show_help(self):
             help_text = """   命令:
@@ -775,12 +1258,21 @@ if TEXTUAL_AVAILABLE:
   /tools         列出工具
   /status        系统状态
   /clear         清屏
+  /vim           开关 Vim 键位
 
   ⌨  快捷键:
   Ctrl+C     退出        Ctrl+L     清屏
   Ctrl+N     新对话      回车       发送消息
   Shift+回车 换行        F1         侧栏
-  Escape     取消生成    Tab        缩进"""
+  Escape     取消生成    Tab        缩进
+  Ctrl+Space 焦点切换    i          聚焦并输入
+  Ctrl+F/B   翻页下/上   Ctrl+D/U   半页下/上
+
+  VIM 输入(默认开): 输入框内 Esc 进普通模式
+    h j k l  移动    w/b  跳词    0/$  行首/行尾
+    i a o    插入/追加/换行      I A O  行首/行尾/上方
+    dd 删行   x 删字   u 撤销     gg/G 文首/文尾
+    普通模式下 回车 = 下移一行"""
             self._system(help_text)
 
         def _handle_manual(self, args):
@@ -845,6 +1337,84 @@ if TEXTUAL_AVAILABLE:
 
             self._system('\n'.join(status_lines))
 
+        # ── Plan 模式 / 会话 委派 ──
+
+        def _tui_plan(self, args):
+            args = (args or "").strip()
+            if args in ('off', 'exit', '退出', 'cancel'):
+                self._set_plan_mode(False)
+                self._system("已退出 PLAN 模式（计划未执行）")
+                return
+            self._set_plan_mode(True)
+            if args:
+                self._system(f"📋 已进入 PLAN 模式，正在只读调研: {args}")
+                self._user_msg(f"/plan {args}")
+                self._show_thinking()
+                self.is_generating = True
+                self._update_status("思考中 (PLAN)...")
+                self.run_worker(self._generate(args), exclusive=True)
+            else:
+                self._system("📋 已进入 PLAN 模式。描述任务，AI 将只做只读调研并给出实施计划；/build 批准执行，/plan off 取消。")
+
+        def _tui_build(self):
+            if not self.bridge.plan_mode() and not self.bridge.current_plan().strip():
+                self._system("当前不在 PLAN 模式，且无已生成的计划")
+                return
+            self._set_plan_mode(False)
+            self._system("✅ 已批准 PLAN，开始执行。")
+
+        def _tui_sessions(self):
+            sessions = self.bridge.sessions()
+            if not sessions:
+                self._system("  无历史会话")
+                return
+            lines = ["  历史会话:"]
+            for i, s in enumerate(sessions[:10], 1):
+                title = (s.title or s.id)[:30]
+                t = self._format_rel_time(s.updated_at)
+                lines.append(f"  {i}. {title}  ({t})")
+            self._system('\n'.join(lines))
+
+        def _tui_session(self, args):
+            args = (args or "").strip()
+            if not args:
+                self._tui_sessions()
+                return
+            if args == 'new':
+                self.action_new_chat()
+                return
+            self._forward_to_cli(f"/resume {args}")
+
+        def _tui_resume(self, args):
+            self._forward_to_cli(f"/resume {args}")
+
+        def _forward_to_cli(self, cmd: str):
+            """把未知/委派命令交给 cli 完整路由处理。"""
+            try:
+                self.cli.process_conversation(cmd)
+            except Exception as e:
+                self._error(f"命令执行失败: {e}")
+            self.call_after_refresh(self._update_sidebar)
+            self.call_after_refresh(self._sync_plan)
+
+        def _set_plan_mode(self, on: bool):
+            self.plan_mode = on
+            try:
+                self.cli.plan_mode = on
+                if not on:
+                    self.cli.current_plan = ""
+            except Exception:
+                pass
+            self._update_sidebar()
+            self._update_status("📋 PLAN" if on else "就绪")
+
+        def _sync_plan(self):
+            """从 cli 同步 plan 状态到 TUI 反应变量。"""
+            remote = self.bridge.plan_mode()
+            if remote != self.plan_mode:
+                self.plan_mode = remote
+                self._update_sidebar()
+
         # ── AI 生成 ──
 
         async def _generate(self, user_input):
@@ -874,8 +1444,10 @@ if TEXTUAL_AVAILABLE:
                 self.call_after_refresh(self._remove_thinking)
                 self.call_after_refresh(self._remove_typing_indicator)
                 self.is_generating = False
+                self.call_after_refresh(self._sync_plan)
                 self.call_after_refresh(lambda: self._update_status("就绪"))
                 self.call_after_refresh(self._update_sidebar)
+                self.call_after_refresh(self._update_header)
 
         def _write_raw(self, msg):
             if '✅' in msg or 'OK 工具' in msg:
@@ -911,6 +1483,10 @@ if TEXTUAL_AVAILABLE:
                 sidebar.add_class("hidden")
                 self.sidebar_visible = False
 
+        def action_toggle_plan(self):
+            self._set_plan_mode(not self.plan_mode)
+            self._system("📋 PLAN 模式: " + ("开" if self.plan_mode else "关"))
+
         def action_cancel(self):
             if self.is_generating:
                 self._remove_thinking()
@@ -918,6 +1494,107 @@ if TEXTUAL_AVAILABLE:
                 self._system("已取消")
                 self.is_generating = False
 
+        # ── Vim 风格：聊天区滚动 / 聚焦切换 ──
+        def _user_input(self):
+            return self.query_one("#user-input")
+
+        def _chat_scroll(self):
+            return self.query_one("#chat-scroll")
+
+        def action_toggle_focus(self):
+            """Ctrl+Space：在输入框与聊天区之间切换焦点。"""
+            ta = self._user_input()
+            if ta.has_focus:
+                self.screen.set_focus(None)
+            else:
+                ta.focus()
+                if getattr(ta, "vim_enabled", False):
+                    ta.vim.mode = "INSERT"
+                self._update_vim_mode_display()
+
+        def action_focus_insert(self):
+            """i：输入框未聚焦时聚焦并进入 INSERT（模仿 vim 按 i 开始输入）。"""
+            ta = self._user_input()
+            if ta.has_focus:
+                return
+            ta.focus()
+            if getattr(ta, "vim_enabled", False):
+                ta.vim.mode = "INSERT"
+            self._update_vim_mode_display()
+
+        def _guard_chat(self):
+            """聊天导航仅在输入框未聚焦时生效，避免与输入冲突。"""
+            return self._user_input().has_focus
+
+        def action_chat_down(self):
+            if self._guard_chat():
+                return
+            self._chat_scroll().scroll_down(3)
+
+        def action_chat_up(self):
+            if self._guard_chat():
+                return
+            self._chat_scroll().scroll_up(3)
+
+        def action_chat_page_down(self):
+            if self._guard_chat():
+                return
+            self._chat_scroll().scroll_page_down()
+
+        def action_chat_page_up(self):
+            if self._guard_chat():
+                return
+            self._chat_scroll().scroll_page_up()
+
+        def action_chat_half_down(self):
+            if self._guard_chat():
+                return
+            self._chat_scroll().scroll_relative(y=20)
+
+        def action_chat_half_up(self):
+            if self._guard_chat():
+                return
+            self._chat_scroll().scroll_relative(y=-20)
+
+        def _update_vim_mode_display(self):
+            """在输入框下方提示栏显示当前 Vim 模式。"""
+            try:
+                hint = self.query_one("#input-hint")
+                ta = self._user_input()
+            except Exception:
+                return
+            if not getattr(ta, "vim_enabled", False):
+                hint.update("  回车发送 | Shift+回车换行 | Ctrl+L 清屏 | F1 侧栏 | Ctrl+P PLAN")
+                return
+            mode = getattr(getattr(ta, "vim", None), "mode", "INSERT")
+            tag = "-- INSERT --" if mode == "INSERT" else "-- NORMAL --"
+            hint.update(f"  {tag}  │  Esc 切模式  │  i插入 a追加 o换行  hjkl移动  dd删行 x删字  u撤销")
+
+        def _toggle_vim(self):
+            ta = self._user_input()
+            ta.vim_enabled = not getattr(ta, "vim_enabled", True)
+            if ta.vim_enabled:
+                self._system("Vim 键位: 开 (Esc 在 NORMAL/INSERT 间切换)")
+            else:
+                self._system("Vim 键位: 关 (普通输入)")
+            self._update_vim_mode_display()
+
+        def _save_screenshot(self):
+            """导出当前 TUI 屏幕截图（SVG）——AI 可据此自验实时渲染（/snapshot）"""
+            import os
+            out_dir = os.path.join(
+                os.path.dirname(os.path.dirname(os.path.abspath(__file__))), ".tui_snapshots"
+            )
+            os.makedirs(out_dir, exist_ok=True)
+            path = os.path.join(out_dir, "live.svg")
+            try:
+                svg = self.export_screenshot()
+                with open(path, "w", encoding="utf-8") as f:
+                    f.write(svg)
+                self._system(f"截图已保存: {path}")
+            except Exception as e:
+                self._error(f"截图失败: {e}")
+
 else:
     # Textual 不可用时的占位类
     class XiaoliTUI:

xcli_core/verbose.py

@@ -0,0 +1,24 @@
+"""
+启动详情开关
+
+默认启动只报结论（加载了几个引擎/几个工具），不再逐条刷屏。
+需要看每个插件/技能的加载明细时，加 --verbose（或 -v），
+也可以设环境变量 XCLI_VERBOSE=1。
+"""
+import os
+import sys
+
+_TRUTHY = ('1', 'true', 'yes', 'on')
+
+
+def is_verbose() -> bool:
+    """是否输出启动明细"""
+    if any(a in ('--verbose', '-v') for a in sys.argv[1:]):
+        return True
+    return str(os.environ.get('XCLI_VERBOSE', '')).lower() in _TRUTHY
+
+
+def vprint(*args, **kwargs):
+    """仅在 --verbose 时打印的 print"""
+    if is_verbose():
+        print(*args, **kwargs)

xcli_core/vim_keys.py

@@ -0,0 +1,275 @@
+"""
+Vim 风格输入编辑状态机（纯逻辑，不依赖 textual）
+================================================
+把“模态编辑”的核心（NORMAL / INSERT 双模式 + 各种 motion / operator）抽成
+纯函数 + 一个小状态类，便于在不依赖 textual 的环境里做单元测试。
+
+widget 侧（tui.py 的 VimSendTextArea）只负责：
+- 把 textual 的 key 名字映射成这里的 token（见 tui.py 的 _vim_token）
+- 调用 VimInputState.feed(text, offset, token) 拿到 EditResult
+- 把 EditResult 应用回 TextArea（设 text / 设光标 / 触发 undo 等）
+
+坐标约定：文本用单一字符串表示，光标用「字符偏移 offset」表示（从 0 开始）。
+widget 负责 offset <-> (row, col) 的转换。
+"""
+
+from dataclasses import dataclass
+from typing import Optional
+
+
+@dataclass
+class EditResult:
+    """一次 vim 动作的执行结果。"""
+    text: str                       # 应用后的文本
+    offset: int                     # 应用后的光标偏移
+    mode: str                       # 结束时的模式: "INSERT" / "NORMAL"
+    pending: Optional[str]          # 多键序列的待定前缀: "d" / "g" / None
+    handled: bool = True            # False 表示“交给输入框默认处理”（仅 INSERT 非 Esc 键）
+    action: Optional[str] = None    # 额外指令: "undo"
+    message: str = ""               # 状态栏提示语
+
+
+# ── 文本/坐标辅助 ──────────────────────────────────────────────────────────
+
+def _line_start_off(text: str, off: int) -> int:
+    """当前行行首偏移。"""
+    s = text.rfind("\n", 0, off)
+    return s + 1
+
+
+def _line_end_off(text: str, off: int) -> int:
+    """当前行行尾（不含换行符）偏移。"""
+    e = text.find("\n", off)
+    return e if e != -1 else len(text)
+
+
+def _first_nonblank_off(text: str, off: int) -> int:
+    """当前行第一个非空白字符偏移，若整行空白则回到行首。"""
+    s = _line_start_off(text, off)
+    e = _line_end_off(text, off)
+    i = s
+    while i < e and text[i].isspace():
+        i += 1
+    return i
+
+
+def _word_forward_off(text: str, off: int) -> int:
+    """跳到下一个 word 的词首（vim 的 w）：越过当前词 + 词间空白。"""
+    n = len(text)
+    i = off
+    # 越过错当前空白
+    while i < n and text[i].isspace():
+        i += 1
+    # 越过当前 word
+    if i < n and (text[i].isalnum() or text[i] == "_"):
+        while i < n and (text[i].isalnum() or text[i] == "_"):
+            i += 1
+    else:
+        while i < n and not text[i].isspace() and not (text[i].isalnum() or text[i] == "_"):
+            i += 1
+    # 越过词间空白，落到下一词首
+    while i < n and text[i].isspace():
+        i += 1
+    return i
+
+
+def _word_back_off(text: str, off: int) -> int:
+    """跳到上一个 word 的词首（vim 的 b）。"""
+    i = off
+    # 越过光标前空白
+    while i > 0 and text[i - 1].isspace():
+        i -= 1
+    # 越过当前 word
+    if i > 0 and (text[i - 1].isalnum() or text[i - 1] == "_"):
+        while i > 0 and (text[i - 1].isalnum() or text[i - 1] == "_"):
+            i -= 1
+    else:
+        while i > 0 and not text[i - 1].isspace() and not (text[i - 1].isalnum() or text[i - 1] == "_"):
+            i -= 1
+    # 越过词间空白，落到上一词首
+    while i > 0 and text[i - 1].isspace():
+        i -= 1
+    return i
+
+
+def _offset_to_rc(text: str, off: int):
+    off = max(0, min(off, len(text)))
+    row = text.count("\n", 0, off)
+    line_start = text.rfind("\n", 0, off)
+    col = off - (line_start + 1)
+    return row, col
+
+
+def _rc_to_offset(text: str, row: int, col: int) -> int:
+    lines = text.split("\n")
+    row = max(0, min(row, len(lines) - 1))
+    line = lines[row]
+    col = max(0, min(col, len(line)))
+    off = 0
+    for l in lines[:row]:
+        off += len(l) + 1
+    return off + col
+
+
+# ── 状态机 ────────────────────────────────────────────────────────────────
+
+class VimInputState:
+    """
+    Vim 输入状态机。
+
+    mode:   "INSERT" 或 "NORMAL"
+    pending: 多键序列待定前缀（"d" / "g"），单键动作时为 None
+
+    用法：
+        st = VimInputState(mode="INSERT")   # 输入框默认进入 INSERT
+        res = st.feed(text, offset, token)
+        # 把 res 应用回输入框
+    """
+
+    def __init__(self, mode: str = "INSERT"):
+        self.mode = mode
+        self.pending: Optional[str] = None
+
+    # 进入 INSERT 时返回的结果构造器
+    @staticmethod
+    def _insert(text, offset, msg) -> EditResult:
+        return EditResult(text, offset, "INSERT", None, message=msg)
+
+    def feed(self, text: str, offset: int, token: str) -> EditResult:
+        """
+        token 是已经映射好的 vim 键名，例如:
+        "h","l","j","k","w","b","e","0","$","^","g","G","d","x","u",
+        "i","a","o","O","I","A","escape","left","right","up","down"
+        """
+        if self.mode == "INSERT":
+            if token == "escape":
+                # 回到 NORMAL，光标左移一格（vim 习惯：Esc 后光标退到本字上）
+                new_off = offset
+                if offset > 0 and text[offset - 1] != "\n":
+                    new_off = offset - 1
+                self.mode = "NORMAL"
+                self.pending = None
+                return EditResult(text, new_off, "NORMAL", None, message="-- NORMAL --")
+            # 其余键交给 TextArea 默认处理（打字）
+            return EditResult(text, offset, "INSERT", None, handled=False)
+
+        # ── NORMAL 模式 ──
+        # 多键序列待定前缀
+        if self.pending == "d":
+            self.pending = None
+            if token == "d":
+                return self._delete_line(text, offset, "dd")
+            if token == "w":
+                return self._delete_word(text, offset, "dw")
+            if token == "$":
+                return self._delete_to_eol(text, offset, "d$")
+            # 非法组合：清掉 pending 后把当前键当单键重新处理
+            # （落到下面的单键逻辑）
+
+        if self.pending == "g":
+            self.pending = None
+            if token == "g":
+                return EditResult(text, 0, "NORMAL", None, message="gg")
+            # 其余：当作普通单键处理
+
+        # ── 单键 ──
+        if token in ("i",):
+            self.mode = "INSERT"
+            return self._insert(text, offset, "-- INSERT --")
+        if token == "I":
+            self.mode = "INSERT"
+            return self._insert(text, _first_nonblank_off(text, offset), "-- INSERT (行首) --")
+        if token == "a":
+            self.mode = "INSERT"
+            return self._insert(text, min(offset + 1, len(text)), "-- INSERT (追加) --")
+        if token == "A":
+            self.mode = "INSERT"
+            return self._insert(text, _line_end_off(text, offset), "-- INSERT (行尾) --")
+        if token == "o":
+            self.mode = "INSERT"
+            e = _line_end_off(text, offset)
+            new_text = text[:e] + "\n" + text[e:]
+            return self._insert(new_text, e + 1, "-- INSERT (新行) --")
+        if token == "O":
+            self.mode = "INSERT"
+            s = _line_start_off(text, offset)
+            new_text = text[:s] + "\n" + text[s:]
+            return self._insert(new_text, s, "-- INSERT (上方新行) --")
+
+        if token in ("h", "left"):
+            new_off = max(_line_start_off(text, offset), offset - 1)
+            return EditResult(text, new_off, "NORMAL", None)
+        if token in ("l", "right"):
+            le = _line_end_off(text, offset)
+            new_off = min(le, offset + 1) if offset < len(text) else offset
+            return EditResult(text, new_off, "NORMAL", None)
+        if token in ("j", "down"):
+            r, c = _offset_to_rc(text, offset)
+            return EditResult(text, _rc_to_offset(text, r + 1, c), "NORMAL", None)
+        if token in ("k", "up"):
+            r, c = _offset_to_rc(text, offset)
+            return EditResult(text, _rc_to_offset(text, r - 1, c), "NORMAL", None)
+        if token == "w":
+            return EditResult(text, _word_forward_off(text, offset), "NORMAL", None)
+        if token == "b":
+            return EditResult(text, _word_back_off(text, offset), "NORMAL", None)
+        if token == "e":
+            i = _word_forward_off(text, offset)
+            new_off = max(offset, i - 1) if i > offset else offset
+            return EditResult(text, new_off, "NORMAL", None)
+        if token == "0":
+            return EditResult(text, _line_start_off(text, offset), "NORMAL", None)
+        if token == "^":
+            return EditResult(text, _first_nonblank_off(text, offset), "NORMAL", None)
+        if token == "$":
+            return EditResult(text, _line_end_off(text, offset), "NORMAL", None)
+        if token == "G":
+            r, c = _offset_to_rc(text, offset)
+            last_row = text.count("\n")
+            return EditResult(text, _rc_to_offset(text, last_row, c), "NORMAL", None)
+        if token == "g":
+            self.pending = "g"
+            return EditResult(text, offset, "NORMAL", "g", message="g")
+        if token == "d":
+            self.pending = "d"
+            return EditResult(text, offset, "NORMAL", "d", message="d")
+        if token == "x":
+            if offset < len(text):
+                new_text = text[:offset] + text[offset + 1:]
+                return EditResult(new_text, offset, "NORMAL", None, message="x")
+            return EditResult(text, offset, "NORMAL", None)
+        if token == "u":
+            return EditResult(text, offset, "NORMAL", None, action="undo", message="u")
+        if token == "escape":
+            return EditResult(text, offset, "NORMAL", None, message="-- NORMAL --")
+
+        # 未知键：忽略（保持 NORMAL）
+        return EditResult(text, offset, "NORMAL", None)
+
+    # ── operator 实现 ──
+    def _delete_line(self, text, offset, msg) -> EditResult:
+        s = _line_start_off(text, offset)
+        e = _line_end_off(text, offset)
+        if e < len(text):
+            e2 = e + 1          # 连同行尾换行一起删
+        else:
+            e2 = e
+            if s > 0:
+                s = s - 1       # 最后一行：删掉上一行的换行
+        new_text = text[:s] + text[e2:]
+        new_off = min(s, len(new_text))
+        return EditResult(new_text, new_off, "NORMAL", None, message=msg)
+
+    def _delete_word(self, text, offset, msg) -> EditResult:
+        s = offset
+        e = _word_forward_off(text, offset)
+        new_text = text[:s] + text[e:]
+        new_off = min(s, len(new_text))
+        return EditResult(new_text, new_off, "NORMAL", None, message=msg)
+
+    def _delete_to_eol(self, text, offset, msg) -> EditResult:
+        s = offset
+        e = _line_end_off(text, offset)
+        new_text = text[:s] + text[e:]
+        new_off = min(s, len(new_text))
+        return EditResult(new_text, new_off, "NORMAL", None, message=msg)