codebee 0.1.32 → 0.1.34

package/CHANGELOG.md

@@ -6,6 +6,29 @@ README 元数据带回，供老版本在「发现新版本」时展示新版更
 
 ## 未发布
 
+## v0.1.34（2026-09-21）
+
+### 🛠 问题修复
+
+- 手机端操作被「其他设备正在控制」挡住时，现在会弹出「接管并重试」确认框——点一下就接管控制权并自动重做刚才的操作，不再连点全是失败。
+- 微信群列表拉取失败时把读取器的真实报错带出来（依赖缺失/解释器位数不对/微信未登录一眼可辨），不再只说「无有效输出」。
+- 连载「章节评审」现在在侧栏进入的任务级详情也会显示，不再只有打开某一次运行时才能看到。
+- 新建任务会一并记住上次的编排模式与思考程度；非法历史值会丢弃，不会污染表单。
+- 修复扫描 Python 后尚未保存时被后台轮询冲回旧路径；现在点选或自动选中的候选都会保持到用户保存。
+
+## v0.1.33（2026-09-21）
+
+### ✨ 新功能
+
+- 创建任务支持快速/自动/专家/手动四档编排，以及独立的自动/快速/标准/深度思考程度；所有预置类型按风险动态决定规划、评审、修复和换将步骤。
+- 对话默认使用系统推荐，也可仅为本次对话指定厂商或模型；指定渠道不可用时明确失败，不修改全局 CLI 绑定。
+- 创建前显示同类历史或流程基线 ETA/P90，运行中显示剩余时间与超出预估状态；满载时不排队，异常退出的运行自动收口为失败。
+
+### 🛠 问题修复
+
+- 修复 Windows GBK 控制台因特殊符号导致服务启动崩溃；Aider 启动前增加 Git 仓库可读性诊断，避免损坏仓库长时间卡住。
+- 修复 Anthropic 兼容接口误收到 OpenAI `reasoning_effort` 字段；修复语言切换覆盖端口标题导致桌宠重复开窗；过滤 CLI 历史和评审中间文件，成果页只展示可交付产物。
+
 ## v0.1.32（2026-09-21）
 
 ### ✨ 新功能

package/README.md

@@ -21,8 +21,30 @@ Kimi Code、MiMo Code、Grok Build、Pi、DeepSeek Harness……），提供
 不会把你的任务内容交给任何第三方。
 
 <!-- relnotes:start -->
+### 🛠 问题修复 · Fixes
+
+- 手机端操作被「其他设备正在控制」挡住时，弹出「接管并重试」——一键接管并自动重做刚才的操作。
+- When a mobile action is blocked by another device in control, a "Take over & retry" prompt appears — one tap retakes control and redoes the action.
+
+- 微信群列表拉取失败时会显示读取器的真实报错（依赖缺失/位数不对/微信未登录一眼可辨）。
+- WeChat group-list failures now surface the reader's real error (missing dependency, wrong bitness, or WeChat not logged in).
+
+- 连载章节评审在任务级详情与单次运行详情都可见；新建任务会记住上次的编排模式和思考程度。
+- Chapter reviews now appear in both task and run details; new tasks also remember the last orchestration mode and reasoning level.
+
+- 扫描 Python 后即使保存前遇到后台刷新，刚选中的路径也不会再被冲掉。
+- A Python selected by Scan now survives background refreshes even before settings are saved.
+
 ### ✨ 新功能 · New
 
+- 创建任务支持快速/自动/专家/手动编排与独立思考程度；代码、写作、翻译、邮件、汇报、调研等全类型按风险动态选择步骤。
+- 新建对话默认系统推荐，也可只对当前对话指定厂商/模型；创建前和运行中显示 ETA/P90 与剩余时间。
+- New tasks support Fast/Auto/Expert/Manual orchestration and an independent reasoning level; every task type adapts its workflow to risk.
+- Conversations use the system recommendation by default, with optional per-conversation provider/model selection plus ETA/P90 visibility.
+
+- 修复 Windows 控制台启动崩溃、损坏 Git 仓库导致 Aider 长时间等待、Anthropic 接口收到不兼容思考字段，以及桌宠语言切换后重复开窗。
+- Fixed Windows console startup crashes, Aider stalls on unreadable Git repositories, incompatible Anthropic reasoning fields, and duplicate pet windows after language changes.
+
 - 连载任务详情新增「章节评审」卡片：每章维度评分（低分标红）与达标状态，大纲要点逐项核对结果红绿灰直读。
 - Serial runs now show a Chapter Reviews card: per-chapter dimension scores (low ones in red) plus beat-by-beat checks color-coded green/red/gray.
 

package/app/core/prefs.py

@@ -14,11 +14,15 @@ from . import paths
 
 _LOCK = threading.Lock()
 
-SAVE_KEYS = ("type", "rounds", "threshold", "best_of", "chapters",
-             "words_per_chapter", "variants")
-_LIMITS = {"rounds": (1, 5), "threshold": (1.0, 10.0), "best_of": (1, 3),
-           "chapters": (1, 2000), "words_per_chapter": (200, 20000),
-           "variants": (1, 3)}
+SAVE_KEYS = ("type", "rounds", "threshold", "best_of", "chapters",
+             "words_per_chapter", "variants", "mode", "thinking")
+_LIMITS = {"rounds": (1, 5), "threshold": (1.0, 10.0), "best_of": (1, 3),
+           "chapters": (1, 2000), "words_per_chapter": (200, 20000),
+           "variants": (1, 3)}
+_CHOICES = {
+    "mode": ("auto", "fast", "expert", "manual"),
+    "thinking": ("auto", "low", "standard", "high"),
+}
 
 
 def _file():
@@ -47,12 +51,17 @@ def record(payload):
             v = payload.get(key)
             if v is None:
                 continue
-            if key == "type":
-                v = str(v)[:32]
-                if v:
-                    cur["type"] = v
-                continue
-            c = _clamp(key, v)
+            if key == "type":
+                v = str(v)[:32]
+                if v:
+                    cur["type"] = v
+                continue
+            if key in _CHOICES:
+                v = str(v).strip().lower()
+                if v in _CHOICES[key]:
+                    cur[key] = v
+                continue
+            c = _clamp(key, v)
             if c is not None:
                 cur[key] = c
         tmp = _file().with_suffix(".tmp")

package/app/core/wxdigest.py

@@ -131,17 +131,25 @@ def refresh_groups():
     _DIR.mkdir(parents=True, exist_ok=True)
     rf = _DIR / "groups_cache.json"
     cmd = [rp, str(READER_SCRIPT), "--list-groups", "--result", str(rf)]
+    stderr = b""
     try:
-        subprocess.run(cmd, timeout=READER_TIMEOUT, capture_output=True,
-                       creationflags=CREATE_NO_WINDOW if os.name == "nt" else 0)
+        r = subprocess.run(cmd, timeout=READER_TIMEOUT, capture_output=True,
+                           creationflags=CREATE_NO_WINDOW if os.name == "nt" else 0)
+        stderr = r.stderr or b""
     except subprocess.TimeoutExpired:
         raise RuntimeError("读取群列表超时（%ds）" % READER_TIMEOUT)
     try:
         res = json.loads(rf.read_text(encoding="utf-8"))
     except Exception:
-        raise RuntimeError("群列表无有效输出（检查 reader_python 环境）")
+        # 带上 sidecar stderr 尾部：依赖缺失/解释器位数不对/脚本异常的真因
+        # 不再被「无有效输出」一句吞掉（用户实测排障反馈）
+        detail = (stderr or b"").decode("utf-8", "replace").strip()[-300:]
+        raise RuntimeError("群列表无有效输出（检查 reader_python 环境）%s"
+                           % ("：" + detail if detail else ""))
     if not res.get("ok"):
-        raise RuntimeError("读取群列表失败")
+        errs = "；".join(str(x) for x in (res.get("errors") or [])) or \
+               str(res.get("error") or "未知错误")
+        raise RuntimeError("读取群列表失败：%s" % errs[:300])
     groups = res.get("groups") or []
     with _LOCK:
         _STATE["groups_cache"] = groups

package/app/ui/app.js

@@ -10,7 +10,7 @@ function codebeeDocumentTitle(lang) {
 document.title = codebeeDocumentTitle((localStorage.getItem("orch.lang") || "zh").toLowerCase());
 
 const $ = (id) => document.getElementById(id);
-const S = { state: null, catalog: null, catSig: "", providers: null, bindings: null, modelsSig: "", bindSig: "", tab: "tasks", detailRunId: null, pollTimer: null, showArchived: false, /* 会话内开关：每次加载默认隐藏已归档、图标不选中（不持久化，见 btn-side-arch） */ selProvs: {}, selModels: {}, selRuns: {}, bindSel: {}, catalogChecking: false, updateCheckAt: 0, control: null, sseLive: false, es: null, flows: null, orch: null, skills: null, orchSig: "", settings: null, sessionAgents: new Set(), atts: [], gitInfo: null, gitWb: null, gitWbKey: "", gitWbAt: 0, gitWbBusy: false, creatingTask: false, inspKey: null, inspData: null, inspSig: "", inspAt: 0, inspTab: "git", inspAutoSig: "", rdTab: null, rdTabSig: "", rdTabPin: false, _rdCtx: {}, lastPrefs: null };
+const S = { state: null, catalog: null, catSig: "", providers: null, bindings: null, modelsSig: "", bindSig: "", tab: "tasks", detailRunId: null, pollTimer: null, showArchived: false, /* 会话内开关：每次加载默认隐藏已归档、图标不选中（不持久化，见 btn-side-arch） */ selProvs: {}, selModels: {}, selRuns: {}, bindSel: {}, catalogChecking: false, updateCheckAt: 0, control: null, sseLive: false, es: null, flows: null, orch: null, skills: null, orchSig: "", settings: null, sessionAgents: new Set(), atts: [], gitInfo: null, gitWb: null, gitWbKey: "", gitWbAt: 0, gitWbBusy: false, creatingTask: false, inspKey: null, inspData: null, inspSig: "", inspAt: 0, inspTab: "git", inspAutoSig: "", rdTab: null, rdTabSig: "", rdTabPin: false, _rdCtx: {}, lastPrefs: null, prefsApplied: false };
 
 /* ---------------------------------------------------------- 任务类型（流程） */
 async function loadFlows() {
@@ -65,10 +65,22 @@ function renderTypeOptions() {
   // 首屏默认落在「直接执行」（快档位：单 CLI 直达，无拆解/评审）；
   // 老数据或该流程被删时保持首项，不硬造一个不存在的值
   else if (flowById("direct")) sel.value = "direct";
+  applyTaskPreferenceDefaults();
   renderTypeMenu();
   onTypeChange();
 }
 
+function applyTaskPreferenceDefaults() {
+  if (S.prefsApplied || !S.lastPrefs) return;
+  const lp = S.lastPrefs;
+  const mode = $("f-mode"), thinking = $("f-thinking");
+  if (mode && ["auto", "fast", "expert", "manual"].includes(lp.mode)) mode.value = lp.mode;
+  if (thinking && ["auto", "low", "standard", "high"].includes(lp.thinking)) thinking.value = lp.thinking;
+  const manual = $("f-manual-only");
+  if (manual && mode) manual.classList.toggle("hidden", mode.value !== "manual");
+  S.prefsApplied = true;
+}
+
 /* 类型选择的可见层：带黑白图标的按钮 + 弹出菜单。原生 <option> 渲染不了 SVG，
  * 所以隐藏 select 只当值真源（既有联动/测试全部照旧），外观全靠这层。 */
 function renderTypeMenu() {
@@ -590,6 +602,7 @@ async function api(path, opts) {
   delete fetchOpts.busy;
   delete fetchOpts.busyElement;
   delete fetchOpts.operation;
+  delete fetchOpts._ctrlRetry;
   try {
     let res;
     try {
@@ -602,7 +615,27 @@ async function api(path, opts) {
     if (res.status === 401) { showTokenGate(t("令牌不正确或已更换，请重新输入")); throw new Error(t("需要访问令牌")); }
     let data = null;
     try { data = await res.json(); } catch (e) { /* ignore */ }
-    if (res.status === 423 && data && data.control) setControl(data.control);
+    if (res.status === 423 && data && data.control) {
+      setControl(data.control);
+      // 手机端接管闭环（用户实测反馈：423 连点全失败不知道怎么办）：
+      // 他人持有时弹「接管并重试」确认，确认后抢夺控制权并自动重试一次。
+      // uiConfirm 弹窗互斥防堆叠；重试标记防递归；GET 轮询不会走到这里。
+      const holder = data.control.holder || "";
+      if (!opts._ctrlRetry && !opts.silent && !api._ctrlPrompting && holder) {
+        api._ctrlPrompting = true;
+        try {
+          if (await uiConfirm(t("「{0}」正在控制，接管后自动重试「{1}」？",
+            holder, operationLabel(path, { ...opts, operation: true }) || t("提交操作")),
+            { ok: t("接管并重试") })) {
+            await ctrlAction("acquire", true);
+            api._ctrlPrompting = false;
+            return api(path, Object.assign({}, opts, { _ctrlRetry: true }));
+          }
+        } finally {
+          api._ctrlPrompting = false;
+        }
+      }
+    }
     if (!res.ok) throw new Error((data && data.error) || ("HTTP " + res.status));
     operationFinish(operationToken, "done");
     return data;
@@ -3672,6 +3705,7 @@ function drawTaskDetail(key, runs) {
     (latest.error ? '<span class="stat err">' + errTag(latest.error) + esc(latest.error.slice(0, 200)) + "</span>" : "");
   $("rd-plan").classList.add("hidden");
   renderRouting(latest);
+  renderChapterScores(latest);
   let html = "";
   ordered.forEach((r, i) => {
     const runNo = runs.length - i;
@@ -11733,6 +11767,7 @@ function beePyRender(list) {
     b.addEventListener("click", () => {
       const p = list[Number(b.dataset.i)];
       $("bee-reader-py").value = p.path;
+      Bee.pyDirty = true;
       box.querySelectorAll(".bee-py-item").forEach((x) => x.classList.remove("on"));
       b.classList.add("on");
       beePyInstallVisibility(p);
@@ -11744,6 +11779,7 @@ function beePyRender(list) {
     list.find((p) => p.bits >= 64);
   if (hit) {
     $("bee-reader-py").value = hit.path;
+    if (hit.path !== cur) Bee.pyDirty = true;
     beePyInstallVisibility(hit);
     const btn = box.querySelector('.bee-py-item[data-i="' + list.indexOf(hit) + '"]');
     if (btn) btn.classList.add("on");

package/app/ui/i18n.js

@@ -409,6 +409,8 @@
     "%（失败 ": "% (failed ",
     "「": "\"",
     "」将变为只读。": "\" will become read-only.",
+    "「{0}」正在控制，接管后自动重试「{1}」？": "\"{0}\" is in control. Take over and retry \"{1}\"?",
+    "接管并重试": "Take over & retry",
     "」恢复为内置默认配置？": "\" restored to built-in defaults?",
     "{0} 正在控制（点击接管）": "{0} is in control (click to take over)",
     "」？已有任务不受影响。": "\"? Existing tasks are not affected.",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codebee",
-  "version": "0.1.32",
+  "version": "0.1.34",
   "description": "CodeBee · 多智能体编排台 — 统一编排本机 AI 编码 CLI（Codex / Claude Code / QwenCode / OpenCode / Aider…）：目标→拆解→路由→执行→验证→跨厂商评审→修复→报告。纯 Python 标准库实现，本地 Web UI。",
   "license": "MIT",
   "bin": {