codebee 0.1.30 → 0.1.32

package/CHANGELOG.md

@@ -6,6 +6,23 @@ 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
 
+## v0.1.32（2026-09-21）
+
+### ✨ 新功能
+
+- 连载任务详情新增「章节评审」卡片：每章的维度评分（低分标红）、达标状态、字数与修订轮数一目了然；每章按大纲逐项核对的判定结果（已完成/未完成/待核实）以红绿灰行直接展示——哪章哪个要点没写到一眼可查。
+
+## v0.1.31（2026-09-21）
+
+### ✨ 新功能
+
+- 群摘要搬进正式页面：设置 →「微信群聊汇报」现在是完整页面——配置（监控文件夹/微信直连/Python 扫描与一键安装/监听群）和各群摘要列表一体呈现，不再挤在右下角悬浮小窗里。
+- 桌面蜜蜂右键「群摘要」直达该页面（已打开的页面只聚焦不重复开）。
+
+### 🛠 问题修复
+
+- 修复微信直连「选不中 Python」：此前每分钟的自动刷新会把用户刚选好但还没保存的路径悄悄冲回旧值，现在改过路径后不再回填覆盖，保存成功后才更新。
+
 ## v0.1.30（2026-09-21）
 
 ### ✨ 新功能

package/README.md

@@ -23,6 +23,17 @@ Kimi Code、MiMo Code、Grok Build、Pi、DeepSeek Harness……），提供
 <!-- relnotes:start -->
 ### ✨ 新功能 · New
 
+- 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.
+
+- 群摘要搬进正式页面：设置 →「微信群聊汇报」里配置与摘要列表一体呈现，右下角悬浮小窗已移除；桌面蜜蜂右键「群摘要」直达。
+- Group digests now live on a proper settings page (config + digest list in one place); the floating bottom-right widget is gone, and the desktop bee links straight to it.
+
+### 🛠 问题修复 · Fixes
+
+- 修复「选不中 Python」：自动刷新不再冲掉你刚选好但还没保存的路径。
+- Fixed "can't pick a Python": background refresh no longer wipes a freshly chosen but unsaved path.
+
 - 微信直连配置更省事：「扫描」自动找本机 Python（标注位数与依赖状态，点选即用），选中后可一键安装 wechatauto-replica。
 - WeChat setup is now effortless: a Scan button finds local Pythons (with bitness and dependency badges), and one click installs wechatauto-replica.
 

package/app/core/builtin_agent.py

@@ -566,7 +566,9 @@ def run(bi, prompt, workdir, timeout=180, cancel_event=None, log=None, images=No
                 url, headers, body = _build_request(proto, pbase, kk["key"], model,
                                                     system, msgs, tools_ok)
                 effort = str(bi.get("reasoning_effort") or "").strip().lower()
-                if effort in ("low", "medium", "high") and proto != "google":
+                # reasoning_effort 是 OpenAI wire 字段；Anthropic thinking 使用
+                # 另一套对象结构，向兼容网关硬塞该字段会直接得到 400。
+                if effort in ("low", "medium", "high") and proto == "openai":
                     body["reasoning_effort"] = effort
                 status, data, err = _post_interruptible(url, headers, body,
                                                         allow_private, timeout,

package/app/core/modelhub.py

@@ -3003,16 +3003,17 @@ def resolve_orchestrator():
     return prov, model
 
 
-def _chat_cache_path(provider_id, model_name, prompt, max_tokens):
+def _chat_cache_path(provider_id, model_name, prompt, max_tokens, reasoning_effort=""):
     """§07 T2.2：精确匹配响应缓存的落盘路径（只缓存 ok 的幂等调用）。"""
     import hashlib as _h
-    key = "|".join([str(provider_id), str(model_name), str(max_tokens), str(prompt)])
+    key = "|".join([str(provider_id), str(model_name), str(max_tokens),
+                    str(reasoning_effort or ""), str(prompt)])
     name = _h.sha256(key.encode("utf-8")).hexdigest()[:24]
     return paths.DATA_DIR / "chat_cache" / (name + ".json")
 
 
 def chat(provider_id, model_name, prompt, max_tokens=2048, timeout=120, cache_ttl=0,
-         on_delta=None):
+         on_delta=None, reasoning_effort=""):
     """直连供应商 API 做一次对话（编排者规划 / 连通性测试）。
 
     支持 anthropic / openai / google 三种协议；复用 SSRF 防护。
@@ -3026,7 +3027,8 @@ def chat(provider_id, model_name, prompt, max_tokens=2048, timeout=120, cache_tt
     """
     if cache_ttl > 0:
         try:
-            cache_path = _chat_cache_path(provider_id, model_name, prompt, max_tokens)
+            cache_path = _chat_cache_path(provider_id, model_name, prompt, max_tokens,
+                                          reasoning_effort)
             if cache_path.is_file():
                 age = time.time() - cache_path.stat().st_mtime
                 if age <= cache_ttl:
@@ -3073,6 +3075,9 @@ def chat(provider_id, model_name, prompt, max_tokens=2048, timeout=120, cache_tt
                 headers = {"Authorization": "Bearer " + use_key}
             body = {"model": model_name, "max_tokens": max_tokens,
                     "messages": [{"role": "user", "content": prompt}]}
+            effort = str(reasoning_effort or "").strip().lower()
+            if proto == "openai" and effort in ("low", "medium", "high"):
+                body["reasoning_effort"] = effort
         return url, headers, body
 
     last_err = ""
@@ -3156,7 +3161,8 @@ def chat(provider_id, model_name, prompt, max_tokens=2048, timeout=120, cache_tt
     # §07 T2.2：cache_ttl>0 时落盘缓存（仅 ok 结果，原子写）
     if cache_ttl > 0:
         try:
-            cache_path = _chat_cache_path(provider_id, model_name, prompt, max_tokens)
+            cache_path = _chat_cache_path(provider_id, model_name, prompt, max_tokens,
+                                          reasoning_effort)
             cache_path.parent.mkdir(parents=True, exist_ok=True)
             tmp = cache_path.with_suffix(".tmp")
             import json as _json

package/app/core/pipeline.py

@@ -961,7 +961,7 @@ def _run_code(run, task, agents, ev, stats, mode):
                   and not task.get("attachments")
                   and user_context_len < 600))
     planned = False
-    if mode == "auto" and not fast_path:
+    if mode in ("auto", "expert") and not fast_path:
         _wait_gate(run_id, ev)
         plan_step, plan_log = store.add_step(run_id, "plan", impl["id"], impl.get("label"),
                                              note=route.get("implementer", ""))
@@ -1003,7 +1003,7 @@ def _run_code(run, task, agents, ev, stats, mode):
     if not workflow["review_required"]:
         reviewer = None
         route["reviewer"] = workflow["reason"]
-    elif mode == "auto":
+    elif mode in ("auto", "expert"):
         reviewer, route["reviewer"] = router.pick_reviewer(agents, impl, "code", stats)
     else:
         reviewer, route["reviewer"] = _pick_reviewer_legacy(agents, impl)
@@ -1235,7 +1235,8 @@ def _run_code(run, task, agents, ev, stats, mode):
         "# 代码任务报告：%s" % task["title"], "",
         "- 结论：**%s**" % ("✅ 通过" if overall_pass else "❌ 未通过"),
         "- 编排模式：%s　实现者：%s　评审者：%s" % (
-            "智能" if mode == "auto" else "手动", impl.get("label"),
+            {"auto": "自动", "fast": "快速", "expert": "专家",
+             "manual": "手动"}.get(mode, mode), impl.get("label"),
             reviewer.get("label") if reviewer else "按验证结果省略"),
         "- 动态步骤：规划 %s；实现 %d 步；验证 %s；评审 %d 人；最多修复 %d 轮；换将 %s" % (
             "独立执行" if workflow["planning"] == "llm" else "并入实现",
@@ -2941,7 +2942,8 @@ def _run_serial_review(run, task, agents, ev, stats, mode, critics, impl, route,
                  "通过" if chapters_pass else "未通过",
                  "通过" if global_pass else "未通过"),
              "- 编排模式：%s　作者：%s　评审组：%s" % (
-                 "智能" if mode == "auto" else "手动",
+                 {"auto": "自动", "fast": "快速", "expert": "专家",
+                  "manual": "手动"}.get(mode, mode),
                  impl.get("label"), "、".join(a.get("label") for a in critics)),
              "", "## 各章得分（章阈值 %.1f）" % threshold_ch, "",
              "| 章 | 标题 | " + " | ".join(dims) + " | 均分 | 达标 | 轮次 | 字数 |",
@@ -3119,7 +3121,7 @@ def _run_content_review(run, task, agents, ev, stats, mode):
                          error="没有可用智能体", ended_at=_now())
         return
     if resume_ctx is not None:
-        if mode == "auto":
+        if mode != "manual":
             critics, route["critics"] = router.pick_critics(
                 agents, task.get("type") or "novel", stats, impl=impl)
         else:
@@ -3151,7 +3153,7 @@ def _run_content_review(run, task, agents, ev, stats, mode):
         except Exception:
             return ""
 
-    draft_note = route.get("author", "") if mode == "auto" else ""
+    draft_note = route.get("author", "") if mode != "manual" else ""
 
     # 编排者大纲：只对真实执行有意义；失败静默退回无大纲（喂入带指令的任务副本）
     outline = (planner.make_review_outline(_steered_task(run_id, task))

package/app/core/planner.py

@@ -327,8 +327,9 @@ def make_serial_outline(task, author_agent=None, workdir=None, ev=None, log_path
             # 否则 stop_reason=max_tokens、正文为空（实测 2048 全被思考吞掉）
             # 流式回调把增量实时写进步骤日志，直连生成不再是一行标题的黑箱
             cb = _log_streamer(log_path)
-            res = modelhub.chat(prov["id"], model, prompt,
-                                max_tokens=16000, timeout=300, on_delta=cb)
+            res = modelhub.chat(prov["id"], model, prompt,
+                                max_tokens=16000, timeout=300, on_delta=cb,
+                                reasoning_effort=_reasoning_effort(task))
             cb.flush()
             _log_usage("outline", "outline", task, res, model=model,
                        provider=prov.get("name", prov.get("id", "")),
@@ -431,12 +432,33 @@ def _plan_workflow(data):
     return out or None
 
 
-def _orchestrator():
-    """编排者可用时返回 (provider, model)，否则 None。"""
-    try:
-        return modelhub.resolve_orchestrator()
-    except Exception:
-        return None
+def _orchestrator():
+    """编排者可用时返回 (provider, model)，否则 None。"""
+    try:
+        return modelhub.resolve_orchestrator()
+    except Exception:
+        return None
+
+
+def _reasoning_effort(task):
+    """把任务级思考程度映射为直连编排者支持的 reasoning 参数。"""
+    task = task if isinstance(task, dict) else {}
+    selected = str(task.get("thinking") or "standard").strip().lower()
+    if selected in ("low", "standard", "high"):
+        return {"low": "low", "standard": "medium", "high": "high"}[selected]
+    mode = str(task.get("mode") or "auto").lower()
+    difficulty = str(task.get("difficulty") or "auto").lower()
+    ttype = str(task.get("type") or "").lower()
+    try:
+        threshold = float(task.get("threshold") or 7.0)
+    except (TypeError, ValueError):
+        threshold = 7.0
+    if mode == "fast" or difficulty == "easy" or threshold <= 6.0:
+        return "low"
+    if (mode == "expert" or difficulty == "hard" or threshold >= 8.5
+            or ttype in ("research", "tech_proposal", "serial_novel")):
+        return "high"
+    return "medium"
 
 
 def make_code_plan(task, planner_agent, workdir, ev=None, resume=None, log_path=None):
@@ -479,7 +501,8 @@ def _orch_code_plan(task, prov, model, log_path=None):
                          .replace("__GOAL__", task["goal"])
                          .replace("__CONTEXT__", task.get("context") or "（无）")
                          .replace("__VERIFY__", task.get("verify_command") or "（未配置）")),
-                        max_tokens=8000, timeout=300, on_delta=cb)
+                        max_tokens=8000, timeout=300, on_delta=cb,
+                        reasoning_effort=_reasoning_effort(task))
     cb.flush()
     _log_usage("plan", "plan", task, res, model=model,
                provider=prov.get("name", prov.get("id", "")))
@@ -507,7 +530,8 @@ def make_review_outline(task):
                         (REVIEW_OUTLINE_PROMPT
                          .replace("__GOAL__", task["goal"])
                          .replace("__CONTEXT__", task.get("context") or "（无）")),
-                        max_tokens=8000, timeout=300)
+                        max_tokens=8000, timeout=300,
+                        reasoning_effort=_reasoning_effort(task))
     _log_usage("outline", "outline", task, res, model=model,
                provider=prov.get("name", prov.get("id", "")))
     if not res["ok"]:

package/app/core/repeat_guard.py

@@ -6,7 +6,7 @@
   监听 tools/post-execute，按 (tool, canonical_args) 跟踪，3/5/8 阈值注入 user-message 提醒。
 
 Tutti 的 "tool" 对应到 "step role"，"canonical_args" 对应到 prompt 模板指纹。
-阈值更宽松（5/10/15），因 Tutti 步骤级别比 dsh 工具调用粗。
+步骤级调用代价远高于普通工具调用，因此第 3 次提醒、第 5 次强制停止。
 """
 from __future__ import annotations
 
@@ -18,7 +18,7 @@ import time
 log = logging.getLogger(__name__)
 
 # 渐进阈值：第 N 次命中时给提醒；最后一次及以上时强制停止。
-DEFAULT_THRESHOLDS = (5, 10, 15)
+DEFAULT_THRESHOLDS = (3, 5)
 
 REMINDER_TEMPLATE = (
     "⚠️ 提醒：上一步使用了几乎相同的输入（已连续 {count} 次调用 {role}）。"
@@ -121,4 +121,4 @@ class RepeatGuard:
 
 
 # 模块级默认实例（pipeline 接线用；run_id 入键天然隔离多任务，测试无需替换）
-guard = RepeatGuard()
+guard = RepeatGuard()

package/app/core/store.py

@@ -486,6 +486,15 @@ def load_all():
 
 # ---------------------------------------------------------------- 运行（含管理操作）
 
+_RUN_ESTIMATOR = None
+
+
+def set_run_estimator(estimator):
+    """由应用入口注入 ETA 估算器，保持存储层不反向依赖用量模块。"""
+    global _RUN_ESTIMATOR
+    _RUN_ESTIMATOR = estimator if callable(estimator) else None
+
+
 def create_run(kind, title, task_id=None, entry_id=None, op=None):
     run = {
         "id": _new_id("r" if kind == "orchestration" else "m"),
@@ -501,8 +510,7 @@ def create_run(kind, title, task_id=None, entry_id=None, op=None):
     if kind == "orchestration" and task_id:
         task = get_task(task_id) or {}
         try:
-            from . import usage
-            eta = usage.estimate(
+            eta = _RUN_ESTIMATOR(
                 task_type=task.get("type") or "", mode=task.get("mode") or "auto",
                 thinking=task.get("thinking") or "standard",
                 rounds=task.get("rounds"))

package/app/core/task_compile.py

@@ -103,8 +103,9 @@ def code_workflow(task, difficulty, plan=None, mode="auto", planned=True):
     risk_text = " ".join(str(task.get(k) or "") for k in ("title", "goal", "context")).lower()
     high_risk = bool(any(k in risk_text for k in (
         "安全", "权限", "鉴权", "认证", "加密", "密钥", "注入", "迁移", "数据库",
-        "并发", "竞态", "支付", "public api", "公共 api", "breaking change",
-        "security", "permission", "auth", "migration", "concurrency")))
+        "并发", "竞态", "支付", "架构", "公共接口", "public api", "公共 api",
+        "breaking change", "security", "permission", "auth", "migration",
+        "concurrency", "architecture")))
     fast_mode = mode == "fast"
     expert_mode = mode == "expert"
     easy_auto = (mode == "auto" and difficulty == "easy") or fast_mode

package/app/main.py

@@ -2386,6 +2386,7 @@ def main():
     except Exception:
         pass
     from core import usage
+    store.set_run_estimator(usage.estimate)
     _step("正在回填用量台账…")
     n_bf = usage.backfill_from_runs()  # 历史运行 token 回填台账（幂等，仅补缺失步骤）
     if n_bf:
@@ -2536,7 +2537,8 @@ def main():
 
     print("[CodeBee] 本机     http://127.0.0.1:%d" % args.port, flush=True)
     print("[CodeBee] 数据目录 %s" % paths.DATA_DIR, flush=True)
-    print("[CodeBee] ✔ 已就绪，浏览器即将自动打开；不要关闭本窗口。", flush=True)
+    # Windows 的默认 GBK 控制台无法编码部分符号，启动日志必须保持控制台安全。
+    print("[CodeBee] [OK] 已就绪，浏览器即将自动打开；不要关闭本窗口。", flush=True)
     try:
         import faulthandler
         faulthandler.cancel_dump_traceback_later()  # 启动完成，看门狗退役（否则运行期每 20s 误报堆栈）