codebee 0.1.65 → 0.1.67

package/app/core/planner.py

@@ -69,7 +69,7 @@ def _log_streamer(log_path, min_chars=400, min_secs=0.8):
     return cb
 
 
-def _outline_timeout():
+def _outline_timeout():
     """大纲/规划的 CLI 兜底超时（秒）：env TUTTI_OUTLINE_TIMEOUT（秒）优先，
     其次设置 orchestrator.outline_timeout_s，缺省 900。"""
     try:
@@ -86,16 +86,16 @@ def _outline_timeout():
             return v
     except Exception:
         pass
-    return DEFAULT_OUTLINE_TIMEOUT
-
-
-def _deadline_timeout(deadline, requested):
-    if deadline is None:
-        return requested
-    try:
-        return max(1, min(float(requested), deadline - time.monotonic()))
-    except (TypeError, ValueError):
-        return requested
+    return DEFAULT_OUTLINE_TIMEOUT
+
+
+def _deadline_timeout(deadline, requested):
+    if deadline is None:
+        return requested
+    try:
+        return max(1, min(float(requested), deadline - time.monotonic()))
+    except (TypeError, ValueError):
+        return requested
 
 
 def _log_usage(source, role, task, res, agent=None, tool="", model="", provider="",
@@ -113,7 +113,7 @@ def _log_usage(source, role, task, res, agent=None, tool="", model="", provider=
                      cost_usd=float((res or {}).get("cost_usd") or 0.0),
                      usage=(res or {}).get("usage"))
         # 告警模块：编排者直连调用的成功/失败上报（provider_id 供「禁用厂商」定位）
-        if provider:
+        if provider and not runner.attempt_cancelled(res):
             from . import health
             if (res or {}).get("ok"):
                 health.report_success(provider)
@@ -133,6 +133,8 @@ CODE_PLAN_PROMPT = """你是技术负责人。请先判定任务难度，再按
 子任务拆分：easy 一律只拆 1 个子任务——一个子任务把目标完整做完并自检通过即收尾，
 禁止拆出独立的核查/联调/收尾子任务；只有 hard 才允许多个子任务（至多 __N__ 个），
 粒度要可独立验证，且最后一个子任务必须包含整体联调/收尾。
+hard 也不做任务没要的防御性扩展子任务（缓存/重试/监控/通用抽象层）——
+发现真实风险写进对应子任务的 detail 里备注，由人决定要不要做。
 
 ## 先探索再计划（重要，借鉴 OpenSpec explore）
 拆解之前先用你的读文件/搜索工具**实际查看工作目录**，找到目标相关的真实文件与函数，再据此拆解。
@@ -248,7 +250,8 @@ def _norm_chapters(data, n, vol_per=0, vol_start=1, vol_plan=None):
     # 分卷：只收本批真正用到的卷的卷名/卷弧光（模型可能多写或乱写卷号，一律按
     # 本批章号范围过滤）。卷边界不在这里——它由 volumes.build_plan 从章号推导。
     if vol_plan:
-        named = volumes.norm_volumes(data.get("volumes"), vol_per, vol_start, n)
+        named = volumes.norm_volumes(data.get("volumes"), vol_per, vol_start, n,
+                                      plan=vol_plan)
         if named:
             res["volumes"] = named
     return res
@@ -355,8 +358,8 @@ def book_volume_plan(task, outline, upto):
     return volumes.merge_titles(plan, named)
 
 
-def make_serial_outline(task, author_agent=None, workdir=None, ev=None, log_path=None,
-                        deadline=None):
+def make_serial_outline(task, author_agent=None, workdir=None, ev=None, log_path=None,
+                        deadline=None, run_id=None):
     """连载大纲：编排者 API 优先 → 作者 CLI → 模板。返回 {book_title, chapters:[{title,beats,hook}]}。
 
     续写批次（serial.start_chapter > 1）：改用续写大纲提示词，注入前情大纲与
@@ -368,7 +371,7 @@ def make_serial_outline(task, author_agent=None, workdir=None, ev=None, log_path
     n = int(serial.get("chapters") or 8)
     wpc = int(serial.get("words_per_chapter") or 2500)
     start = int(serial.get("start_chapter") or 1)
-    sk_block, _ = skills.block_for(task)
+    sk_block, _ = skills.block_for(task, run_id=run_id)
     kb_block = knowledge.block_for(task)
     if kb_block:
         sk_block = (sk_block + "\n\n" + kb_block) if sk_block else kb_block
@@ -453,8 +456,8 @@ def make_serial_outline(task, author_agent=None, workdir=None, ev=None, log_path
             # prompt 逐字节相同，命中省一次全量规划调用。第 2 次尝试故意
             # 旁路缓存——首试 ok 但 JSON 不合法时，重试要的是「再抽一次样
             # 本」而不是同一份坏文本。
-            res = modelhub.chat(prov["id"], model, prompt,
-                                max_tokens=16000, timeout=_deadline_timeout(deadline, 300), on_delta=cb,
+            res = modelhub.chat(prov["id"], model, prompt,
+                                max_tokens=16000, timeout=_deadline_timeout(deadline, 300), on_delta=cb,
                                 reasoning_effort=_reasoning_effort(task),
                                 cache_ttl=3600 if _attempt == 1 else 0)
             cb.flush()
@@ -483,8 +486,8 @@ def make_serial_outline(task, author_agent=None, workdir=None, ev=None, log_path
         _append_log(log_path, "===== 作者 CLI（%s）=====" % author_agent.get("id", "?"))
         res = runner.run_agent(modelhub.bind_agent(author_agent), prompt,
                                workdir=workdir or task.get("workdir"), readonly=True,
-                               timeout=_deadline_timeout(deadline, _outline_timeout()),
-                               deadline=deadline, cancel_event=ev,
+                               timeout=_deadline_timeout(deadline, _outline_timeout()),
+                               deadline=deadline, cancel_event=ev,
                                log_path=log_path)
         _log_usage("outline", "outline", task, res, agent=author_agent)
         if not res["ok"]:
@@ -602,13 +605,13 @@ def _reasoning_effort(task):
     return "medium"
 
 
-def make_code_plan(task, planner_agent, workdir, ev=None, resume=None, log_path=None,
-                   deadline=None):
+def make_code_plan(task, planner_agent, workdir, ev=None, resume=None, log_path=None,
+                   deadline=None):
     """代码任务计划：编排者 API 优先 → CLI 智能体 → 单步模板。"""
     orch = _orchestrator()
     if orch:
-        plan = _orch_code_plan(task, orch[0], orch[1], log_path=log_path,
-                               deadline=deadline)
+        plan = _orch_code_plan(task, orch[0], orch[1], log_path=log_path,
+                               deadline=deadline)
         if plan:
             return plan
     if planner_agent is None:
@@ -620,8 +623,8 @@ def make_code_plan(task, planner_agent, workdir, ev=None, resume=None, log_path=
               .replace("__CONTEXT__", task.get("context") or "（无）")
               .replace("__VERIFY__", task.get("verify_command") or "（未配置）"))
     res = runner.run_agent(planner_agent, prompt, workdir=workdir, readonly=True,
-                           timeout=_deadline_timeout(deadline, 300), deadline=deadline,
-                           cancel_event=ev, resume=resume,
+                           timeout=_deadline_timeout(deadline, 300), deadline=deadline,
+                           cancel_event=ev, resume=resume,
                            log_path=log_path)
     _log_usage("plan", "plan", task, res, agent=planner_agent)
     data = runner.extract_json(res.get("text") or "")
@@ -638,7 +641,7 @@ def _fallback_code_plan(task, note):
             "steps": [{"title": "实现任务", "detail": task["goal"]}]}
 
 
-def _orch_code_plan(task, prov, model, log_path=None, deadline=None):
+def _orch_code_plan(task, prov, model, log_path=None, deadline=None):
     cb = _log_streamer(log_path)
     # §07 T2.2：代码计划同任务同 prompt（断点续跑/重试），TTL 1h 精确缓存
     res = modelhub.chat(prov["id"], model,
@@ -646,12 +649,13 @@ def _orch_code_plan(task, prov, model, log_path=None, deadline=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=_deadline_timeout(deadline, 300), on_delta=cb,
+                        max_tokens=8000, timeout=_deadline_timeout(deadline, 300), on_delta=cb,
                         reasoning_effort=_reasoning_effort(task),
                         cache_ttl=3600)
     cb.flush()
     _log_usage("plan", "plan", task, res, model=model,
-               provider=prov.get("name", prov.get("id", "")))
+               provider=prov.get("name", prov.get("id", "")),
+               provider_id=prov.get("id", ""))
     if not res["ok"]:
         _append_log(log_path, "编排者计划（%s · %s）失败：%s" % (
             prov.get("name", prov["id"]), model, (res.get("error") or "未知错误")[:300]))
@@ -666,7 +670,7 @@ def _orch_code_plan(task, prov, model, log_path=None, deadline=None):
             "workflow": _plan_workflow(data)}
 
 
-def make_review_outline(task, deadline=None):
+def make_review_outline(task, deadline=None):
     """review 类任务：编排者产出写作大纲（失败返回 None，起草退回无大纲）。"""
     orch = _orchestrator()
     if not orch:
@@ -676,11 +680,12 @@ def make_review_outline(task, deadline=None):
                         (REVIEW_OUTLINE_PROMPT
                          .replace("__GOAL__", task["goal"])
                          .replace("__CONTEXT__", task.get("context") or "（无）")),
-                        max_tokens=8000, timeout=_deadline_timeout(deadline, 300),
+                        max_tokens=8000, timeout=_deadline_timeout(deadline, 300),
                         reasoning_effort=_reasoning_effort(task),
                         cache_ttl=3600)   # §07 T2.2：同任务重试幂等，TTL 1h
     _log_usage("outline", "outline", task, res, model=model,
-               provider=prov.get("name", prov.get("id", "")))
+               provider=prov.get("name", prov.get("id", "")),
+               provider_id=prov.get("id", ""))
     if not res["ok"]:
         return None
     data = runner.extract_json(res.get("text") or "")

package/app/core/portguard.py

@@ -1,111 +1,111 @@
-# -*- coding: utf-8 -*-
-"""启动端口清场（用户拍板 2026-09-21）：端口被占时先清场再启动。
-
-- **自家旧实例**（命令行含本包 main.py 完整路径，或 npm 包内 app/main.py）
-  → 杀树（TerminateProcess 直杀）——升级/重启最常见的占用者就是没退干净
-  的 CodeBee 自己；控制台进程对温和信号（WM_CLOSE）无反应，温和关不掉
-  正是用户「旧进程太难杀」的痛点，所以自家实例直接强杀。
-- **别人的进程** → 只报告占用者，不发送任何信号；启动流程不得替用户
-  关闭可能正在开发中的服务。
-- 系统/自身进程拒关（portscan 内置护栏）。
-"""
-from __future__ import annotations
-
-import logging
-import os
-import shlex
-import subprocess
-
-from . import portscan, runner
-
-log = logging.getLogger(__name__)
-
-PS_EXE = "C:\\Windows\\System32\\WindowsPowerShell\\v1.0\\powershell.exe"
-
-
-def proc_cmdline(pid):
-    """读进程命令行；失败返回空串。仅用于确认旧 CodeBee 实例身份。"""
-    try:
-        pid = int(pid)
-    except (TypeError, ValueError):
-        return ""
-    if pid <= 0:
-        return ""
-    if os.name != "nt":
-        try:
-            with open("/proc/%d/cmdline" % pid, "rb") as fh:
-                return fh.read().decode("utf-8", "replace").replace("\x00", " ").strip()
-        except Exception:
-            return ""
-    try:
-        r = subprocess.run(
-            [PS_EXE, "-NoProfile", "-Command",
-             "(Get-CimInstance Win32_Process -Filter 'ProcessId = %d').CommandLine" % pid],
-            capture_output=True, timeout=10)
-        return r.stdout.decode("utf-8", "replace").strip() if r.returncode == 0 else ""
-    except Exception:
-        return ""
-
-
-def _command_tokens(cmdline):
-    try:
-        return [token.strip().strip('"').strip("'")
-                for token in shlex.split(str(cmdline or ""), posix=False)]
-    except (TypeError, ValueError):
-        return []
-
-
-def is_own_instance(cmdline, main_script, port=None):
-    """按启动器、脚本独立参数和端口精确确认 CodeBee 实例。"""
-    tokens = _command_tokens(cmdline)
-    if len(tokens) < 2:
-        return False
-    launcher = os.path.splitext(os.path.basename(tokens[0]))[0].lower()
-    if launcher not in ("python", "python3", "py"):
-        return False
-    script = ""
-    for token in tokens[1:]:
-        if token.startswith("-"):
-            continue
-        script = token
-        break
-    if not script or not os.path.isabs(script):
-        return False
-    actual = os.path.normcase(os.path.normpath(os.path.abspath(script)))
-    expected = os.path.normcase(os.path.normpath(os.path.abspath(str(main_script))))
-    packaged = actual.replace("\\", "/").lower().endswith(
-        "/node_modules/codebee/app/main.py")
-    if actual != expected and not packaged:
-        return False
-    if port is None:
-        return True
-    try:
-        wanted = str(int(port))
-    except (TypeError, ValueError):
-        return False
-    explicit_port = None
-    for i, token in enumerate(tokens):
-        if token == "--port" and i + 1 < len(tokens):
-            explicit_port = tokens[i + 1]
-        elif token.startswith("--port="):
-            explicit_port = token.split("=", 1)[1]
-    # argparse 默认端口可不出现在旧实例参数中；非默认端口必须显式相符。
-    return ((explicit_port == wanted) if explicit_port is not None
-            else wanted == "8765")
-
-
-def clear_stale_port(port, main_script):
-    """清掉占用端口的进程。返回 (是否清掉, 人话说明)。"""
-    holders = [p for p in portscan.listening_ports() if p.get("port") == int(port)]
-    if not holders:
-        return True, ""
-    pid = int(holders[0].get("pid") or 0)
-    if pid <= 4 or pid == os.getpid():
-        return False, "占用者是系统进程/自身，拒绝清理"
-    if is_own_instance(proc_cmdline(pid), main_script, port=port):
-        try:
-            runner._kill_tree(pid)
-            return True, "已结束旧实例 PID %d" % pid
-        except Exception as e:
-            return False, "旧实例 PID %d 清理失败 %s" % (pid, str(e)[:80])
-    return False, "占用者非 CodeBee（PID %d），未自动关闭" % pid
+# -*- coding: utf-8 -*-
+"""启动端口清场（用户拍板 2026-09-21）：端口被占时先清场再启动。
+
+- **自家旧实例**（命令行含本包 main.py 完整路径，或 npm 包内 app/main.py）
+  → 杀树（TerminateProcess 直杀）——升级/重启最常见的占用者就是没退干净
+  的 CodeBee 自己；控制台进程对温和信号（WM_CLOSE）无反应，温和关不掉
+  正是用户「旧进程太难杀」的痛点，所以自家实例直接强杀。
+- **别人的进程** → 只报告占用者，不发送任何信号；启动流程不得替用户
+  关闭可能正在开发中的服务。
+- 系统/自身进程拒关（portscan 内置护栏）。
+"""
+from __future__ import annotations
+
+import logging
+import os
+import shlex
+import subprocess
+
+from . import portscan, runner
+
+log = logging.getLogger(__name__)
+
+PS_EXE = "C:\\Windows\\System32\\WindowsPowerShell\\v1.0\\powershell.exe"
+
+
+def proc_cmdline(pid):
+    """读进程命令行；失败返回空串。仅用于确认旧 CodeBee 实例身份。"""
+    try:
+        pid = int(pid)
+    except (TypeError, ValueError):
+        return ""
+    if pid <= 0:
+        return ""
+    if os.name != "nt":
+        try:
+            with open("/proc/%d/cmdline" % pid, "rb") as fh:
+                return fh.read().decode("utf-8", "replace").replace("\x00", " ").strip()
+        except Exception:
+            return ""
+    try:
+        r = subprocess.run(
+            [PS_EXE, "-NoProfile", "-Command",
+             "(Get-CimInstance Win32_Process -Filter 'ProcessId = %d').CommandLine" % pid],
+            capture_output=True, timeout=10)
+        return r.stdout.decode("utf-8", "replace").strip() if r.returncode == 0 else ""
+    except Exception:
+        return ""
+
+
+def _command_tokens(cmdline):
+    try:
+        return [token.strip().strip('"').strip("'")
+                for token in shlex.split(str(cmdline or ""), posix=False)]
+    except (TypeError, ValueError):
+        return []
+
+
+def is_own_instance(cmdline, main_script, port=None):
+    """按启动器、脚本独立参数和端口精确确认 CodeBee 实例。"""
+    tokens = _command_tokens(cmdline)
+    if len(tokens) < 2:
+        return False
+    launcher = os.path.splitext(os.path.basename(tokens[0]))[0].lower()
+    if launcher not in ("python", "python3", "py"):
+        return False
+    script = ""
+    for token in tokens[1:]:
+        if token.startswith("-"):
+            continue
+        script = token
+        break
+    if not script or not os.path.isabs(script):
+        return False
+    actual = os.path.normcase(os.path.normpath(os.path.abspath(script)))
+    expected = os.path.normcase(os.path.normpath(os.path.abspath(str(main_script))))
+    packaged = actual.replace("\\", "/").lower().endswith(
+        "/node_modules/codebee/app/main.py")
+    if actual != expected and not packaged:
+        return False
+    if port is None:
+        return True
+    try:
+        wanted = str(int(port))
+    except (TypeError, ValueError):
+        return False
+    explicit_port = None
+    for i, token in enumerate(tokens):
+        if token == "--port" and i + 1 < len(tokens):
+            explicit_port = tokens[i + 1]
+        elif token.startswith("--port="):
+            explicit_port = token.split("=", 1)[1]
+    # argparse 默认端口可不出现在旧实例参数中；非默认端口必须显式相符。
+    return ((explicit_port == wanted) if explicit_port is not None
+            else wanted == "8765")
+
+
+def clear_stale_port(port, main_script):
+    """清掉占用端口的进程。返回 (是否清掉, 人话说明)。"""
+    holders = [p for p in portscan.listening_ports() if p.get("port") == int(port)]
+    if not holders:
+        return True, ""
+    pid = int(holders[0].get("pid") or 0)
+    if pid <= 4 or pid == os.getpid():
+        return False, "占用者是系统进程/自身，拒绝清理"
+    if is_own_instance(proc_cmdline(pid), main_script, port=port):
+        try:
+            runner._kill_tree(pid)
+            return True, "已结束旧实例 PID %d" % pid
+        except Exception as e:
+            return False, "旧实例 PID %d 清理失败 %s" % (pid, str(e)[:80])
+    return False, "占用者非 CodeBee（PID %d），未自动关闭" % pid